ORI GINAL RESEARCH
Conservatism measures that control for the effectsof economic rents on stock returns
Judson A. Caskey • Kyle Peterson
� Springer Science+Business Media New York 2013
Abstract Recent studies show that regression-based estimates of accounting conserva-
tism reflect both differences in the asymmetric recognition of bad news and differences in
asset composition. In particular, a firm’s market value and returns reflect both assets-
in-place and expected future rents, while book values tend to reflect only assets-in-place.
We propose two tests that remove the effect of asset composition on cross-sectional
comparisons of accounting conservatism. First, a test based on a ratio of regression
coefficients allows for valid cross-sectional comparisons of conservatism relative to overall
news recognition. Second, in some cases, researchers can separately identify and make
cross-sectional comparisons of the fraction of good news recognized and the fraction of
bad news recognized. The estimates in this second scenario use a regression of earnings on
returns interacted with a book-to-market ratio. We validate our model by deriving and
testing several predictions based on it.
Keywords Accounting conservatism � Asymmetric timeliness � Book-to-market ratio �Returns-earnings relation
JEL Classification M41 � G10 � G30 � N20
1 Introduction
A large and growing body of accounting research studies the causes and consequences of
the tendency of accounting earnings to recognize unrealized losses earlier than unrealized
gains (Watts 2003a, b). The accounting treatment of unrealized gains and losses depends
J. A. CaskeyMcCombs School of Business, University of Texas, Austin, 1 University Station,B6400, Austin, TX 78712, USAe-mail: [email protected]
K. Peterson (&)Lundquist College of Business, University of Oregon, Eugene, OR 97403, USAe-mail: [email protected]
123
Rev Quant Finan AccDOI 10.1007/s11156-013-0360-1
on the nature of assets to which they pertain. Overall firm value consists of both assets-
in-place and future economic rents such as profits that derive from customer loyalty or
corporate culture (Roychowdhury and Watts 2007). Accounting standards, both across
industries and across countries, differ in the extent to which accounting book values reflect
future rents, if at all, whereas returns reflect investor expectations about both components
of firm value. As a result, cross-sectional differences in estimates of accounting conser-
vatism reflect both differences in the composition of firm value and, holding constant the
extent to which balance sheets include future rents, differences in firms’ policies for
recognizing gains and losses.
Many studies estimate accounting conservatism using a regression of earnings on
returns (Basu 1997) in which the incremental coefficient on negative returns serves as a
proxy for conservatism (hereafter ‘AT coefficient’). This measure reflects the extent to
which book values recognize gains and losses on both assets-in-place and future rents. If
the objective of a study is to compare the overall relationship between book and market
values (e.g. across accounting standards regimes), the effect of asset composition on this
measure poses no problems. However, in studies that examine firm-specific patterns in
recognizing gains and losses (e.g. for debt contracting purposes), it is important to consider
whether the accounting regime allows for the recognition of future rents. If not, as is the
case with US GAAP, the AT coefficient does not allow researchers to separately identify
the effects of recognition policies and the effects of asset composition. In particular, a firm
could have a high (low) AT coefficient simply because a small (large) portion of its market
value is due to expected future economic rents.
For example, consider a firm that has separable assets with a resale value of $100 and a
book value, equaling the book value of equity, of $90. Now suppose that this firm expe-
riences a negative shock in the demand for its products that reduces the resale value of its
net assets to $85 and therefore records a $5 charge to earnings for the net impairment. If
this firm does not earn abnormal profits, its market value will be roughly equal to the
market value of its separable assets and will therefore experience a decline of $15 = $100–
$85. Thus, 1/3 = $5/$15 of the decline in the market value of equity will be recorded as a
charge to earnings.
Now consider the same example, except that the firm is able to earn abnormal profits so
that its market value includes an extra $20 representing the present value of the expected
future abnormal profits. If the negative demand shock also adversely impacts this com-
ponent of firm value by, say, $5, then total decline in market value is then $20, equal to the
$15 reduction in assets in place plus the $5 reduction in expected future abnormal profits.
