Accounting, Organizations and Society xxx (2015) 1e18

Contents lists available at ScienceDirect

Accounting, Organizations and Society

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

The accuracy of disclosures for complex estimates: Evidence from

reported stock option fair values*
Brian Bratten a, Ross Jennings b, *, Casey M. Schwab c
Gatton College of Business & Economics, University of Kentucky, 423FB&E, Lexington, KY, 40506, USA
McCombs School of Business, University of Texas at Austin, B6400, Austin, TX, 78712, USA
Terry College of Business, University of Georgia, 242 Brooks Hall, Athens, GA, 30602, USA

articleinfo abstract

Article history: In this study, we exploit the unique reporting requirements for employee stock options to provide large
Received 30 May 2014 sample evidence on the accuracy of footnote disclosures related to a specic complex estimate, the fair
Received in revised form
value of options granted. We rst document the frequency and magnitude of differences between (1) the
2 July 2015
reported weighted-average fair value of options granted and (2) the calculated option fair value using the
Accepted 2 September 2015
disclosed weighted-average valuation model inputs and the Black-Scholes option pricing model. In a
Available online xxx
sample of 23,358 rm-year observations between 2004 and 2011, we nd that 23.9 percent have re-
ported and calculated option fair values that differ by more than ten percent, and that these differences
Keywords:
Employee stock options are sticky and are frequently signicant as a percentage of net income. We also nd that fair value
Fair values differences are larger for rms that (1) exhibit anomalous stock option footnote disclosures that likely
Disclosure quality result from disclosure errors, (2) have more complex and hence error-prone stock option programs, and
Complex estimates (3) have lower quality nancial reporting. Taken together this evidence is consistent with large fair value
differences that are primarily due to unintentional errors in the stock option footnote disclosures. To
document the consequences of these fair value differences, we provide evidence that errors in the re-
ported fair values prevent nancial statement users from using the reported values to reliably estimate
future stock option expense for many rms. Consistent with this result, we also nd that analyst forecasts
are less accurate and more disperse for rms with larger fair value differences.
2015 Elsevier Ltd. All rights reserved.

1. Introduction estimates (SEC, 2003; PCAOB, 2011; Bratten, Gaynor, McDaniel,

As the use of fair values and other complex estimates in nancial
reporting has increased, regulators, the media, and researchers
have expressed concerns about the reporting and auditing of these
*
We gratefully acknowledge helpful comments from Mark Peecher (editor), two
anonymous reviewers, Steve Baginiski, John Campbell, Preeti Choudhary, Dain
Donelson, Theodore Goodman, Jackie Hammersley, John McInnis, Laura Wang, Teri
Yohn, Yong Yu, and workshop participants at Miami University, the University of
Toronto, the 2014 FARS midyear meeting, and the 2014 AAA annual meeting. Brian
Bratten acknowledges the nancial support of the Von Allmen School of Accoun-
tancy and the Gatton College of Business and Economics. Ross Jennings acknowl-
edges the nancial support of the Deloitte & Touche Professorship of Accounting
and the McCombs School of Business. Casey Schwab acknowledges the nancial
support of the Terry College of Business and the J.M. Tull School of Accounting. We
also acknowledge the excellent research assistance of Philip Chung, Anne Ehinger,
Ying Huang, Jonathan Ross, and Russell Williamson.
* Corresponding author.
E-mail address: ross.jennings@mccombs.utexas.edu (R. Jennings).
Montague, & Sierra, 2013; Rapoport, 2013; Grifth, Hammersley,
& Kadous, 2015). These estimates are often accompanied by foot-
note disclosures providing the earliest information to nancial
statement users about the calculation of the estimates and how the
estimates impact net income. Despite the importance of these
supporting disclosures, there is virtually no research on their
accuracy.
In this study, we conduct a comprehensive examination of
footnote disclosures related to a specic complex estimate, the fair
value of employee stock options granted. We focus on stock option
disclosures to exploit the unique reporting requirements for stock
options that, unlike disclosures for other estimates, require
disclosure of not only the calculation outputdthe estimated fair
value of employee stock options granteddbut also the calculation
inputs and the method of calculation. These requirements allow us
to evaluate the internal consistency of these disclosures by
comparing the disclosed fair value to a calculated fair value based
on the rm's disclosed inputs and the Black-Scholes valuation

http://dx.doi.org/10.1016/j.aos.2015.09.001
0361-3682/ 2015 Elsevier Ltd. All rights reserved.

Please cite this article in press as: Bratten, B., et al., The accuracy of disclosures for complex estimates: Evidence from reported stock option fair
values, Accounting, Organizations and Society (2015), http://dx.doi.org/10.1016/j.aos.2015.09.001
2
B. Bratten et al. / Accounting, Organizations and Society xxx (2015) 1e18
model. Specically, we construct a measure that we refer to as the
fair value difference that equals the absolute value of the differ-
ence between the reported and calculated fair values, scaled by the
calculated fair value. We use this measure to investigate both the
accuracy of these disclosures and the consequences of inaccurate
disclosures. Our analyses proceed in three stages.
In the rst stage, we document the frequency and materiality of
large fair value differences (greater than ten percent). We focus
on large differences because small differences can occur when
multiple grants are issued in the same year and the inputs vary
across grants. In our sample of 23,358 rm-year observations be-
tween 2004 and 2011, we nd that 5573 observations (23.9
percent) have large differences between the reported and calcu-
lated fair values. For observations with large differences, 69.5
percent of the differences are greater than one percent of reported
income. These results suggest the potential for a signicant number
of material errors in a high prole nancial statement footnote.
In the second stage, we conduct analyses to determine whether
unintentional disclosure errors are the primary source of these fair
value differences. We rst evaluate Compustat footnote data and
uncover two anomalies in the stock option disclosure that suggest
possible errors. Specically, we nd that 1109 rm-year observa-
tions (4.7 percent of the sample) disclose a reported fair value for
their options granted equal to the exercise price of the options, a
virtual impossibility without a disclosure error. We also nd that
1932 rm-year observations (10.8 percent of the 17,959 rm-year
observations for which we have disclosure data from the prior
year) disclose the same volatility, risk-free-rate, exercise price, or
option grant date fair value for option grants in consecutive years,
an unlikely occurrence. For both of these anomalies, the fair value
difference is larger for observations exhibiting the anomaly, sug-
gesting that disclosure errors are a likely source of at least some of
the fair value differences.
To provide further evidence that fair value differences represent
disclosure errors, we then conduct a multivariate analysis to
determine whether fair value differences are associated with the
complexity of the rm's stock option program and with rm
reporting and operating environment characteristics that suggest
low quality nancial reporting. We nd that rms have larger fair
value differences when they have more complex, and likely more
error-prone, stock option programs. Specically, rms exhibit
larger fair value differences when they grant relatively more op-
tions to rank-and-le employees, have higher rates of cancellations,
have larger changes in their employee base, split their stock during
the current year, or make more extensive use of options. We also
nd that fair value differences are larger when nancial reporting
quality is lower. Specically, fair value differences are larger in years
in which stock option expense is disclosed rather than expensed
and for rms that have an internal control weakness, have an ac-
counting restatement, do not use a Big 4 auditor, or do not have a
chief accounting ofcer as one of its most highly compensated
executives. Finally, we nd that fair value differences are larger for
rms that have less developed accounting systems or operate in a
more complex reporting environment. Specically, rms making a
corporate acquisition in the current year, as well as smaller and
younger rms, exhibit larger fair value differences.
Although these results suggest that, on average, large fair value
differences result from unintentional disclosure errors, we also
consider several alternative explanations for the fair value differ-
ences that do not involve unintentional disclosure errors. Based on
an analysis of hand-collected data for 200 rm-year observations
with large fair value differences, we nd that 20 large fair value
differences are due to data entry errors in the Compustat database.
For the remaining 180 observations, we nd that at most 30 of the
fair value differences are potentially attributable to a combination
Please cite this article in press as: Bratten, B., et al., The accuracy of disclosures for complex estimates: Evidence from reported stock option fair
values, Accounting, Organizations and Society (2015), http://dx.doi.org/10.1016/j.aos.2015.09.001
of (1) use of a valuation model other than the Black-Scholes model,
(2) options issued ineor out-of-the-money, (3) discounts for post-
vesting restrictions, and (4) disclosure of ranges for valuation
model inputs when midpoints are poor proxies for the weighted
averages of these inputs. Collectively, these analyses suggest that a
substantial majority of large fair value differences are likely
attributable to unintentional disclosure errors.
In our third stage, we investigate potential consequences of
inaccurate footnote disclosures by examining how these errors map
into future share-based compensation expense and impact ana-
lysts' earnings forecasts. We nd that the relation between ex-
pected option expense based on reported fair values and future
reported share-based compensation expense is declining in the
absolute fair value difference. This result indicates that, on average,
reported fair values are measured with increasing error as the ab-
solute fair value difference increases. We also nd that the calcu-
lated fair value is positively and signicantly related to the future
expense after controlling for the reported fair value. This result
indicates that the calculated fair value is used to compute option
expense for many rms and the reported fair value is incorrect.
These results suggest that footnote disclosure errors likely impair
investors' abilities to predict future option expense. Focusing on
analysts' forecasts, we nd a positive and signicant relation be-
tween absolute fair value differences and both absolute forecast
errors and forecast dispersion. These results indicate that analysts'
forecasts are negatively affected by the stock option footnote errors
represented by the absolute fair value differences.
Taken together, the results reported in this paper provide evi-
dence that a substantial number of rms have errors in their stock
option footnote. As such, this study makes several contributions for
regulators, nancial statement users and researchers. First, for
regulators, our evidence suggests that material errors occur in stock
option footnote disclosures that result in disclosed information that
is either not relevant or not faithfully representative. This is espe-
cially important given that PCAOB guidance recommends that audit
rms conduct the same comparison of reported and calculated fair
values that we conduct in this study (PCAOB, 2006, p. 26). Further,
these ndings validate concerns that fair value determinations
based on unobservable inputs are particularly challenging for au-
ditors (PCAOB, 2009, p. 5) and that auditors are not adequately
evaluating all of the inputs into an estimate for the collective
impact on the estimate (PCAOB, 2011; Peecher, Schwartz, &
Solomon, 2007; Grifth et al. 2015). The fact that the errors we
identify can be detected with little effort leaves open the question
of the extent of errors in more difcult-to-verify disclosures of
complex estimates.
Second, for nancial statements users, our evidence on the ac-
curacy of the stock option footnote disclosure is important because
these disclosures provide the earliest information about the
magnitude of stock option expense in future years. Indeed, one of
the FASB's objectives for employee stock option footnote disclo-
sures is to enable users of the nancial statements to under-
stand the effect of compensation cost arising from share-based
payment arrangements on the income statement (ASC 718-10-50-
1). Substantial differences between the disclosed and calculated fair
values that cannot be reconciled based on information provided in
the footnote leave nancial statement users with no guidance on
which value is a more accurate measure of the grant date fair value
and, thus, which value to use to estimate future option expense.
Third, this study contributes to prior research. Prior studies
examine either material but infrequent errors (i.e., restatements) or
general measures of nancial reporting quality (e.g., internal con-
trol weaknesses, accruals quality). Unlike these studies, we exploit
the unique reporting requirements for stock options to provide
large sample evidence on the incidence and materiality of errors in
 B. Bratten et al. / Accounting, Organizations and Society xxx (2015) 1e18
a common nancial statement disclosure that are not publicly
disclosed by the rm. This study also contributes to the literature
examining stock option disclosures. Unlike prior studies which
generally investigate bias in the selection of valuation model inputs
(e.g., Aboody, Barth, & Kasznik, 2006; Hodder, Mayew, McAnally, &
Weaver, 2006; Bartov, Mohanram, & Nissim, 2007; Johnston, 2006;
Choudhary, 2011) and the valuation model (Bratten, Jennings, &
Schwab, 2015), we examine the accuracy of those disclosures by
examining whether the reported stock option fair value is inter-
nally consistent with the disclosed valuation inputs and model.
Note that selection of biased inputs alone would not result in the
fair value differences we document as long as the fair value of
options granted is consistently computed using those biased inputs.
Thus, our study complements prior literature to provide a detailed
analysis of the estimation and nancial reporting of employee stock
option fair values.
2. The fair value difference and the likelihood of
unintentional disclosure errors
2.1. The fair value difference
ASC 718-10-50 requires rms to disclose the number and
weighted-average grant date fair value of options granted during
the year. Firms are also required to disclose the valuation method
(e.g., the Black-Scholes model) and the valuation model inputs used
to estimate the fair value of these options. The valuation model
inputs include the weighted-average exercise price, expected life,
expected volatility, expected dividend yield, risk-free rate, and
discount for post-vesting restrictions. To meet these disclosure
requirements, rms must track and report a considerable amount
of information, creating ample opportunities for error.
In this study, we use the required stock option footnote disclo-
sures to compare two alternative measures of the fair value of
employee stock options granted in the current year. The rst is the
weighted-average fair value of options granted as reported in the
rm's notes to their nancial statements. We refer to this value as
the reported fair value (FV Reported). The second is the calculated fair
value based on the weighted-average valuation model inputs dis-
closed in the rm's notes to their nancial statements and the
Black-Scholes option pricing model. We refer to this value as the
calculated fair value (FV Calculated).1 We then use these values to
compute a rm's fair value difference (FV Difference), the difference
between the rm's reported and calculated fair values, scaled by
the calculated fair value [(FV Reported e FVCalculated)/FVCalculated].
It is unlikely that these fair value differences result from pur-
poseful earnings management for several related reasons. First, to
knowingly report incorrect fair values or option model inputs in the
stock option footnote would likely be an SEC violation and risk
serious negative consequences. Second, management is likely to
derive little to no gain from knowingly reporting incorrect fair
values or option model inputs because such misreporting by itself
would not affect reported expense unless the rm also used the
incorrect information reported in the footnote disclosure to
compute stock option expense, a further SEC violation. Third,
fraudulently manipulating the valuation of stock options granted is
likely not even necessary given ndings in prior research that rms
are able to decrease reported option expense (and thus manage
earnings upwards) by exercising discretion when selecting both the
When computing FVCalculated, we assume the current market price of the un-
derlying stock on the grant day is equal to the disclosed option exercise price. This
is consistent with prior research that suggests virtually all rms issue options that
are at the money (Bebchuk et al., 2002; Chang et al., 2013; Hall & Murphy, 2002).
Please cite this article in press as: Bratten, B., et al., The accuracy of disclosures for complex estimates: Evidence from reported stock option fair
values, Accounting, Organizations and Society (2015), http://dx.doi.org/10.1016/j.aos.2015.09.001
valuation model (Bratten et al., 2015) and valuation model inputs
(Aboody et al., 2006; Bartov et al., 2007; Choudhary, 2011; Hodder
et al., 2006; Johnston, 2006). These actions may push the bound-
aries of legitimate reporting, but fall short of illegal activity. Finally,
as we discuss below, we nd that the fair value differences are
skewed toward positive (expense-increasing) values, which would
permanently decrease net income since overstated stock option
expense does not later reverse through net income like most ac-
cruals. For these reasons we focus on the absolute value of FV Dif-
ference,which we label AbsFVDifference, and investigate whether these
absolute fair value differences are due to unintentional disclosure
errors or have alternative, non-error explanations.
Minor differences between the reported and calculated fair
values can occur when valuation model inputs vary across multiple
equal-weighted option grants during the year because the Black-
Scholes model is non-linear in most valuation model inputs (life,
volatility, dividend yield, and risk-free rate). When these inputs
vary across grants, the weighted-average fair value from the model
(the reported fair value) will not equal the calculated fair value
based on the weighted-average inputs (the calculated fair value).
However, as the examples we present in Appendix A demonstrate,
using weighted-average valuation inputs to compute an option fair
value is not likely to produce a fair value difference that is more
than a few percent as long as the different grants are relatively
equally-weighted. We address the potential for larger fair value
differences that can arise from multiple grants that are not equally-
weighted in detail below.
2.2. The likelihood of unintentional disclosure errors
To meet the stock option disclosure requirements each rm
must successfully track multiple variables, often for multiple stock
option grants, input these variables into a complex non-linear
formula and compute and report weighted averages for the
various inputs and the outputs. The steps involved in preparing
these footnote disclosures range from simple (e.g., if the exercise
price for the option equals the closing stock price on the grant date,
nd that price) to more complex (e.g., estimate the expected life of
an option given the previous exercise behavior of employees, the
changes in employee proles, the economic environment and the
terms of the current options granted). There are ample opportu-
nities for error throughout this process. For example, an accountant
preparing the footnote disclosure may compute a weighted-
average as a simple average or not correctly incorporate the effect
of a stock split in the middle of the year. Alternatively, an accoun-
tant may incorrectly report the exercise price of options granted as
the fair value of the options granted.2
Two additional features of the nancial reporting process in the
stock option setting increase the likelihood of errors. First, rms
often engage external valuation consultants to estimate stock op-
tion fair values and other complex estimates because management
lacks the necessary expertise to perform these estimates internally.
This lack of valuation expertise can be problematic given man-
agement is still responsible for accurately preparing the footnote
disclosures related to the estimated stock option fair values. Based
on interviews of auditors and a review of PCAOB inspection reports,
Grifth et al. (2015) argue that a division of knowledge between
auditors and valuation experts is a root cause of challenges arising
when auditing complex estimates because auditors do not have
We report below that many rms do indeed report the exercise price as the fair
value of the options granted. As further evidence of the inherent complexity in
accounting for employee stock options, a PricewaterhouseCoopers guide (PwC,
2013) on how to account for stock-based compensation is 362 pages.
4
B. Bratten et al. / Accounting, Organizations and Society xxx (2015) 1e18
sufcient valuation knowledge to engage in the necessary critical
analysis of management or specialist models. Similar issues likely
exist for managers who prepare the stock option disclosures and
choose to engage an external valuation consultant to assist in the
valuation of stock options rather than acquiring the requisite
knowledge to understand and perform the valuation themselves.
Second, many rms rely on internal spreadsheets when ac-
counting for stock options. Panko and Ordway (2005) summarize
evidence from several sources and report that spreadsheets are
ubiquitous in nancial reporting by U.S. rms. Moreover, prior
empirical studies examining spreadsheet accuracy nd that the
majority of spreadsheets examined have signicant errors. Powell,
Baker, and Lawson (2008) cite studies by KPMG and Coopers &
Lybrand in which both rms found that 91 percent of the spread-
sheets they examined had a signicant error. 3
Various strategies are available to reduce or eliminate these
errors. Indeed, testing spreadsheets is one focus of the imple-
mentation of SarbaneseOxley Act section 404 internal controls
requirements (PwC, 2004). However, studies suggest that such
systematic testing of spreadsheets is not common (Caulkins,
Morrison, & Weidenmann, 2007; Leon, Kalbers, Coster, &
Abraham, 2012). In fact, Deloitte (2009) surveyed over 3000 pro-
fessionals during a 2008 webcast on balancing spreadsheet risks
and found that 70.1 percent of respondents indicated that spread-
sheets were relied on for critical portions of the business, but only
33.9 percent indicated that their rm had specialized techniques to
ensure the spreadsheets were functioning properly.
In addition, PCAOB audit guidance recommends that for com-
panies using the Black-Scholes model the auditor should verify
that the company is using the correct formula and recalculate the
fair value (PCAOB, 2006, p. 26). Following this recommendation for
each stock option grant should help detect many unintentional
errors at the grant level. However, if this recommendation is not
also followed after combining multiple grants, unintentional errors
can still go undetected in the weighted-average footnote
disclosures.
2.3. An example
Before empirically testing the likelihood of unintentional
disclosure errors, we provide an example based on a rm in our
sample reporting an option fair value equal to the option's exercise
price, a likely disclosure error. In their Form 10-K for the scal year
ended December 31, 2010, Hughes Communications disclosed that
both the fair value of options granted and the exercise price of
options granted equaled $53.67 for 2008, $16.77 for 2009 and
$28.99 for 2010. In contrast, the calculated fair values based on the
disclosed inputs for those years equal $24.26 for 2008, $7.25 for
2009 and $11.79 for 2010, resulting in an AbsFV Difference greater than
100 percent in each year. It seems likely that either (1) the reported
fair value is in error and grossly overstates the fair value for options
granted during the year or (2) the exercise price is in error so that
the calculated fair value understates the fair value of the options
granted during the year.
To determine which of these two possibilities is true, we use the
reported fair values and the calculated fair values to estimate future
expense and compare these estimates with the actual expense re-
ported by Hughes Communications. Hughes' disclosures indicate
3 percent and 0.45 percent in 2008, 2009, and 2010, respectively.
To conrm that spreadsheets are used to prepare stock option footnote dis-
closures, we asked ve senior partners at four major audit rms how commonly
spreadsheets are used to prepare stock option footnote disclosures. Three partners
responded very common, one responded virtually universal, and one responded
common.
Please cite this article in press as: Bratten, B., et al., The accuracy of disclosures for complex estimates: Evidence from reported stock option fair
values, Accounting, Organizations and Society (2015), http://dx.doi.org/10.1016/j.aos.2015.09.001
that 2008 is the rst year they issue stock options and that they
made one grant per year in 2008, 2009 and 2010, about four
months into the year. We assume that (a) options granted in the
current year vest over four years with two-thirds of a year occurring
in the grant year, (b) 100,500 options are granted in 2009 (net of
options retired), and (c) forfeitures in 2008 relate to options gran-
ted in 2008, forfeitures in 2009 relate half to options granted in
2008 and half to options granted in 2009, and forfeitures in 2010
relate one third to each of the current and preceding two years.
Using these assumptions and the reported fair values, we estimate
option expense for 2008, 2009, and 2010 to be $4.9 million, $7.7
million, and $10.4 million, respectively. These estimates are much
greater than the actual stock option expense reported by Hughes of
$2.0 million, $3.5 million, and $4.3 million for these three years. In
contrast, using these same assumptions and the calculated fair
values, we estimate option expense for 2008, 2009, and 2010 to be
$2.2 million, $3.5 million, and $4.6 million, respectively. These es-
timates are very close to those reported by Hughes for their actual
option expense, and indicate that Hughes used the calculated value,
not the reported value, to compute future option expense.4
A nancial statement user relying on the reported fair value of
options granted to estimate stock option expense would have had
an error relative to a forecast based on the calculated fair value of
$2.7 ($5.8) million in 2008 (2010), which was nearly 30 (25) percent
of their net income of $9.3 ($23.0) million.5 This analysis indicates
that the reported fair values in Hughes stock option footnotes are
not consistent with reported stock option expense, and that a
nancial statement user relying on reported stock option fair values
would be misled.
3. The sample and descriptive statistics
Our sample is drawn from the Compustat database for scal
years 2004e2011 and includes all rms incorporated in the United
States with available data to compute the fair value difference
(FVDifference). We begin in 2004 because this is the rst year for
which Compustat contains detailed information on the weighted-
average fair value of employee stock option grants during the cur-
rent year as well as the weighted-average inputs to the Black-
Scholes pricing model used to value the current year's option
grants.6 This provides us with an initial sample of 25,439 obser-
vations. We then require the additional explanatory variables used
in our multivariate analysis, resulting in our main sample of 23,358
rm-year observations. For some of our analyses, we also require
lagged Compustat data or analyst forecast data from I/B/E/S,
reducing our sample for these analyses. The sample construction is
detailed in Table 1.
Panel A of Table 2 reports descriptive statistics for the reported
fair value (FVReported), the calculated fair value (FVCalculated), the fair
value difference (FVDifference), the absolute value of the fair value
difference (AbsFVDifference), and measures of the materiality of
4
The Hughes example provides anecdotal evidence of a disclosure error. This
example is unique because Hughes has a short history of option grants and clearly
discloses both the timing of the grants and the subsequent actual stock option
expense. Our ability to perform this calculation is much more limited for the ma-
jority of rms that have long-standing option programs and do not separately
disclose stock option expense.
5
In 2009, the forecast error based on the calculated value is $4.2 million less than
the forecast error based on the reported fair value, and Hughes reported a loss of
$51.6 million. As a percentage of assets, these forecast errors are 0.24 percent, 0.35
6
The Black-Scholes inputs necessary to compute the calculated value of options
granted include life (Compustat OPTLIFE), volatility (OPTVOL), dividend yield
(OPTDR), risk-free rate (OPTRFR), and the exercise price (OPTPRCGR), which we
assume to be equal to the market price of the underlying stock on the grant date.
Table 1
Sample selection.
B. Bratten et al. / Accounting, Organizations and Society xxx (2015) 1e18
Data restrictions
Primary Sample
Starting Compstat sample of U.S. Firms from scal years 2004e2011 with data to compute AbsFV Difference
(i.e., FVOPTGR, OPTLIFE, OPTVOL, OPTDR, OPTRFR, OPTPRCGR)
Less rms without data to compute deteriminants of AbsFVDifference
Future share-based compensation expense sample
Starting sample (i.e., Primary sample)
Less rms without data to compute share-based compensation expense
Less rms without AbsFVDifference in year t-1
Less rms without AbsFVDifference in year t-2
Analyst forecast accuracy and dispersion sample
Starting sample (i.e., Primary sample)
Less pre-SFAS 123R observations
Less rms without I/B/E/S data to compute current and lagged forecast errors
Less rms without data to compute additional control variables
Less rms without I/B/E/S data to compute current and lagged forecast dispersion
Detailed variable denitions are presented in Appendix C.
AbsFVDifference. The distribution of the signed difference (FV Difference) is
centered near zero (median 0.001) but also includes many fair
value differences far from zero. In addition, the mean FV Difference
(0.055) is signicantly positive (p-value < 0.01, untabulated), indi-
cating the distribution is skewed toward positive values. As dis-
cussed above, this distribution is more consistent with
unintentional error than managerial manipulation because manip-
ulation to lower option expense would result in a FV Difference distri-
bution skewed toward negative values (i.e., FVReported < FVCalculated).
Panel A also reports that the mean AbsFV Difference is 11.9 percent.
As a point of reference, this is much larger than the mean error
reported in prior research in auditing. For example, Durney, Elder
and Glover (2014) examine 160 post-SOX audit sampling applica-
tions and report a mean error (computed as the absolute difference
between management's reported value and the auditor's proposed
value, scaled by management's reported value) of 0.2 percent. They
also report a mean error for ten studies conducted prior to SOX of
3.08 percent.
We also report in Table 2 that 23.9 percent of rm-year obser-
vations have an AbsFVDifference greater than ten percent. In some of
our analyses, we divide the sample into two groups: those with
AbsFVDifference greater than ten percent (the high group) and those
with AbsFVDifference less than ten percent (the low group). We
focus on rm-year observations with AbsFVDifference greater than ten
percent because these differences are more likely due to errors than
to some alternative explanation. If we lower (raise) the threshold
for large differences to ve (20) percent, then AbsFV Difference is
large for 33.2 (15.4) percent of the rms in our sample.
We next evaluate the materiality of these absolute differences.
MaterialityIncome (MaterialityAsset) equals the absolute value of the
annual difference between the reported and calculated fair values
multiplied by the number of options granted, scaled by the absolute
value of net income (scaled by total assets). Average Materi-
alityIncome (MaterialityAsset) equals 7.9 (0.7) percent. Moreover,
MaterialityIncome (MaterialityAsset) exceeds one percent for 24.6 (6.7)
percent of the sample. These statistics suggest that AbsFVDifference is
material for many rms.
Panel B of Table 2 presents univariate statistics for materiality
after partitioning the sample into high and low AbsFVDifference groups.
MaterialityIncome (MaterialityAsset) is greater than one percent for 69.5
(25.3) percent of rm-year observations in high AbsFV Difference group
relative to 10.5 (0.8) percent of observations in the low group, and
greater than ve percent for 37.0 (7.2) percent of observations in high
group relative to 2.4 (0.1) percent in the low group.
Panel C of Table 2 presents the distribution of AbsFVDifference over
We also
time. The percentage of observations with AbsFV Difference greater
Please cite this article in press as: Bratten, B., et al., The accuracy of disclosures for complex estimates: Evidence from reported stock option fair
values, Accounting, Organizations and Society (2015), http://dx.doi.org/10.1016/j.aos.2015.09.001
6
5
N Used in table(s)
25,439
2081 23,358 2, 3, 4
23,358
16 23,342
5504 17,838 7
4557 13,281 7
23,358
6563 16,795
7496 9299
2672 6627 8
887 5740 8
than ten percent declines from 32.6 percent in 2004 to 17.2 percent
in 2011. The largest decline, from 29.6 percent in 2005 to 22.6
percent in 2006, coincides with recognition of the fair value of
options granted on the income statement. The last two columns of
Panel C present the percentage of observations with an AbsFV Dif-
greater than one percent of net income for the high and low
ference
AbsFVDifference groups, respectively. For the high (low) group this
percentage declines from 77.7 (16.6) percent in 2004 to 67.9 (8.3)
percent in 2011. Despite this decrease, a substantial portion of the
sample still exhibits a material difference between reported and
calculated fair values in the nal year of our study period.
Finally, we examine whether the large fair value differences are
one-time events or recur through time for individual companies
using a subsample (12,052 rm-year observations) with data
available to compute AbsFVDifference for three consecutive years. The
rst column of panel D reports that for this sample 22.3 percent of
the rm-year observations in the rst year are in the high AbsFV-
Difference group and 41.4(21.8) percent of these observations are still
in the high group in the second year (both the second and third
years). Random assignment of rms to the high and low groups in
subsequent years based on the frequency in the current year would
produce 22.3 percent of the current-year high group still in the high
group in the second year, and 5.0 percent (0.223 0.223) still in the
high group in both the second and third years. Thus, we observe
nearly twice the expected frequency in the second year and more
than four times the expected frequency in the third year. 7 The
second and third columns of panel D present results separately for
observations with positive and negative fair value differences in the
rst year. The results are very similar to those reported for the
combined sample, with somewhat greater stickiness for positive
fair value differences.
Taken together, these results suggest that large fair value differ-
ences are not random events, but recur with the same sign across time
for many rms. In the next section we investigate whether the fair
value differences are likely due to unintentional reporting errors
arising from complex option programs, lower quality nancial
reporting, and challenging accounting environments. We also inves-
tigate alternative potential explanations for fair value differences that
do not necessarily suggest the presence of disclosure errors.
7
regressing rms' AbsFVDifference in year t on their AbsFVDifference in year t-1. The co-
efcient on the prior-year AbsFVDifference is positive and signicant (b 0.3338; p-
value
< 0.01),the
estimated providing evidence
persistence of that AbsFV
absolute Difference exhibits some persistence from
fair value differences by
one year to the next.
 B. Bratten et al. / Accounting, Organizations and Society xxx (2015) 1e18