The earnings charge remains the same because book values do not include expected future
profits. Thus, earnings include 1/4 = $5/$20 of the reduction in market value. A typical
regression-based measure of conservatism would identify the former firm as having more
conservative earnings even though the two firms have identical accounting policies. This
represents a form of mismeasurement to the extent that the research objective is to compare
recognition policies as opposed to the overall extent to which book values track overall
news that impacts market values.
In this study, we propose two conservatism estimates that mitigate the effects of asset
composition, allowing for cross-sectional comparisons based solely on firms’ recognition
policies. The first measure utilizes a ratio of regression coefficients from an earnings-
returns regression (hereafter ‘AT ratio’). This measure is based on the fact that asset
composition affects the recognition of both good and bad news. In order to address con-
cerns that the AT ratio may have poor econometric properties, we report confidence
intervals based on both the delta method and Fieller’s Theorem (Fieller 1954), which is a
J. A. Caskey, K. Peterson
123
more robust means to determine confidence intervals for ratios of regression coefficients,
particularly in small samples. Our results suggest that, at least in large samples, the delta
method confidence intervals approximate the Fieller confidence intervals very well.
The second measure utilizes a regression of earnings on returns interacted with the
firm’s book-to-market ratio. The book-to-market ratio serves as a control for the fraction of
returns attributable to future rents rather than assets-in-place. The benefit of this measure is
that it allows researchers to determine whether one group of firms incorporates more bad
news in earnings than another, while the AT ratio can only compare the relative degree of
conservatism between firms (i.e. bad news recognition relative to good news recognition).1
A comparison of ratios is unable to quantify how much good or bad news about separable
assets is included in earnings (e.g. a firm’s earnings include X % of the value of negative
shocks to separable assets and Y % of the value of positive shocks).
For researchers that want to capture cross-sectional differences in the magnitude of bad
news recognition in earnings we suggest using the regression where returns are interacted
with the book-to-market ratio. This approach uses book value as a proxy for the market
value of assets-in-place, which potentially creates measurement error. In order to assess
whether the measurement error has a major impact on AT coefficient estimates, researchers
can utilize the fact that the bias also impacts an AT ratio computed from a regression of
earnings on returns interacted with the book-to-market ratio. If the AT ratio from a
regression with book-to-market interacted returns closely matches the AT ratio from a
regression without the interaction, then this provides evidence that bias has little impact on
the AT coefficients from the regression with book-to-market interacted returns, allowing
researchers to use these coefficients for cross-sectional comparisons. If the ratios are not
similar, researchers cannot obtain a direct measure of the fraction of bad news recognized,
but can obtain a relative comparison of bad news recognition using the AT ratio from the
regression without the interaction with book-to-market.
We analyze an econometric model of the relation between market and book values to
assess the validity of conservatism estimates.2 The firm’s book value incorporates the
recognized value of separable assets, while market value includes investors’ assessment of
the value of separable assets plus the value of future rents and growth opportunities.3
Within the context of our model, we show that regression tests of asymmetric timeliness
identify the presence of conservative accounting policies, but also reflect assets
composition.
Our econometric model implies that the effect of rents on returns dampens the asso-
ciation between earnings and returns to the extent that book values reflect only assets-
in-place, consistent with the intuition of Roychowdhury and Watts (2007). Thus, firms with
high book-to-market ratios, which presumably have a large portion of value and returns
related to assets-in-place, have higher AT coefficients. This is consistent with our tests
(significant at 1 %). In contrast, we find that high book-to-market firms have an AT ratio
that is statistically equivalent to the ratio of low book-to-market firms. In other words, the
difference in AT coefficients between high and low book-to-market firms appears to be
driven by differences in firms’ composition of assets-in-place versus future rents as
opposed to differences in accounting recognition policies.