Table 2
Univariate statistics: the magnitude and materiality of fair value differences.

Panel A: basic univariate statistics

Variable N Mean P5 P10 Q1 Median Q3 P90 P95

FVReported 23,358 7.369 0.460 0.930 2.360 5.260 9.790 15.390 19.400
FVCalculated 23,358 7.180 0.456 0.905 2.304 5.086 9.579 15.023 18.960
% of sample % of sample
FVDifference FVDifference
Variable N Mean P5 P10 Q1 Median Q3 P90 P95 <10% >10%
FVDifference 23,358 0.055 0.199 0.095 0.016 0.001 0.018 0.217 0.549 9.6% 14.2%
% of sample
AbsFVDifference
Variable N Mean P5 P10 Q1 Median Q3 P90 P95 >10%
AbsFVDifference 23,358 0.119 0.001 0.001 0.004 0.017 0.091 0.326 0.578 23.9%
% of sample % of sample
Materiality Materiality
Variable N Mean P5 P10 Q1 Median Q3 P90 P95 >1% >5%
MaterialityIncome 23,358 0.079 0.000 0.000 0.000 0.001 0.010 0.056 0.151 24.6% 10.7%
MaterialityAsset 23,358 0.007 0.000 0.000 0.000 0.000 0.001 0.005 0.016 6.7% 1.8%

Panel B: materiality univariate statistics in high and low AbsFV Difference subsamples

Variable AbsFVDifference> 10% subsample AbsFVDifference <10% subsample

N Mean Median >1% >5% N Mean Median >1% >5%

MaterialityIncome 5574 0.298 0.028 69.5% 37.0% 17,784 0.010 0.001 10.5% 2.4%
MaterialityAsset 5574 0.028 0.002 25.3% 7.2% 17,784 0.001 0.000 0.8% 0.1%

Panel C: absolute fair value differences (AbsFV Difference) over Time

Year N Mean P5 P10 Q1 Median Q3 P90 P95 Full sample AbsFVDiff>10% AbsFVDiff <10%

% Of sample % Of sample % Of sample

AbsFVDifference Materiality Income MaterialityIncome

>10% >1% >1%

2004 2902 0.194 0.001 0.001 0.005 0.029 0.165 0.487 0.870 32.6% 77.7% 16.6%
2005 3481 0.147 0.001 0.002 0.006 0.026 0.131 0.420 0.707 29.6% 74.9% 14.1%
2006 3259 0.110 0.001 0.001 0.005 0.016 0.081 0.312 0.516 22.6% 69.1% 11.5%
2007 3169 0.113 0.001 0.001 0.005 0.016 0.087 0.305 0.545 23.1% 69.1% 9.8%
2008 2945 0.111 0.001 0.001 0.004 0.018 0.088 0.307 0.572 23.2% 62.0% 8.9%
2009 2710 0.104 0.001 0.001 0.003 0.014 0.076 0.303 0.539 21.8% 62.1% 7.2%
2010 2566 0.081 0.000 0.001 0.003 0.011 0.052 0.223 0.416 17.8% 63.8% 7.3%
2011 2326 0.073 0.000 0.001 0.003 0.011 0.050 0.211 0.402 17.2% 67.9% 8.3%
All Years 23,358 0.119 0.001 0.001 0.004 0.017 0.091 0.326 0.578 23.9% 69.5% 10.5%

Panel D: stickiness of fair value differences in subsample with three consecutive years of data availability (N 12,052)

Large absolute diff. Large positive diff. Large negative diff.

AbsFVDifference > 10% FVDifference > 10% FVDifference < 10%

% of subsample with large differences in the rst year 22.3% 10.7% 11.6%
% of rst-year large differences with a large difference in the second year 41.4% 51.2% 32.3%
% of rst-year large differences with a large difference in the second and third year 21.8% 23.5% 13.8%

The sample includes rm-year observations from 2004 to 2011 with sufcient disclosure information to re-calculate the fair value of stock options granted, and to estimate
Equation (1). FVDifference equals the difference between the reported (FV Reported) and calculated (FVCalculated) fair value of stock options granted, scaled by the calculated fair value
(FVCalculated). FVReported equals OPTFVGR as reported by Compustat. FV Calculated is the calculated fair value based on the Black-Scholes valuation model and the disclosed valuation
model inputs reported by Compustat. AbsFV Difference equals the absolute value of FVDifference. MaterialityIncome (MaterialityAsset) equals the ratio of absolute value of the difference
between the reported and calculated stock option fair values multiplied by the number of option granted to net income (total assets) [Abs(FV Reported e FVCalculated)*# Options
Granted/Net Income] ([Abs(FVReported - FVCalculated)*# Options Granted/Assets]). Panel D is limited to a subsample of rms for which there are at least three consecutive rm-
year observations.

4. Do large fair value differences result from disclosure
errors?
In this section, we conduct a variety of analyses to determine
whether the fair value differences documented above appear to
result from disclosure errors.
4.1. Direct evidence of disclosure errors
The most direct evidence we provide comes from two anomalies
in the stock option footnote itself that suggest a disclosure error.
First, we observe that 1109 rm-year observations (4.7 percent of
the sample) disclose a reported fair value equal to the exercise price
of the options granted. While not mathematically impossible, the
option fair value will not equal the exercise price unless one or
more of the other model inputs are both incorrect and highly un-
usual. For example, in the Hughes Communications example dis-
cussed above, both of these values could be correct only if the
weighted-average volatility was 340 percent, the expected life
was 230 years, or the weighted-average risk-free rate was 500
percent. As such, this equality suggests an incorrect transfer of data
from a spreadsheet or other source to the footnote. The mean

Please cite this article in press as: Bratten, B., et al., The accuracy of disclosures for complex estimates: Evidence from reported stock option fair
values, Accounting, Organizations and Society (2015), http://dx.doi.org/10.1016/j.aos.2015.09.001

B
S

absolute fair value difference is 59.3 (8.9) percent for observations
for which the reported fair value equals (does not equal) the ex-
ercise price. These means are signicantly different (p-
value < 0.01).
Second, among 17,959 rm-year observations with available
data, we observe that 1932 observations (10.8 percent of this sub-
sample) exhibit a repeat disclosure in which they disclose the
same volatility, risk-free rate, exercise price, or grant date fair value
for option grants in consecutive years. While it is not impossible for
these values to be the same from one year to the next, these values
are generally expected to vary from year to year, so a repeated
disclosure for one of these values may indicate that a value was not
properly updated. For observations with (without) a repeat
disclosure, the mean absolute fair value difference is 13.0 percent
(9.5 percent). These means are signicantly different (p-
value < 0.01). Both of these anomalies suggest that disclosure errors
are a likely source of large fair value differences.
4.2. Additional large sample evidence of disclosure errors
Next we estimate Equation (1) below to examine whether fair
value differences vary with rm characteristics indicating more
complex option programs, lower quality nancial reporting, or a
more challenging nancial reporting environment. Because these
characteristics represent conditions that are more prone to
disclosure errors, this analysis provides additional, large sample
evidence on whether fair value differences represent disclosure
errors.
AbsFVDifference a0 a1RankandFile% a2Cancellation%
a3AbsDEmployees a4Split a5Extent
a6Post123R a7Restate a8ICW a9CAO
a10Big4 a11 Acquisition a12 Restructure
a13Size a14 Age a15 MissingICW
a16MissingExecucomp
(1)
The rst ve variablesdRank-and-File%, Cancellation%, Abs-
DEmployees, Split, and Extentdcapture features that likely increase
the complexity of the rm's stock option program and the
weighted-average computations. Rank-and-File% equals the ratio of
non-executive (i.e., rank-and-le) stock options granted to total
options granted in the current year. Cancellation% equals the ratio of
options canceled, forfeited and expired during the current year to
options outstanding at the beginning of the year. Abs DEmployees
equals the absolute value of the percent change in employee count
in the current year relative to the previous year. Split is an indicator
variable that equals one if the rm split its stock in the current year
and zero otherwise. Extent equals the annual decile rank of the ratio
of the fair value of the stock options granted in the current year to
market value of equity.
If larger fair value differences represent disclosure errors, we
expect larger differences for rms with more difcult weighted-
average computations and more complex disclosures. As such, we
expect larger differences for rms (a) with stock option programs
that include more employees (Rank-and-File%), (b) with higher
employee turnover (Cancellation%), (c) with a growing or shrinking
workforce (AbsDEmployees), and (d) that have split their stock (Split).
This suggests positive coefcients on these four variables. In the case
of Extent, on one hand, rms that use stock options extensively are
more likely to have more complicated programs that may be more
prone to error, resulting in larger fair value differences. On the other
hand, rms that use stock options extensively may have more
Please cite this article in press as: Bratten, B., et al., The accuracy of disclosures for complex estimates: Evidence from reported stock option fair
values, Accounting, Organizations and Society (2015), http://dx.doi.org/10.1016/j.aos.2015.09.001
8
7
experience in making these complex calculations and may be more
careful with their disclosures. As a result, we make no prediction for
the sign of the coefcient for Extent.
The next ve variablesdPost123R, Restate, ICW, Big4 and
CAOdcapture nancial reporting quality. Post123R equals one after
the implementation of the SFAS 123R (FASB, 2004) requirement to
recognize an expense for the fair value of options granted and zero
otherwise. ICW equals one if the rm had an internal control
weakness in the current year and zero otherwise. Restate equals
one if the rm restated current year earnings and zero otherwise.
Big4 equals one if the rm engages a Big 4 auditor in the current
year, and zero otherwise. CAO equals one if the rm has an ofcer
with the title Chief Accounting Ofcer among its top ve executives
as reported by Execucomp.
If large fair value differences represent disclosure errors, we
expect that large differences are more likely for rms with lower
nancial reporting quality. In the case of Post123R, prior research
nds that investors perceive recognized stock option expense to be
more reliable than disclosed stock option expense (Frederickson,
Hodge, & Pratt, 2006; Johnston, 2006), so we expect a negative
coefcient estimate for Post123R.8 A weak internal control system
(ICW) increases the likelihood of an undetected error affecting a
rm's reported nancial statements or their notes, while a
restatement (Restate) provides direct evidence of a failure in the
rm's accounting system. Thus, we expect positive coefcients for
Restate and ICW. Prior research provides evidence that audit quality
is associated with larger audit rms (Ashbaugh-Skaife, Collins, &
Kinney, 2007; Becker, Defond, Jiambalvo, & Subramanyam, 1998;
Palmrose, 1988; Teoh & Wong, 1993), and higher quality audits
are likely to detect more errors before the nancial statements are
published. As a result, we expect a negative coefcient on Big4.
Finally, CAO captures the stature of the person responsible for ac-
counting decisions in the rm. Rhodes and Russomanno (2013) nd
that rms with higher prole chief accounting ofcers exhibit
higher reporting quality. As a result, we expect a negative coef-
cient on CAO.
We also include four independent variablesdAcquisition,
Restructure, Size, and Aged that prior literature nds to be asso-
ciated with restatements of nancial reporting errors (Cao, Myers,
& Omer, 2012; Kinney, Palmrose, & Scholz, 2004) or material in-
ternal control weaknesses (Ashbaugh-Skaiffe et al., 2007; Doyle,
Ge, & McVay, 2007; Ge & McVay, 2005). Acquisition equals one if
the rm acquired another rm during the current year, and zero
otherwise. Restructure equals the cumulative dollar amount of
restructuring charges taken in the current and previous year
scaled by the market value of equity. Size equals the log of the
market value of the rm's equity. Age equals the log of the number
of years the rm has available CRSP stock return data. Based on
prior research we expect positive coefcients on Acquisition and
Restructure and a negative coefcient on Age if fair value differ-
ences represent disclosure errors. In the case of Size, because this
variable may proxy for either more complex option programs or
more developed accounting systems, we make no directional
prediction.
Finally, we include two additional variables to account for the
fact that several of the variables included in the regression (i.e., ICW,
Rank-and-File%, and CAO) do not have data available for many ob-
servations in the sample. Rather than reduce our sample for
8
more reliable in a variety of other settings including nonpension postretirement
benets (Davis-Friday, Folami, Liu, & Mittelstaedt, 1999; Davis-Friday, Liu, & Mit-
Prior telstaedt, 2004),
studies also derivative
provide nancial
evidence instruments
that recognized (Ahmed, Kilic,
information & Lobo,
is viewed as 2006) and
asset revaluations (Cotter & Zimmer, 2003).
 B. Bratten et al. / Accounting, Organizations and Society xxx (2015) 1e18

Table 3
Determinants of absolute fair value differences: Univariate analyses.

Variable Pred. (High e low) High Low Difference (high e low) Sig. diff.

AbsFVDifference AbsFVDifference

Mean Mean

Stock option program complexity variables

Rank-and-File% 0.2167 0.3191 0.1024
Cancellation% 0.1274 0.0926 0.0348 ***
AbsDEmployees 0.2006 0.1475 0.0531 ***
Split 0.0678 0.0540 0.0138 ***
Extent ? 5.2090 4.2783 0.9307 ***
Weak accounting system quality variables
Post123R e 0.6453 0.7522 0.1069 ***
ICW 0.0983 0.0697 0.0286 ***
Restate 0.0814 0.0557 0.0258 ***
CAO e 0.0459 0.0897 0.0438 ***
Big4 e 0.5640 0.7262 0.1621 ***
Additional variables correlated with accounting system quality
Acquisition 0.4591 0.4892 0.0301
Restructure 0.0113 0.0099 0.0014 **
Size ? 5.2381 6.1229 0.8849 ***
Age e 2.5571 2.7639 0.2068 ***
Variables to account for missing observations
Missing(Execucomp) ? 0.7076 0.5417 0.1658 ***
Missing(ICW) ? 0.2377 0.1149 0.1228 ***
N 5574 17,784

*** and ** indicates that the difference between the means in the high and low group are signicant at the one and ve percent level, respectively, using a one-tailed test when
predications are made, and a two-tailed test otherwise. The sample includes rm-year observations from 2004 to 2011 with sufcient disclosure information to re-calculate
the fair value of stock options granted, and to estimate Equation (1). High (Low) AbsFVDifference observations indicate those with AbsFVDifference greater than (less than or equal to)
ten percent. All variables are dened in Appendix C. Continuous variables are winsorized at the 1st and 99th percentiles.

observations that lack data for these variables, when one of these
variables is missing we set the variable's value to zero and include
an indicator variable set equal to one when the data is missing and
zero otherwise. Missing(ICW) is the variable set equal to one when
ICW is missing. Since Rank-and-File% and CAO are missing because
of limitations in Execucomp coverage, Missing(Execucomp) is set
equal to one when Rank-and-File% and CAO are missing. We make
no predictions for these indicator variables. 9
Table 3 presents univariate statistics for our regression variables
for the high and low AbsFV Difference groups. We nd that rms
exhibiting a large AbsFVDifference have more complex stock option
programs, as indicated by more stock option cancellations
(Cancellation%), greater changes in the size of their workforce
(AbsDEmployees), splitting their stock (Split), or more extensive use
of employee stock options (Extent). Large fair value difference rms
are also more likely to have lower accounting quality as indicated
by disclosing rather than recognizing stock option expense
(Post123R), reporting an internal control weakness (ICW), restating
earnings (Restate), not having a Chief Accounting Ofcer among
their top ve executives (CAO), or engaging a non-Big4 auditor
(Big4). Finally, we also nd that large fair value difference rms
exhibit more restructuring of operations (Restructure) and are
smaller (Size) or younger (Age), all indicators from previous
research of internal control weaknesses. Taken together, this
pattern of results suggests that internal inconsistencies in stock
option footnote disclosures captured by large values for AbsFV Dif-
ference are unintentional disclosure errors.
The rst column of Table 4 presents the results from estimating
Equation (1). Overall, the results are consistent with the univariate
Data needed to compute ICW and Execucomp-based variables (i.e., Rank-and-
File% and CAO) are missing for 5541 and 13,578 observations, respectively. Including Act enacted in July 2010 also caused managers to account for employee stock op-
Missing indicator variables has been used to preserve sample size when certain
independent variables would substantially reduce the sample size (e.g., see
Himmelberg, Hubbard, & Palia, 1999; Hanlon & Slemrod, 2009). If we omit these
variables from the regression, results are qualitatively similar for the remaining
variables.
results reported in Table 3. In particular, all of the coefcients are
highly signicant in the predicted direction, with the exception of
Restructure and Size. We also note that the two characteristics that
were not signicant in the univariate analysis, Rank-and-File% and
Acquisition, are highly signicant in the predicted direction in the
multivariate analysis.10
We conduct three additional analyses to conrm our results and
interpretation. First, we add two variables, SamePrice and Repeat-
Disclosure, to Equation (1) that control for the two anomalous
disclosures discussed above that suggest a likely disclosure error.
SamePrice is an indicator variable equal to one if the rm discloses
the same grant date fair value and exercise price for stock options
granted in year t, and zero otherwise. RepeatDisclosure is an indi-
cator variable equal to one when the risk-free rate, the expected
volatility, the exercise price, or the reported grant date fair value is
identical in year t to the same disclosure in year t-1, and zero
otherwise. By adding these additional controls for SamePrice and
RepeatDisclosure, we are ensuring that our results are not purely
driven by the rm-years exhibiting the most egregious disclosure
errors. We report the results in column two of Table 4. The co-
efcients on SamePrice and RepeatDisclosure are positive and sig-
nicant. In addition, all inferences with respect to the other
variables are unchanged, except that the coefcient on Size is now
signicant.11
Second, we estimate Equation (1) using a logistic regression
The signicant negative coefcient on Post123R indicates that AbsFV Difference is
smaller after rms were required to expense employee stock options, consistent
with rms being more focused on accounting for employee stock options that are
expensed rather than only disclosed. However, it is possible that other regulatory
actions such as new executive compensation disclosure rules issued by the SEC in
August 2006 and the DoddeFrank Wall Street Reform Act and Consumer Protection
tions more carefully and contributed to this result.
11
observations for which we do not have data for RepeatDisclosure. If we omit this
variable from the regression, results are qualitatively similar for the remaining
variables.
This regression also includes a variable, Missing(RepeatDisclosure), for the 5399

Please cite this article in press as: Bratten, B., et al., The accuracy of disclosures for complex estimates: Evidence from reported stock option fair
values, Accounting, Organizations and Society (2015), http://dx.doi.org/10.1016/j.aos.2015.09.001

B
S
 9

Table 4 received only three replies. The most revealing reply came from a
Determinants of absolute fair value differences: multivariate analysis.
particularly candid CFO who noted that the disclosures had been
Variable Pred. (1) (2) prepared under his predecessor and audited by their external
Coef. (p-value) Coef. (p-value) auditor. He agreed the disclosures were in error. He stated that
using the company records he and his staff were unable to
Intercept ? 0.1211 (<0.001) 0.0522 (<0.001)
determine how the error had occurred. He also noted that pre-
Rank-and-File% 0.0382 (<0.001) 0.0369 (<0.001)
Cancellation% 0.0978 (<0.001) 0.0764 (<0.001)
paring these disclosures was beyond the technical capacity of his
AbsDEmployees 0.0375 (<0.001) 0.0220 (0.008) staff and that they now outsource this function to a professional
Split 0.0191 (0.008) 0.0226 (<0.001) accounting rm. The second reply conrmed that the company
Extent ? 0.0069 (<0.001) 0.0037 (<0.001) uses a lattice model, which we had noted in the email as a po-
Post123R e 0.0387 (<0.001) 0.0186 (<0.001)
Restate 0.0257 (0.004) 0.0196 (0.010) tential source of the difference. The nal reply corrected a
ICW 0.0192 (0.004) 0.0155 (0.012) misinterpretation of their footnote disclosure that accounted for
CAO e 0.0133 (0.008) 0.0117 (0.005) their large fair value difference. The other 47 rms we contacted
Big4 e 0.0341 (<0.001) 0.0220 (<0.001) neither admitted to an error or offered an alternative explanation
Acquisition 0.0078 (0.025) 0.0113 (<0.001)
for their fair value difference.
Restructure 0.0332 (0.527) 0.0348 (0.233)
Size ? 0.0012 (0.392) 0.0055 (<0.001)
To systematically examine a variety of alternative, non-error
Age e 0.0134 (<0.001) 0.0165 (<0.001) explanations for large fair value differences, we then randomly
Missing(Execucomp) ? 0.0275 (0.001) 0.0311 (<0.001) selected 200 rm-year observations from each of the high and low
Missing(ICW) ? 0.0447 (<0.001) 0.0279 (<0.001)
AbsFVDifference groups. For each observation, we hand-collected from
SamePrice 0.4781 (<0.001)
RepeatDisclosure 0.0198 (<0.001)
the rm's 10-K the valuation model, valuation model inputs (both
Missing(RepeatDisclosure) ? 0.0325 (<0.001) standard inputs, e.g. risk-free rate, volatility, etc., and less common
N 23,358 23,358 inputs, e.g., post-vesting restrictions, etc.), and whether the rm
Adj R2 0.0524 0.2161
issues ineor out-of-the money options (if disclosed). We use this
The sample is discussed in Table 1. The dependent variable is AbsFVDifference. All data to test a variety of alternative explanations and report the
variables are dened in Appendix C. Continuous variables are winsorized at the 1st results in Table 5.
and 99th percentiles. Standard errors are clustered by rm. One-tailed p-values are
The rst potential explanation we examine is whether some
presented when predictions are made.
large absolute fair value differences may be due to data errors in the
Compustat database. To test for this, we compare the Compustat
data with our hand-collected data and determine how many large
fair value differences are accounted for by replacing the Compustat
with HighAbsFVDifference as the dependent variable, where High- data with our hand-collected data. This procedure reduces the
AbsFVDifference equals one when AbsFVDifference is greater than 0.10, absolute fair value difference below ten percent for 20 of the 200
and zero otherwise. By focusing on large absolute fair value dif- observations in the high AbsFVDifference group.12
ferences, we ensure that our results are not due to variation in The second potential explanation we examine is whether some
smaller differences that can occur when multiple grants are issued large absolute fair value differences are due to use of a valuation
in the same year and the inputs vary across grants. The results model other than the Black-Scholes option pricing model (e.g., a
(untabulated) are qualitatively similar to those reported in Table 4. lattice model) to estimate the fair value of employee stock options.
Third, we estimate Equation (1) separately for rms with negative We nd only 14 (9) observations in our high (low) fair value dif-
and positive fair value differences. If the large fair value differ- ference group do not use the Black-Scholes model, and two of the
ences are due to unintentional disclosure errors, we expect that non-Black-Scholes rms in the high group also had a Compustat
the independent variables in Equation (1) should explain both error.
positive and negative fair value differences. For the positive sub- The third explanation we examine is whether large fair value
sample, we nd similar results to those reported in Table 4. For the differences result from options issued in or out of the money. If an
negative subsample we also nd similar results with the excep- option is issued in or out of the money, our calculated value, which
tions of the coefcients on Split and Acquisition, which are no is based on the assumption that options are issued at the money,
longer signicant, and the coefcient on Extent, which is negative will be incorrect and potentially result in a large fair value differ-
and signicant. ence. A careful reading of the 400 footnotes in our hand-collected
Taken together, the results in Table 3, Table 4 and the three sample reveals only 4 (1) observations in the high (low) group
additional analyses indicate that large fair value differences are with explicit disclosures indicating that options were not granted at
more likely for rms with more complex option programs, lower the money. While the absence of explicit statements that options
nancial reporting quality, and a more challenging nancial are issued in or out of the money does not denitively disprove this
reporting environment. These results are consistent with large fair explanation, prior research suggests that virtually all rms issue
value differences likely resulting from unintentional errors. options that are at the money (Bebchuk, Fried, & Walker, 2002;
Chang, Chen, & Fuh, 2013; Hall & Murphy, 2002). As such, it is