1 See Guay and Verrecchia (2006) for a discussion of the importance of distinguishing between estimates ofa firm’s bad news recognition versus its recognition of bad news relative to good news.2 See the contemporaneous paper by Ball et al. (2009) for an alternative econometric model.3 This assumption relates to Roychowdhury and Watts’ (2007) development of Watts’ (2003a) frameworkin which accounting attempts to represent the value of a firm’s separable assets.
The effects of economic rents on stock returns
123
We derive and test several implications of our econometric model in order to assess its
empirical validity. We predict and find that interacting returns with the book-to-market
ratio increases AT coefficients, which is due to both a reduction in the dampening effect
that future rents have on AT coefficients and bias related to unrecognized gains on assets-
in-place. We predict a positive bias on the AT ratio, as well, but the test of that prediction
is statistically insignificant. Roychowdhury and Watts (2007) state that the inclusion of
goodwill violates the assumption that book values only reflect separable assets. We predict
and verify that the inclusion of goodwill does bias AT coefficients upward in the sense that
acquisitions that create recognized goodwill can increase the AT coefficient vis-a-vis a firm
with identical accounting policies but that did not have acquisitions. We also predict a
similar effect on the AT ratio, but fail to find statistically significant evidence in support of
that prediction.
Many prior studies utilize ratio-based measures of conservatism, but without analytical
justification for the practice. In our examination of studies that cite Basu (1997), we found
eight that measure conservatism or describe results using a ratio of regression coefficients.4
Seven of these studies define the ratio using the coefficient on positive returns in the
denominator, which raises econometric concerns because the coefficient on positive returns
often lacks statistical significance yielding an unbounded confidence interval for the ratio.5
None of these seven studies assess the statistical significance of the ratio. Callen et al.
(2010) develop a ratio measure of conservatism in an alternative framework based on a
decomposition of returns into discount rate and cash flow news. Our approach of inter-
acting returns with book-to-market also differs from Khan and Watts (2009), who develop
a measure based on linear combinations of firm size, market-to-book and leverage that are
associated with conservatism.
Of the eight studies that use a ratio measure of conservatism, only Pope and Walker
(1999) discuss the use of a ratio to correct for biases in cross-sectional comparisons of AT
coefficients. They use a ratio to correct for cross-country differences in expected returns.
Our model is distinct from theirs in that we allow for the distinction between separable
assets and future rents and we allow for a more general relation between returns and
earnings. We demonstrate that comparisons based on the AT ratio control for cross-
sectional differences in the amount of firms’ market values attributable to rents, thus
showing that the ratio controls for the biases that have yielded a negative relation between
AT coefficients and market-to-book ratios.
Roychowdhury and Watts (2007) provide an alternative means to correct for the effect
of future rents on the estimation of AT coefficients. They propose extending the returns
measurement horizon to, for example, 3 years. Beatty (2007) notes that this approach
reduces sample sizes due to data availability requirements and that the requirement of
negative returns for a multi-year period reduces the number of bad-news observations.
Moreover, Givoly et al. (2007) discuss how the aggregation of both positive and negative
news events reduces the power of asymmetric timeliness tests. The use of multi-year
returns exacerbates this effect. Lastly, while extending the measurement horizon may
reduce bias, Roychowdhury and Watts (2007) do not discuss whether or not this approach
allows for valid cross-sectional comparisons.
4 We searched articles in Contemporary Accounting Research, Journal of Accounting and Economics,Journal of Accounting Research, Review of Accounting Studies, and The Accounting Review through March2012.5 We show that controlling for lagged earnings yields statistically significant coefficients on positiveearnings in earnings/returns regressions, which is important for using a ratio to infer conservatism.
J. A. Caskey, K. Peterson
123
The paper proceeds as follows. Section 2 describes measures of conservatism that
remove the effects of differences in asset composition (future rents vs. assets-in-place),
develops our econometric model and predictions, and describes how our model relates to
prior literature. Section 3 describes our data. Section 4 presents our primary tests and Sect.