4.3. Consideration of alternative non-error explanations

12
The results presented above suggest that, on average, large fair
below ten percent for 20 of the 200 high AbsFVDifference observations, we identied
value differences result from unintentional disclosure errors. Next,
data entry errors for 43 (9) rm-year observations within the high (low) AbsFVDif-
we investigate whether non-error explanations may explain some Note ference subsamples.
that while This
correction of suggests
Compustata Compustat
errors only data error
reduces entry
the rate
AbsFV of 13 percent
Difference

large fair value differences. We rst selected fty random rms
with scal year ends in 2010 and 2011 and AbsFV Difference greater
than ten percent and sent an e-mail to investor relations at each of
these companies explaining in detail our calculation of AbsFV Dif-
ference. We asked them for help in understanding the difference but
(52/400) overall for these data. This error rate is generally higher than data entry
error rates documented in prior research examining the data integrity of nancial
accounting data (Kern & Morris, 1994; Kinney & Swanson, 1993; Mills, Newberry, &
Novack, 2003; San Miguel, 1977; Shi & Zhang, 2012; Thomas & Swanson, 1986). This
is likely due to the difculty of collecting detailed data from non-standard nancial
statement footnote disclosures.

Please cite this article in press as: Bratten, B., et al., The accuracy of disclosures for complex estimates: Evidence from reported stock option fair
values, Accounting, Organizations and Society (2015), http://dx.doi.org/10.1016/j.aos.2015.09.001

10
 B. Bratten et al. / Accounting, Organizations and Society xxx (2015) 1e18

Table 5
Determinants of absolute fair value differences: explanations not suggesting disclosure errors.

Variable Pred. (High e low) High Low Difference (high e low)

AbsFVDifference AbsFVDifference

Frequency Frequency

Compustat error 20 n/a n/a

Non-black scholes 14 9 5
NotAtMoney 4 1 3
Post-vesting discount 2 0 2
Range only re-weighting 14 n/a n/a
Overlapping explanations (4)
Potential alternative explanations 50
N 200 200

The sample used in this table consists of 400 rm-year observations from our main sample for which we hand-collected Form 10-K data for the purpose of evaluating
alternative explanations for high absolute fair value differences. High (Low) AbsFV Difference indicates an AbsFV Difference greater than (less than or equal to) 0.10. A Compustat Error
is determined to be present when a high fair value difference is observed based on Compustat data but a high fair value difference is not observed after replacing Compustat
data with hand-collected valuation model inputs and the reported grant date option fair value. Non-Black Scholes indicates that the rm disclosed using a stock option
valuation model other than a Black-Scholes model. NotAtMoney indicates an explicit disclosure that options were granted with a strike price higher than or lower than the
exercise price. Post-Vesting Discount indicates that the rm disclosed discounting the computed fair value for post-vesting restrictions or post-vesting forfeitures. Range Only
Re-Weighting indicates observations for which a high fair value difference is observed based on using the midpoint of the range-only disclosure but a high fair value difference
is not observed after weighting the high (low) ends of range-only disclosures for each input to 25 (75), 75 (25), 90 (10), or (10) 90. Overlapping explanations indicates the
number of observations for which multiple alternative explanations were present.

unlikely that ineor out-of-the-money options explain a meaningful
portion of large fair value differences.
The fourth explanation we examine is whether large fair value
differences result from a discount for post-vesting restrictions such
as restrictions on the sale or transfer to a third party after vesting
(ASC 718-10-30-10), or for expected post-vesting forfeitures. 13 This
required disclosure is not collected by Compustat, so our calculated
value will not reect this discount. We nd that two observations in
the high group disclosed such a discount, but both also used a
model other than the Black-Scholes model to value their stock
options.
The fth explanation we examine is whether large fair value
differences result from multiple option grants within a single year.
As explained above and in Appendix A, multiple option grants with
different valuation model inputs are not likely to result in an
AbsFVDifference greater than a few percent as long as the grants are
relatively equally-weighted. However, multiple option grants in the
same year that are not equally-weighted can result in large fair
value differences if (1) the valuation model inputs are substantially
different across the different grants and (2) the rm discloses only a
range for the inputs rather than a weighted average. 14 Under these
circumstances, Compustat reports the midpoint of the range as the
disclosed input, but the midpoint may be a poor estimate for the
weighted average.
We identied 50 (40) observations in the high (low) group not
accounted for by earlier explanations that disclosed a range but not a
weighted average for at least one of the Black-Scholes model inputs.
For the 50 observations in the high group, we recomputed the
calculated value, weighting the end points of the disclosed range four
different ways (25/75, 75/25, 90/10, and 10/90) as an alternative to the
50/50 weighting used in our main analysis. We then computed an
alternative AbsFVDifference for each observation for each of these
alternative calculated values. Only 14 of the 50 observations in the
high group subjected to this analysis had an alternative AbsFV Difference
Pre-vesting forfeitures are not a plausible alternative explanation for fair value
differences because, although they reduce the amount of expense the rm recog-
nizes, they cannot affect the grant-date fair value.
This was the main explanation suggested during several independent conver-
sations with Big 4 audit partners. ASC 718 requires disclosure of a weighted average
for volatility and dividend yield when those are allowed to change over the life of
the option when using a lattice model, but does not explicitly require disclosure of a
weighted average when these inputs vary across multiple option grants made
during the same year.
less than ten percent for at least one of these calculations.15
After accounting for observations exhibiting more than one
potential non-error alternative explanation, these results indicate
that at most 50 of the 200 observations with large fair value dif-
ferences (25 percent) are potentially accounted for by a combina-
tion of these alternative explanations. This leaves the majority of
large fair value differences still unexplained and is consistent with
our interpretation of the results in Table 4d unintentional disclo-
sure errors are likely the primary source of the large fair value
differences we observe.16
4.4. A closer view of the fty largest fair value differences
Table 6 provides details for the observations with the largest 25
positive (Panel A) and negative (Panel B) fair value differences from
our hand-collected sample. The rst three columns report AbsFV-
Difference (as a percentage), the dollar amount (in millions) of fair
value difference and the difference as a percent of assets. Of the 25
positive (negative) fair value differences 14 (8) are greater than one
percent of total assets. The next two columns report potential ex-
planations for the large fair value differences.17 Only two (ve) of
the positive (negative) fair value differences exhibit no error-based
explanation, consistent with the results reported in the previous
section. Further, 16 (16) of the positive (negative) fair value differ-
ences exhibit only an error-based explanation, and many of them
exhibit multiple error-based explanations.
The remaining four columns present reported stock option
expense (in millions), forecast errors based on reported and
calculated fair values, respectively, and an indicator of whether the
This represents an upper limit because it is unlikely that all 14 observations
made such extreme grants.
16
disclosure errors were corrected in subsequent disclosures. To do this, we examined
177 subsequent-year 10-Ks that we were able to locate for the 200 rms with large
fair
We value
also differences.
examined We found
the extent that 18
to which therms madeobservations
rm-year substantive changes to their
with likely
footnote disclosures by adding or removing information. However, if we use the
newly disclosed information to re-estimate the prior year's fair value difference,
only three of the rms would no longer have a large fair value difference. Note that
none of these 18 rms that changed their disclosure discussed the change or drew
any attention to the change.
As possible error-based explanations we only use the following variables from
Equation (1) that indicate either a disclosure error or a specic weakness in
nancial reporting: SamePrice, ICW, Restate, Pre123R (dened as one minus
Post123R), and NotBigN (dened as one minus BigN).

Please cite this article in press as: Bratten, B., et al., The accuracy of disclosures for complex estimates: Evidence from reported stock option fair
values, Accounting, Organizations and Society (2015), http://dx.doi.org/10.1016/j.aos.2015.09.001

B
S
Table 6
Small sample examination of the source of large fair value differences.

11

Obs. No. AbsFVDifference: Explanations suggesting: Reported stock opt. exp. Forecast error based More accurate forecast
on:

(%) ($) % of Assets No error Disclosure error FVReported FVCalc

Panel A: largest 25 positive fair value differences

1 494.2% 10.968 1.29% a 4
2 246.0% 24.951 0.32% a, b 4
3 235.4% 2.540 51.13% 5
4 224.8% 0.680 2.32%
5 219.6% 7.255 4.05% 4
6 212.3% 0.188 0.34% b 5
7 189.8% 4.441 0.22% a
8 187.2% 0.163 9.24%
9 168.6% 0.025 0.11% 5
10 158.4% 9.794 5.58% 4, 5
11 155.6% 1.551 4.51% 5 0.503 109.5% 8.0% FVCalc
12 154.2% 0.086 0.07% 1
13 152.3% 0.541 1.57% 3, 4 0.102 439.5% 9.1% FVCalc
14 152.0% 0.076 0.01% a 4
15 150.7% 6.127 14.96% 4
16 143.4% 1.910 2.25% 2, 3, 4, 5
17 138.8% 3.087 0.02%
18 133.2% 8.277 2.58%
19 130.0% 3.316 1.58% a 5
20 125.7% 0.389 1.09% 1, 5
21 125.2% 0.208 0.56% a 1, 2, 5
22 124.6% 24.670 0.36%
23 122.8% 7.779 10.11% 1, 4
24 122.1% 0.719 0.16% 4, 5
25 121.2% 1.220 0.08% 5 0.247 9.8% 20.8% FVReported
Panel B: largest 25 negative fair value differences
1 66.3% 601.812 0.05% a
2 62.5% 18.860 24.68% 2
3 61.6% 0.071 1.11% 5 0.076 7.2% 1.0% FVCalc
4 59.5% 0.730 0.32% 5
5 57.9% 0.008 0.01% b 5
6 56.6% 1.214 0.43% 5
7 55.4% 0.469 2.74% 0.456 2.4% 104.3% FVReported
8 54.8% 0.436 6.29% a 5
9 50.2% 0.025 0.01% 5
10 49.8% 0.585 0.02% 4 0.994 51.2% 34.8% FVCalc
11 47.9% 0.670 7.92% 5 0.189 38.1% 61.2% FVReported
12 42.7% 6.084 16.38% e 5
13 42.5% 0.045 0.01% 4
14 40.8% 0.980 0.28% 0.604 24.1% 46.0% FVReported
15 39.8% 10.474 1.04% 2, 3, 4
16 39.0% 0.007 0.00% 5
17 38.6% 0.457 0.49% b 5
18 38.4% 0.123 0.17% 3, 5
19 37.3% 1.260 0.03%
20 35.4% 271.186 0.44% a
21 35.0% 1.910 0.00%
22 33.9% 0.022 0.01% 5 0.051 32.6% 32.6% Tie
23 33.2% 0.253 0.02% 4 0.472 47.8% 31.0% FVCalc
24 33.1% 1.825 0.04% 2, 3 2.4 15.7% 34.7% FVReported
25 32.8% 1.368 3.55% 3, 5 0.74 4.3% 66.9% FVReported

This table provides details on (1) potential explanations for rm-year observations with the largest 25 positive and negative fair value differences from our hand-collected
sample and (2) how the fair value difference may effect forecasted stock option expense for rms that disclose stock option expense. Notation for potential non-error ex-
planations follows: Compustat data error (a), fair value estimated using a non-Black-Scholes valuation model (b), options not issued at the money (c), potential additional post-
vesting forfeiture assumptions (d), and disclosure of input ranges but no weighted-average (e). Notation for potential error explanations follows: Fair value equals exercise
price (1), ICW (2), Restatement (3), Disclosed, Not Expensed (4), and Non-Big4 Auditor (5).

forecast based on the reported or calculated values results in the
smaller forecast error. Reported stock option expense was hand-
collected from the 10-K and is missing for many rms that report
only total stock-based compensation expense and do not separately
disclose the component attributable to stock options. The forecasts
were based on hand-collecting reported and calculated fair values
for as many prior years as necessary to account for current period
expense depending on the expected life of the options granted. In
each case, we assumed that current grants did not contribute to
current expense unless there was disclosure to the contrary. As a
result, the forecasted expense is equal to the sum of prior year
grants divided by the expected life for that year's grant, multiplied
by one minus the estimated forfeiture rate where the estimated
forfeiture rate is equal to the sum of forfeitures over the estimation
period divided by the sum of options granted over the estimation
period. These data requirements permitted us to estimate a stock-
option expense forecast for three (nine) of the positive (negative)
fair value observations.
The accuracy of these forecasts is limited by simplifying as-
sumptions about the lives of the options, the exact date of grants,

Please cite this article in press as: Bratten, B., et al., The accuracy of disclosures for complex estimates: Evidence from reported stock option fair
values, Accounting, Organizations and Society (2015), http://dx.doi.org/10.1016/j.aos.2015.09.001

12
 B. Bratten et al. / Accounting, Organizations and Society xxx (2015) 1e18

and the effect of forfeitures. Even so, we nd that forecast errors for Table 7
six of the 12 observations are less than ten percent. For these six, Do fair value differences map into a rm's recognized expense?

the forecast error is smaller using reported values half the time and Variable (1) (2) (3)
using calculated values half the time. For the six observations for Coeff. Coeff. Coeff.
which the smaller forecast error is greater than ten percent, the
(p-value) (p-value) (p-value)
reported forecast errors are smaller in some cases and the calcu-
lated forecast errors are smaller in others. Taken together the Intercept 0.0060 0.0059 0.0043

pattern presented in this table indicates that it is not possible, ex
ante, to predict whether a particular rm with a large fair value
difference will use the reported fair value or the calculated fair
value to compute future stock option expense. These results are
also consistent with large fair value differences that are largely due
to unintentional disclosure errors that may occur in the reported
fair value or may occur in the assumptions used to compute the
calculated fair value.
5. The consequences of fair value differences
In this section we rst examine whether the fair value differ-
ences have an effect on the relation between current stock option
grants and future stock-based compensation expense. We then
examine whether fair value differences are associated with the
accuracy of analysts' forecasts of future earnings and the dispersion
of analysts' earnings forecasts.
5.1. Fair value differences and future expense
The accuracy of the stock option footnote is important because
this footnote provides the earliest information available to nan-
cial statement users about the magnitude of stock option expense
in future years. Indeed, one of the FASB's objectives for employee
stock option footnote disclosures is to enable users of the nan-
cial statements to understand the effect of compensation cost
arising from share-based payment arrangements on the income
statement (ASC 718-10-50-1). If the reported fair value is accurate
and errors in the stock option footnote are conned to mis-
reporting the Black-Scholes model inputs, these errors may not
reduce the usefulness of the reported grant date fair value for
estimating future option expense. However, substantial errors in
the reported fair values of the options granted would leave in-
vestors with little guidance for estimating future option expense
unless they used the disclosed inputs to determine the calculated
fair value and knew to use that value rather than the reported fair
value.
To provide evidence on this issue, we estimate the following OLS
regression:
ShareBasedCompExpt a0 a1 ReportedGrantValuet1
a2AbsFVDifference;t1
(2)
a3ReportedGrantValuet1
*AbsFVDifference;t1
where ShareBasedCompExp equals share-based compensation
(<0.001) (<0.001) (<0.001)
ReportedGrantValuet-1 0.7322 0.2057 0.1732
(<0.001) (<0.001) (0.043)
AbsFVDifference,t-1 0.0009
(0.447)
ReportedGrantValuet-1*AbsFVDifference,t-1 0.2132
(<0.001)
ReportedGrantValuet-2 0.0677
(0.202)
CalculatedGrantValuet-1 0.5113 0.2850
(<0.001) (0.001)
CalculatedGrantValuet-2 0.2368
(<0.001)
Adj R2 0.5885 0.5935 0.6396
N 17,838 17,838 13,281
The sample consists of rm-year observations with sufcient data to re-compute the
fair value of stock options granted and sufcient data to estimate Equations (2) and
(3). The dependent variable is ShareBasedCompExp, which equals the share-based
compensation reported by the company, scaled by total assets. Reported-
GrantValue equals the product of the reported option fair value and the number of
options granted (i.e., the rm's estimated future option expense at the grant date),
scaled by total assets. CalculatedGrantValue equals the calculated fair value multi-
plied by the number of options granted (i.e., our estimated future option expense at
the grant date), scaled by total assets. Subscripts indicate the year. Standard errors
are clustered by rm. Two-tailed p-values are presented.
reported fair value is not used to compute future expense. To aid in
interpreting the coefcients, consider two scenarios. First, if the
reported fair value is used to compute future expense, we expect a
signicant positive coefcient on ReportedGrantValue and an
insignicant coefcient on the ReportedGrantValue*AbsFVDifference
interaction. Alternatively, if the calculated fair value is used to
compute future expense, we may still observe a signicant positive
coefcient on ReportedGrantValue, but we should also observe a
signicant negative coefcient on the Reported-
GrantValue*AbsFVDifference interaction. This would be consistent with
measurement error in the reported fair value that is increasing in
the absolute fair value difference. This would be particularly
problematic because these errors would prevent nancial state-
ment users from using the reported grant date fair values to
reasonably estimate future stock option expense.
We present the results from estimating Equation (2) in the rst
column of Table 7. The coefcient estimate for Reported-
GrantValuet-1 is positive and signicant, indicating that fair values
are related to future expense. However, the coefcient estimate for
the ReportedGrantValue*AbsFVDifference interaction is negative and
signicant, indicating greater errors in reported fair values as they
deviate more from the calculated fair value. Thus, these results
indicate that the reported fair values are in error for many rms in
our sample with reported fair values that deviate substantially from

(including disclosed option expense in years prior to SFAS 123R's the calculated fair value. This suggests that for some rms with
large fair value differences the calculated fair values are used to
requirement to recognize an expense for the fair value of options),
compute future expense. To test this directly, we estimate the
scaled by total assets. ReportedGrantValue equals the product of the
following OLS regression:
reported option fair value and the number of options granted,
scaled by total assets. 18
Thus, this analysis provides evidence on the extent to which the
ShareBasedCompExpt a0 a1ReportedGrantValuet1
a2 CalculatedGrantValuet1 (3)
18

share-based compensation as our dependent variable in Equation (2) which in- where CalculatedGrantValue equals the product of the calculated
cludes stock option expense and other share-based compensation. fair value and the number of options granted, scaled by total assets.
Compustat does not separately report stock option expense. As such, we use

Please cite this article in press as: Bratten, B., et al., The accuracy of disclosures for complex estimates: Evidence from reported stock option fair
values, Accounting, Organizations and Society (2015), http://dx.doi.org/10.1016/j.aos.2015.09.001

B
S

We present the results from estimating Equation (3) in the
second column of Table 7. The coefcient estimate for Reported-
GrantValuet-1 is positive and signicant. However, the coefcient
estimate for CalculatedGrantValuet-1 is more than twice the
magnitude of the coefcient for ReportedGrantValuet-1 and is also
positive and signicant. This provides evidence that calculated fair
values are not noisy error around an accurate reported fair value,
but rather that for many rms the calculated fair value is used to
compute future expense. In the third column, we estimate Equation
(3) after including an additional lag (from year t-2) for both inde-
pendent variables. Results are similar to the second column; the
coefcient estimates for the calculated grant values are greater than
the coefcient estimates for the reported grant values. Collectively,
these results suggest that nancial statement users who rely on the
reported fair value will be unable to reasonably estimate future
stock option expense for many rms.
5.2. Fair value differences and analyst forecast accuracy and
dispersion
In this section we examine whether the fair value differences we
observe are related to the accuracy and dispersion of analysts'
forecasts. If nancial analysts use the information in the stock op-
tion footnote in forming their forecasts and, as suggested above, fair
value differences increase the difculty of forecasting future option
expense, we anticipate that larger fair value differences will
decrease (increase) the accuracy (dispersion) of analyst forecasts.
To examine this, we estimate the following regression:
AbsFEor Dispersion t1 a0 a1AbsFVDiffPerSharet
Controlst (4)
where AbsFE proxies for forecast accuracy and equals the absolute
value of the difference between the median forecast for earnings
per share for year t1 and the IBES actual for earnings per share for
year t1.19 To ensure that all nancial information is publicly
available, we compute the median forecast in the fourth month
following the scal year end for year t. Dispersion proxies for fore-
cast dispersion and equals the standard deviation of individual
forecasts for earnings per share for year t1. Similar to AbsFE, we
base dispersion on forecasts made during the fourth month
following the scal year end for t. The main independent variable,
AbsFVDiffPerShare, is equal to the absolute value of (FVReported e
FVCalculated) times the number of options granted, divided by shares
outstanding. AbsFE, Dispersion, and AbsFVDiffPerShare are all scaled
by stock price at the end of year t. For this analysis we use only
observations made in the post-SFAS 123R period because analyst
forecasts of earnings prior to SFAS 123R did not include an expense
for employee stock options.
We include a variety of additional variables to control for rm
characteristics that are known to be associated with forecast accu-
racy and forecast dispersion. Specically, earnings forecasts are less
accurate and/or more disperse for rms that are smaller (Size), have
larger earnings surprises (Surprise), experience a loss in the current
year (Loss), are more nancially-distressed (Distress), have more
volatile earnings (StdDevROE), and have a lower analyst following
(LnAnalysts). Forecasts are also generally less accurate and/or more
disperse as the forecast horizon grows (Horizon). Because forecast
accuracy and dispersion are somewhat sticky over time, we also
We examine the relation between unsigned forecast errors and unsigned fair
value differences because, as indicated by the Hughes example above and the re-
sults presented in Tables 5 and 6, we cannot ex ante determine whether the re-
ported or the calculated fair value is in error or which will be used to compute
future option expense.
Please cite this article in press as: Bratten, B., et al., The accuracy of disclosures for complex estimates: Evidence from reported stock option fair
values, Accounting, Organizations and Society (2015), http://dx.doi.org/10.1016/j.aos.2015.09.001
14
13
control for the prior year forecast accuracy and dispersion. 20
The results from estimating Equation (4) using AbsFE (Dispersion)
as the dependent variable are presented in column 1 (column 2) of
Table 8. In both specications, we nd a positive and signicant
coefcient estimate for AbsFVDiffPerShare. This suggests that fair
value differences are associated with increased forecast errors and
greater forecast dispersion when the stock option footnote infor-
mation is not internally consistent. To assess the economic signi-
cance of these results, we compute the magnitude of the absolute
fair value difference per share (AbsFVDiffPerShare) necessary to
result in a full one cent increase in a rm's absolute forecast error per
share. This corresponds to the 83rd percentile of the AbsFVDiffPer-
Share distribution, which suggests that about 17 percent of stock
option footnotes have potential disclosure errors that are associated
with an analyst forecast error of one cent or more. 21 Taken together
the results presented in Table 8 indicate that analyst behavior is
affected by the internal inconsistency in stock option footnote dis-
closures that are attributable to unintentional errors.
6. Conclusion
This study examines the internal consistency of employee stock
option disclosures to provide evidence on the accuracy of supporting
footnote disclosures for complex estimates. We examine the fre-
quency and magnitude of differences between (1) the reported
weighted-average fair value for employee stock options granted
during the year and (2) the calculated fair value using the disclosed
weighted-average option model inputs and the Black-Scholes option
pricing model. We nd that 23.9 percent of observations with
available data have reported and calculated option fair values that
differ by more than ten percent, and that these differences are sticky
and are often material as a percentage of net income.
We then examine potential explanations for these differences.
First, we identify two anomalous disclosures that are likely errors
and nd that these are more common for rms with large fair value
differences. We also nd that fair value differences are larger for
rms with more complex stock option programs or with reporting
and operating environment characteristics that indicate lower
quality nancial reporting. In contrast, we nd limited evidence
that large fair value differences are due to explanations that do not
imply disclosure errors.
Finally, we provide evidence on the consequences of these un-
intentional disclosure errors. First, we provide evidence that, for
many rms, the reported stock option fair value is measured with
error making it a less useful predictor of future expense. This in-
dicates that nancial statement users who rely on this disclosure to
forecast future stock option expense may be misled. Next, we
provide evidence that analysts' forecasts are less accurate and more
disperse across analysts for rms with larger fair value differences,
suggesting that they are indeed being misled by these uninten-
tional disclosure errors.
These results are important because they provide the rst large
sample evidence on the accuracy of supporting footnote disclosures
for complex estimates used increasingly in nancial reporting. The
fact that the errors we identify can be detected with little effort
leaves open the question of the extent of errors in more difcult-to-
verify disclosures. Moreover, the accuracy of the stock option
For evidence on the determinants of analyst forecast accuracy and dispersion,
see Kross, Ro, and Schroeder (1990), Lang and Lundholm (1996), Brown (2001),
Eames and Glover (2003), and Behn, Choi, and Kang (2008).
21
1) AbsFVDiffPerShare, so that if DAbsFE 0.0100 and b1 0.9131, then the
AbsFVDiffPerShare associated with a one cent increase in AbsFE is 0.0110, which
This corresponds to
computation the 83rd
is based on percentile of the
the equation: AbsFVDiffPerShare
DAbsFE b1 (Table 8,distribution.
Column
 B. Bratten et al. / Accounting, Organizations and Society xxx (2015) 1e18

Table 8
The relation between absolute fair value differences and analyst forecast accuracy and dispersion.

Variable Pred. Dependent variable AbsFE Dependent variable Dispersion

Coef. Coef.