5 discusses the empirical properties of the AT measures. Section 6 concludes.
2 Measures of conservatism
2.1 Econometric model
This section illustrates how to control for the effect of asset composition on estimates of
accounting conservatism. The simple model we present here establishes the intuition
necessary to progress to our predictions and empirical tests. An expanded econometric
model is available from the authors upon request. Researchers often estimate accounting
conservatism using a regression of earnings scaled by market value of equity (xit) on
returns (rit) as in Basu (1997) for firm i and year t:
xit ¼ b0 þ b11rit\0 þ b2rit þ b31rit\0rit þ eit: ð1ÞThe coefficient b3 (hereafter ‘AT coefficient’) is used as a measure of the asymmetric
timeliness of earnings with respect to recognizing bad news versus good news. In this
regression total returns are used as a proxy for news, but include the change in both the
value of separable assets and future rents as described in (2). However, earnings generally
do not include the value of future rents until they are realized. Denoting the market value of
firm i at time t by vit, we have:
vit ¼ Assestit þ Rentsit rit ¼Dvit
vi;t�1
¼ DAssetsit
vi;t�1
þ DRentsit
vi;t�1
ð2Þ
If the firm’s market value maintains the same proportion of separable assets and future
rents
(k ¼ Assetsit�1=vit�1), then the change in separable assets can be written:
DAssetsit
vi;t�1
¼ krit ð3Þ
If earnings incorporate a portion of the change in separable assets but not future rents and
the proportion recognized is asymmetric for positive and negative changes (h and �h), then
scaled earnings can be written as6:
xit ¼ 1rit\0hkrit þ 1rit � 0�hkrit: ð4Þ
If earnings as defined in (4) are estimated using the typical Basu (1997) regression as in (1)
the coefficients b2 and b3 asymptotically approach the following:
plim b2 ¼ �hk plim b3 ¼ ðh� �hÞk: ð5Þ
6 Although not modeled here, current earnings could also recognize a portion of prior unrecognized changesin separable assets and realized rents. The exclusion of these elements from the model does not affect ourinferences because we assume the current recognition of prior returns is uncorrelated with current returns.Our more elaborate econometric model available from the authors does incorporate these additionalelements.
The effects of economic rents on stock returns
123
The coefficients in (5) identify the existence of conservatism (h and �h), and provide for
valid cross-sectional comparisons to the extent the researcher wishes to examine the
overall relation between returns and earnings. However, cross-sectional comparisons
between the coefficients b2 and b3 do not allow researchers to compare the asymmetric
timeliness of earnings with respect to separable assets, as described in Roychowdhury and
Watts (2007), because these coefficients reflect not only accounting recognition, h and �h,
but also k, the proportion of separable assets to firm value. For example, firms with a high
(low) proportion of firm value from future rents will have lower (higher) AT coefficients
because they have lower (higher) k. However, this difference in AT coefficients is not due
to more conservative accounting policies but instead results from differences in asset
composition.
As mentioned in the introduction, Roychowdhury and Watts (2007) suggest extending
the windows for returns to correct for the effect of future rents on AT coefficients. We
provide two alternative options for researchers interested in removing the effects of asset
composition from conservatism estimates. The first option involves taking a ratio of
coefficients. Since both b2 and b3 include the effect of k, taking a ratio of the slope
coefficients can eliminate the effect of k. We recommend including the sum of the coef-
ficients on positive and negative returns in the denominator to mitigate small denominator
effects (Ryan and Zarowin 2003).7 This gives:
plimb3
b2 þ b3
¼ ðh��hÞk
hk¼ h� �h
h: ð6Þ
This approach achieves consistent estimates of asymmetric timeliness but only allows for
comparisons of bad news relative to good news recognition. Researchers interested in
measuring the absolute differences in conservatism could potentially use a second option
for correction. This involves estimating a slightly altered regression specification where the
book-to-market ratio, as a proxy for k, is multiplied by returns, which attempts to remove
the effect of k on the coefficients. This altered specification is:
xit ¼ c0 þ c11rit\0 þ c2
bi;t�1
vi;t�1
rit þ c31rit\0
bi;t�1
vi;t�1
rit þ eit: ð7Þ
However, because book value is a noisy estimate of the value of separable assets, the
estimates of c now have an errors-in-variables problem. Therefore, the estimates c3 and
c2 ? c3 now have an additive bias that depends on the accuracy of bit�1
vit�1as a proxy for k. In
particular, the book-to-market ratio is:
bi;t�1
vi;t�1
¼ kbi;t�1 �Market value of assetsi;t�1
vi;t�1
: ð8Þ
From (8), we see that the bias in the coefficient estimates in (7) depend on the extent to
which the book value of separable assets differs from the market value of separable assets.