(p-value) (p-value)

AbsFVDiffPerShare 0.9131 0.3654

(0.012) (0.001)
Size - 0.0070 0.0022
(<0.001) (<0.001)
Surpise 0.0003 0.0001
(<0.001) (<0.001)
Loss 0.0124 0.0046
(<0.001) (<0.001)
Distress 0.0009 0.0002
(<0.001) (<0.001)
Horizon 0.0064 0.0022
(0.598) (0.759)
StdDevROE 0.0035 0.0014
(0.041) (<0.001)
LnAnalysts - 0.0016 0.0008
(0.060) (0.868)
EPS ? 0.0001 0.0001
(0.808) (0.251)
AbFEt-1 0.3165
(<0.001)
Dispersiont-1 0.4056
(<0.001)

N 6627 5740
Adj. R2 0.2988 0.4649

This sample consists of rms that have data necessary to compute our proxies for forecast accuracy, the absolute value of forecast errors (AbsFE), and forecast dispersion, the
standard deviation of analyst forecasts (Dispersion), as well as known determinants of forecast accuracy and dispersion. All variables are dened in Appendix C. Continuous
variables are winsorized at the 1st and 99th percentiles. Standard errors are clustered by rm. One-tailed p-values are presented when predictions are made.

footnote disclosure is important because it provides the earliest
information to nancial statement users about the magnitude of
future stock option expense.
Appendix A. Multiple grants
Minor differences between the reported and calculated fair
values occur when there are multiple equal-weighted option grants
during the year and the valuation model inputs vary between the
different grants. This does not occur if only the stock price differs
between grants because the Black-Scholes model is linear in price.
However, the Black-Scholes model is non-linear in other valuation
model inputs (life, volatility, dividend yield, and risk-free rate). As a
result, when these inputs vary across equal-weighted grants, the
weighted-average fair value from the model (the reported fair
value) will not equal the fair value that results from inputting the
weighted-average inputs into the model (the calculated fair value).
The examples presented in Table A1 demonstrate the effect of
inputs that vary across multiple equal-weighted option grants. The
rst base case presents a scenario in which a rm makes two
equally-sized option grants with a grant date stock price equal to the
exercise price of $20.00 for both grants and the following valuation
model inputs for both grants: dividend yield equal to two percent;
risk-free rate equal to four percent; expected option life equal to ve
years; and volatility equal to 25 percent. These option grants have a
weighted-average reported fair value and a calculated fair value both
equal to $4.71 resulting in a fair value difference of zero.
The following four cases consider alternative scenarios for these
two grants, each with extreme values for one input and average
values for the other inputs. For example, case 1 assumes two grants
that have volatility of 10 percent and 40 percent (weighted average
of 25 percent) and the rest of the inputs as in the base case. For this
case, the reported weighted-average option value is $4.70, which is
less than the calculated option value by less than one percent.
Similarly, in cases two, three, and four, dividend yield, risk-free rate
and expected life are separately allowed to vary between the two
option grants, resulting in absolute fair value differences less than
three percent. These examples suggest that, when a rm makes
multiple equal-weighted option grants, the effect of using the
weighted average of valuation inputs to compute an option fair
value (rather than averaging fair values computed separately using
the inputs for each grant) is not likely to produce a fair value dif-
ference that is more than a few percent.
We also examined various combinations of multiple input
changes, most of which did not result in a fair value difference
greater than our threshold of ten percent. If we allow all of the
inputs to vary simultaneously by the amounts indicated in Table A1
and only in directions that maximize the difference in the fair
values of the two grants, we observe a fair value difference greater
than ten percent (20.6 percent). However, this example has very
large differences in model inputs for option grants made by the
same company in the same year where these differences are all in a
fair-value-difference-maximizing direction, which is a highly un-
likely occurrence. In addition, 15.1 percent of the observations in
our sample (untabulated) report a fair value difference greater than
even this extreme fair value difference of 20.6 percent.

Please cite this article in press as: Bratten, B., et al., The accuracy of disclosures for complex estimates: Evidence from reported stock option fair
values, Accounting, Organizations and Society (2015), http://dx.doi.org/10.1016/j.aos.2015.09.001

B
S
Table A1
Examples of the effect on the fair value difference of varying inputs between two equally-weighted option grants.

15

Case Current price Exercise price Volatility Dividend yield Risk-free rate Life

Base Grant-1 20.00 20.00 25 2.0 4.0 5.0

Grant-2 20.00 20.00 25 2.0 4.0 5.0
FVReported $4.71
FVCalculated $4.71
FVDifference 0.0%
1-Volatility Grant-1 20.00 20.00 40 2.0 4.0 5.0
Grant-2 20.00 20.00 10 2.0 4.0 5.0
FVReported $4.70
FVCalculated $4.71
FVDifference 0.2%
2-Dividend Yield Grant-1 20.00 20.00 25 0.0 4.0 5.0
Grant-2 20.00 20.00 25 4.0 4.0 5.0
FVReported $4.83
FVCalculated $4.71
FVDifference 2.5%
3-Risk-Free Rate Grant-1 20.00 20.00 25 2.0 2.0 5.0
Grant-2 20.00 20.00 25 2.0 6.0 5.0
FVReported $4.73
FVCalculated $4.71
FVDifference 0.4%
4-Life Grant-1 20.00 20.00 25 2.0 4.0 3.0
Grant-2 20.00 20.00 25 2.0 4.0 7.0
FVReported $4.58
FVCalculated $4.71
FVDifference 2.8%

Each case compares the reported value with the calculated value for two equally-weighted employee stock option grants. The base case presents a scenario in which the
inputs are identical for both grants. The next four cases each change one of the four inputs so that the weighted-average for that input is the same, but the values for that input
for each stock option grant are very different. The four inputs permitted to change are the expected volatility, the expected dividend yield, the expected risk-free rate, and the
expected life of the option. The reported fair value (FV Reported) is the (equally) weighted average of the Black-Scholes option value for each of the two grants. The calculated fair
value (FVCalculated) is the Black-Scholes option value based on the (equally) weighted average for each input. The fair value difference (FV Difference) is the reported fair value minus
the calculated fair value divided by the calculated fair value.

Appendix B. Excerpt from 10-K of Hughes communications
for the period ending December 31, 2010
Note 16: Employee share-based payments
Stock option program
On April 24, 2008, our Compensation Committee made stock
options awards under the Plan (the Stock Option Program), which
consisted of the issuance of non-qualied stock options to em-
ployees of the Company and its subsidiaries. A total of 1,250,000
options (the Option Pool) have been authorized under the Stock
Option Program for option awards during the period of April 24,
2008 to December 31, 2011. The grant and exercise price of the stock
options is the closing price of the Company's common stock on the
date of the grant.
Any options forfeited or canceled before exercise will be
deposited back into the Option Pool and will become available for
award under the Stock Option Program. In accordance with the
terms of the Stock Option Program, the Compensation Committee
delegated to our Chief Executive Ofcer (CEO) and President the
authority to award options, at his discretion, to the current and
future employees of the Company and its subsidiaries. Each grant
has a 10 year life and vests 50% on the second anniversary of the
grant date and 25% on each of the third and fourth anniversaries of
the grant date. The fair value of each option award was estimated
on the date of grant using a Black-Scholes option valuation model
based on the assumptions noted in the table below.
Since we became a public registrant in February 2006, we did
not have sufcient history to measure expected volatility using our
own stock price history nor have the history to compute the
expected term of the stock options at the time of issuing stock
options to employees. As a result, we utilized an average volatility
based on a group of companies identied as our peers. We esti-
mated the expected term of the stock options, which is closely
aligned with the identied peer group, based upon the current
anticipated corporate growth, the currently identied market value
of the stock price at issuance and the vesting schedule of the stock
options. The risk-free interest rate is based on the published U.S.
Treasury Yield Curve as of the grant date for the period of 5 years
which most closely correlates to the expected term of the option
award. Dividend yield is zero as we have not, nor do we currently
plan to, issue dividends to our shareholders.
On March 19, 2009, we offered eligible participants in the Stock
Option Program the opportunity to exchange (the Exchange
Offer) all or a portion of their eligible outstanding stock options for
new stock options, on a one-for-one basis, through an exchange
offer, which expired on April 16, 2009. Each new option (the New
Option) has an exercise price of $14.47, which was the closing price
of our common stock on April 15, 2009, and a new vesting schedule
to reect the new grant date of April 16, 2009.
As a result of the Exchange Offer, which was completed on April
16, 2009, 546,900 outstanding stock options (representing 100%
participation) were exchanged, and the estimated fair value of the
New Options of $2.3 million was computed using a Black-Scholes
option valuation model based on the new grant date. The
compensation expense related to the New Options is recognized on
a straight-line basis over the four-year vesting period beginning on
the date of grant.
The key assumptions for the option awards for the years ended
December 31, 2010, 2009 and 2008 are as follows:

Please cite this article in press as: Bratten, B., et al., The accuracy of disclosures for complex estimates: Evidence from reported stock option fair
values, Accounting, Organizations and Society (2015), http://dx.doi.org/10.1016/j.aos.2015.09.001

16
 B. Bratten et al. / Accounting, Organizations and Society xxx (2015) 1e18

Year ended December 31,

2010 2009 2008

Volatility range 45.33%e45.90% 45.97%e47.92% 47.60%e55.0%

Weighted-average volatility 45.40% 47.62% 47.67%
Expected term 5 years 5 years 5 years
Risk-free interest rate range 1.18%e2.59% 1.71%e2.20% 3.15%e1.50%
Weighted-average risk-free interest rate 1.20% 1.79% 3.14%

Hughes communications, inc. notes to the consolidated straight-line basis over the four-year vesting period beginning on
nancial statements (continued) the date of grant. We recorded $4.3 million, $3.5 million and $2.0
million compensation expense for the years ended December 31,
A summary of option activity under the Stock Option Program is 2010, 2009 and 2008, respectively. As of December 31, 2010, we had
presented below: $12.6 million of unrecognized compensation expense for non-

Option shares Weighted- Weighted average remaining contractual life Aggregate

average intrinsic
exercise price value(1)

Outstanding at January 1, 2008 e $ e

Granted 562,400 $ 53.67
Forfeited or expired (10,000) $ 54.00
Outstanding at December 31, 2008 552,400 $ 53.67 9.32 $ e
Retired (546,900) $ 54.00
Granted 647,400 $ 16.77
Forfeited or expired (4850) $ 20.84
Outstanding at December 31, 2009 648,050 $ 16.77 9.37 $ 6326
Granted 572,500 $ 28.99
Forfeited or expired (22,350) $ 19.56
Exercised (1250) $ 17.03
Outstanding at December 31, 2010 1,196,950 $ 22.56 9.04 $ 21,412
Vested and expected to vest at December 31, 2010 1,077,255 $ 22.56 9.04 $ 19,271
Exercisable at December 31, 2010 1250 $ 17.03 7.96 $ 29

In thousands.

The weighted-average grant date fair value of options granted
during the years 2010, 2009 and 2008 were $28.99, $16.77 and
$53.67, respectively. The total intrinsic value of options exercised
for the year ended December 31, 2010 was minimal. The compen-
sation expense related to stock option awards is recognized on a
vested stock options, which is expected to be recognized over a
weighted average period of 3.08 years. As of December 31, 2010,
there were 1250 stock options outstanding and exercisable, which
have an exercise price of $17.03.

Appendix C. Variable denitions

Variable Denition (data source is Compustat unless indicated)

Measuring fair value differences
FVReported The rm's reported stock option fair value (FVOPTGR)
FVCalculated Calculated stock option fair value based the Black-Scholes valuation model and the weighted-average valuation model inputs disclosed in the rm's
stock option footnote; the valuation model inputs include the weighted-average exercise price (OPTPRCGR), expected life (OPTLIFE), expected
volatility (OPTVOL), expected dividend yield (OPTDR), risk-free rate (OPTRFR).
FVDifference [(FVReported) e (FVCalculated)]/FVCalculated.
AbsFVDifference The absolute value of FVDifference
HighAbsFVDifference An indicator variable equal to one if the rm's AbsFV Difference in the current year is greater than or equal to ten percent, and zero otherwise
MaterialityIncome The product of AbsFVDifference and the number of stock options granted in the current year divided by the absolute value of net income
[AbsFVDifference*OPTGR/abs(NI)]
MaterialityAsset The product of AbsFVDifference and the number of stock options granted in the current year divided by total assets (AbsFV Difference*OPTGR/AT)
Stock option disclosure variables
SamePrice An indicator variable equal to one when the disclosed fair value (OPTFVGR) and exercise price of stock options granted during the current year
(OPTPRCGR) are equal, and zero otherwise
RepeatDisclosure An indicator variable equal to one when the disclosed risk free rate (OPTRFR), volatility (OPTVOL), fair value (OPTFVGR), or exercise price of stock
options granted during the current year (OPTPRCGR) the same as the disclosed value for the same item in the prior year, and zero otherwise
Determinants of fair value differences
Stock option program complexity variables
Extent The annual decile rank of the ratio of the fair value of the stock options granted in the current year to the market value of equity
Rank-and-File% One minus the ratio of the sum of options issued to a rm's executives in the current year [sum(Execucomp OPT_AWARDS_NUM)] to total options
granted in the current year (OPTGR)
Cancellation% The ratio of options canceled in the current year (OPTCA) to options outstanding at the beginning of the year (OPTOSBY)

Please cite this article in press as: Bratten, B., et al., The accuracy of disclosures for complex estimates: Evidence from reported stock option fair
values, Accounting, Organizations and Society (2015), http://dx.doi.org/10.1016/j.aos.2015.09.001

(continued )
B. Bratten et al. / Accounting, Organizations and Society xxx
(2015) 1e18
17

Variable Denition (data source is Compustat unless indicated)

AbsDEmployees The absolute value of employee growth [(EMPt - EMPt-1)/EMPt-1]

Split An indicator equal to one if the rm experienced a stock split in the current year (based on share adjustment data from CRSP), and zero otherwise
Weak accounting system quality variables
Post123R An indicator variable equal to one following the enactment of SFAS 123R (scal years beginning after June 15, 2005), and zero otherwise
ICW An indicator variable equal to one if the rm disclosed an internal control weakness was present in the current year, and zero otherwise (as reported
by Audit Analytics)
Restate An indicator variable equal to one if the rm restated current year earnings, and zero otherwise (as reported by Audit Analytics)
CAO An indicator equal to one if the rm's top ve most compensated executives includes a chief accounting ofcer (as reported by Execucomp), and zero
otherwise
Big4 An indicator variable equal to one if the rm engages a Big 4 auditor in the current year, and zero otherwise
Additional variables correlated with accounting system quality
Acquisition An indicator variable equal to one if the rm acquired another rm in the current year, and zero otherwise
Restructure The cumulative dollar amount of restructuring charges taken in the current and previous year scaled by the current year market value of equity
Size The log of the market value of the rm's equity (PRCC_FF*CSHO)
Age The log of the number of years the rm has available CRSP stock return data
Predicting future option expense
ShareBasedCompExp Share-based compensation (including implied option expense in years prior to SFAS 123R's requirement to expense the grant-date fair value of
options), scaled by total assets
ReportedGrantValue The product of the reported stock option fair value (FV Reported) and the number of options granted, scaled by total assets
CalculatedGrantValue The calculated reported fair value (FV Calculated) multiplied by the number of options granted, scaled by total assets
Forecast accuracy and dispersion analysis
AbsFE The absolute value of the forecast error. Forecast error equals the actual t1 earnings (IBES ACTUAL) minus the median of each individual analyst's
latest forecast (IBES VALUE) made available in the fourth month following the end of year t, divided by shares outstanding (IBES SHOUT) during the
fourth month following the end of year t, scaled by the stock price at the end of year t-1 (IBES PRICE).
Dispersion The standard deviation of the individual analyst's latest forecasts for earnings per share for year t1 available in the fourth month following the scal
year end for year t, scaled by the stock price at the end of year t-1 (IBES PRICE).
AbsFVDiffPerShare The absolute value of (FVReported e FVCalculated) times the number of options granted, divided by shares outstanding (IBES SHOUT) during the fourth
month following the end of year t, scaled by the stock price at the end of year t-1 (IBES PRICE).
Surprise This year's earnings minus last years' earnings deated by stock price [(NI t e NIt-1)/PRCC_F]
Loss An indicator equal to one for rm-year observations with positive net income (NI), and zero otherwise
Distress The Altman's Z-Score, multiplied by negative one so that higher values indicate greater distress [Altman's Z 1.2*(WCAP/AT) 1.4*(RE/
AT) 3.3*((PI XINT)/AT) 0.6*((PRCC_F * CSHO)/LT) 0.9*(SALE/AT)]
Horizon The natural log of the average of the number of calendar days between forecast announcement dates (IBES ANNDATS) for forecast made during the
aggregation window and subsequent scal year end (IBES FPEDATS)
StdDevROE The standard deviation of return on equity (NI/CEQ) over the previous three years
LnAnalysts The natural log of the number of analysts following the rm (as reported by IBES)
EPS Earnings per share (EPSFX)

References End User Computing, 19(3), 1e23.

Chang, C., Chen, L., & Fuh, C. (2013). The pricing of risk and sentiment: a study of
Aboody, D., Barth, M. E., & Kasznik, R. (2006). Do rms understate stock option- executive stock options. Financial Management, 42(1), 79e99.
Choudhary, P. (2011). Evidence on differences between recognition and disclosure:
based compensation expense disclosed under SFAS 123? Review of Accounting
a comparison of inputs to estimate fair values of employee stock options.
Studies, 11(4), 429e461.
Ahmed, A. S., Kilic, E., & Lobo, G. J. (2006). Does recognition versus disclosure Journal of Accounting and Economics, 51(1e2), 77e94.
Cotter, J., & Zimmer, I. (2003). Disclosure versus recognition: the case of asset re-
matter? Evidence from value-relevance of banks' recognized and disclosed
valuations. Asia-Pacic Journal of Accounting and Economics, 10, 81e99.
derivative nancial instruments. The Accounting Review, 81(3), 567e588.
Davis-Friday, P. Y., Folami, L. B., Liu, C.-S., & Mittelstaedt, H. F. (1999). The value
Ashbaugh-Skaife, H., Collins, D. W., & Kinney, W. R. (2007). The discovery and
relevance of nancial statement recognition vs. disclosure: evidence from SFAS
reporting of internalcontrol deciencies prior to SOX-mandated audits. Journal
No. 106. The Accounting Review, 74(4), 403e423.
of Accounting and Economics, 44(1e2), 166e192.
Davis-Friday, P. Y., Liu, C., & Mittelstaedt, H. F. (2004). Recognition and disclosure
Bartov, E., Mohanram, P., & Nissim, D. (2007). Managerial discretion and the eco-
reliability: evidence from SFAS No. 106. Contemporary Accounting Research,
nomic determinants of the disclosed volatility parameter for valuing ESOs.
21(2), 399e429.
Review of Accounting Studies, 12(1), 155e179.
Bebchuk, L. A., Fried, J. M., & Walker, D. I. (2002). Managerial power and rent Deloitte. (2009). Spreadsheet management. Not what you gured. Deloitte.
Doyle, J., Ge, W., & McVay, S. (2007). Determinants of weaknesses in internal control
extraction in the design of executive compensation. The University of Chicago
over nancial reporting. Journal of Accounting and Economics, 44(1e2),
Law Review, 69, 751e846.
Becker, C. L., Defond, M. L., Jiambalvo, J., & Subramanyam, K. R. (1998). The effect of 193e223.
Durney, M., Elder, R., & Glover, S. (2014). Field data on accounting error rates and
audit quality on earnings management. Contemporary Accounting Research,
audit sampling. Auditing: A Journal of Practice and Theory, 33(2), 79e110.
15(1), 1e24.
Eames, M., & Glover, S. (2003). Earnings predictability and the directing of analysts'
Behn, B. K., Choi, J., & Kang, T. (2008). Audit quality and properties of analyst
earnings forecast errors. The Accounting Review, 78(3), 707e724.
earnings forecasts. The Accounting Review, 83(3), 327e349.
Financial Accounting Standards Board (FASB). (2004). Share based payment. In
Bratten, B., Gaynor, L. M., McDaniel, L., Montague, N. R., & Sierra, G. E. (2013). The
audit of fair values and other estimates: the effects of underlying environ- Statement of nancial accounting standards No. 123 (revised). Norwalk, CT: FASB.
Frederickson, J. R., Hodge, F. D., & Pratt, J. H. (2006). The evolution of stock option
mental, task, and auditor-specic factors. Auditing: A Journal of Practice and
accounting: disclosure, voluntary recognition, and management disavowels.
Theory, 32(Suppl. 1), 7e44.
Bratten, B., Jennings, R., & Schwab, C. M. (2015). The effect of using a lattice model to The Accounting Review, 81(5), 1073e1093.
Ge, W., & McVay, S. (2005). The disclosure of material weaknesses in internal
estimate reported option values. Contemporary Accounting Research, 32(1),
control after the SarbaneseOxley Act. Accounting Horizons, 19(3), 137e158.
193e222.
Brown, L. (2001). How important is past analyst forecast accuracy? Financial Analyst Grifth, E. E., Hammersley, J. S., & Kadous, K. (2015). Audits of complex estimates as
verication of management numbers: how institutional pressures shape prac-
Journal, 57(6), 44e49.
Cao, Y., Myers, L. A., & Omer, T. C. (2012). Does company reputation matter for tice. Contemporary Accounting Research, 32(3), 833e863.
Hall, B. J., & Murphy, K. (2002). Stock options for undiversied executives. Journal of
nancial reporting quality? Evidence from restatements. Contemporary Ac-
Accounting and Economics, 33(1), 3e42.
counting Research, 29(3), 956e990.
Hanlon, M., & Slemrod, J. (2009). What does tax aggressiveness signal? Evidence
Caulkins, J. P., Morrison, E. L., & Weidenmann, T. (2007). Spreadsheet errors and
from stock price reactions to news about tax shelter involvement. Journal of
decision making: evidence from eld interview. Journal of Organizational and
Public Economics, 93(1e2), 126e141.

Please cite this article in press as: Bratten, B., et al., The accuracy of disclosures for complex estimates: Evidence from reported stock option fair
values, Accounting, Organizations and Society (2015), http://dx.doi.org/10.1016/j.aos.2015.09.001

18 B. Bratten et al. / Accounting, Organizations and Society xxx (2015) 1e18

Himmelberg, C. P., Hubbard, R. G., & Palia, D. (1999). Understanding the determi-
natns of managerial ownership and the link between ownership and perfor-
mance. Journal of Financial Economis, 53(3), 353e384.
Hodder, L., Mayew, W. J., McAnally, M. L., & Weaver, C. D. (2006). Employee stock
option fair-value estimates: do managerial discretion and incentives explain
accuracy? Contemporary Accounting Research, 23(4), 933e975.
Johnston, D. (2006). Managing stock option expense: the manipulation of option-
pricing model assumptions. Contemporary Accounting Research, 23(2), 395e425.
Kern, B. B., & Morris, M. H. (1994). Differences in the COMPUSTAT and expanded
value line databases and the potential impact on empirical research. The Ac-
counting Review, 69(1), 274e284.
Kinney, W. R., Palmrose, Z., & Scholz, S. (2004). Auditor independence, non-audit
services, and restatements: was the U.S. government right? Journal of Ac-
counting Research, 42(3), 561e588.
Kinney, M. R., & Swanson, E. P. (1993). The accuracy and adequacy of tax data in
COMPUSTAT. Journal of American Taxation Association, 15(1), 121e135.
Kross, W., Ro, B. T., & Schroeder, D. (1990). Earnings expectations: the analysts'
information advantage. The Accounting Review, 65(2), 461e476.
Lang, M., & Lundholm, R. (1996). Corporate disclosure policy and analyst behavior.
The Accounting Review, 71, 467e492 (October).
Leon, L., Kalbers, L., Coster, N., & Abraham, D. (2012). A spreadsheet life cycle
analysis and the impact of Sarbanes-Oxley. Decision Support Systems, 54,
452e460.
Mills, L. F., Newberry, K. J., & Novack, G. F. (2003). How well do compustat NOL data
identify rms with U.S. tax return loss carryovers? Journal of the American
Taxation Association, 25(2), 1e17.
Palmrose, Z.-V. (1988). An analysis of auditor litigation and audit service quality. The
Accounting Review, 63(1), 55e73.
Panko, R. R., & Ordway, N. (2005). Sarbanes-Oxley: what about all the spreadsheets?
controlling for errors and fraud in nancial reporting. In EuSpRIG 2005 Pro-
ceedings, Greenwich, London, sponsored by the European spreadsheet risks
interest
group (EuSpRIG), July 6e8, 2005.
Peecher, M. E., Schwartz, R., & Solomon, I. (2007). It's all about audit quality: per-
32(4e5), 463e485.
Powell, S. G., Baker, K. R., & Lawson, B. (2008). A critical review of the literature on
spreadseheet errors. Decision Support Systems, 46, 128e138.
PricewaterhouseCoopers (PwC). (2004). The use of spreadsheets: Considerations for
section 404 of the SarbandeseOxley Act. PricewaterhouseCoopers.
PricewaterhouseCoopers (PwC). (2013). Guide to accounting for stock-based
compensation: A multidisciplinary approach. PricewaterhouseCoopers.
Public Company Accounting Oversight Board (PCAOB). (2006). Staff questions and
answers: Auditing the fair value of share options granted to employees. October 17.
Washington, DC: PCAOB.
Public Company Accounting Oversight Board (PCAOB). (2009). Auditing fair value
measurements and using the work of a specialist. In Standing advisory group
meeting. October 14e15. Washington, DC: PCAOB.
Public Company Accounting Oversight Board (PCAOB). (December 6, 2011). Staff
audit practice alert No. 9. Assessing and responding to risk in the current economic
environment. Washington, DC: PCAOB.
Rapoport, M. (2013). Regulator rebukes PricewaterhouseCoopers. The Wall Street
Journal. March 7, 2013.
Rhodes, A., & Russomanno, D. (2013). One person can make a difference: The positive
assocaiton between chief accounting ofcers (CAOs) and accounting quality.
Working paper. Pennsylvania State University.
San Miguel, J. G. (1977). The reliability of R&D data in COMPUSTAT and 10-K reports.
The Accounting Review, 52(3), 638e641.
Securities and Exchange Commission (SEC). (2003). Summary by the division of
corporation nance of signicant issues addressed in the review of the periodic
reports of the fortune 500 companies. Washington, DC: SEC.
Shi, L., & Zhang, H. (2012). On alternative measures of accruals. Accounting Horizons,
25(4), 811e836.
Teoh, S. H., & Wong, T. J. (1993). Perceived auditor quality and the earnings response
coefcient. The Accounting Review, 68(2), 346e366.
Thomas, L. R., & Swanson, E. P. (1986). Additional considerations when using the
FASB bank of changing price information. The Accounting Review, 61(2),
330e336.
 spectives on strategic-systems auditing. Accounting, Organizations and Society,

Please cite this article in press as: Bratten, B., et al., The accuracy of disclosures for complex estimates: Evidence from reported stock option fair
values, Accounting, Organizations and Society (2015), http://dx.doi.org/10.1016/j.aos.2015.09.001