7 This does not eliminate the need to have a fairly precise measure of the coefficient on positive returns. TheAppendix discusses the Fieller method procedure for computing confidence intervals for ratios of regressioncoefficients. The regression coefficients must be statistically significant in order to have bounded confidenceintervals. For example, a 95 % confidence interval requires that the regression coefficients be significant atthe 5 % level. This implies that the ratios will be statistically insignificant in settings where the asymmetrictimeliness coefficient is statistically insignificant, such as Eng and Lin (2012) examination of Chinese cross-listing firms.
J. A. Caskey, K. Peterson
123
For example, Coca-Cola’s book value includes very little of the value of its trademark,
driving a wedge between the book value and market value of its separable assets.
Researchers should be aware of this when making cross-sectional comparisons between
sets of firms with substantial differences in the extent to which their book values over- or
under-value separable asset. Because the effects of the estimation bias in (7) are difficult to
gauge ex ante, in the next section we describe a simple two-step process to determine
whether researchers can rely on the coefficient estimates in (7).
2.2 Empirical predictions
In this section, we develop several predictions that we test in Sect. 4. The coefficients in (5)
imply a positive association between the AT coefficient and the proportion k of firm value
associated with assets-in-place. Our first prediction is:
P1 The estimated AT coefficient b3 is higher for firms with a higher proportion of value
based on separable assets (high k)
Assuming a positive relation between book-to-market ratios and the proportion of value
associated with assets-in-place, P1 implies a positive relation between book-to-market
ratios and the AT coefficient as in Pae et al. (2005), Givoly et al. (2007) and
Roychowdhury and Watts (2007). To the extent that the book-to-market ratio proxies for
the portion of value associated with assets-in-place, we can also utilize regression (7) to
estimate conservatism. Given that k dampens the coefficient estimates from a typical Basu
regression, this yields the following prediction:
P2 The estimated AT coefficient c3 from (7) is higher than the estimated AT coefficient
b3 from (1). The ratio c3=ðc2 þ c3Þ from regression (7) equals the ratio b3=ðb2 þ b3Þfrom (6) if the error in estimating k is small.
Given our focus on making cross-sectional comparisons, we note that regression (7) can
serve as a basis for estimating differences in conservatism between groups of firms. While
using a ratio of coefficients from (1) removes the effects of asset composition from cross-
sectional comparisons of the relative degree of conservatism, researchers who are interested
in obtaining an estimate of the amount of bad news recognition could use the coefficients from
(7), subject to the caveat that the coefficients provide a biased measure of conservatism.
Because the bias is additive, it affects both the coefficient and the ratio from (7); therefore,
researchers can check the magnitude of the bias by comparing the ratios from (1) and (7). If
the coefficient bias in (7) is small, the ratios from (1) and (7) will be roughly equal and the
coefficient estimates from (7) can be used as an estimate of the degree of conservatism.
Our remaining predictions address the effect of goodwill recognition on conservatism
estimates. The derivations in Sect. 2.1 assume that book value reflects only separable
assets. Goodwill is a notable exception to this assumption and may result in apples-
to-oranges comparisons between companies that grew via acquisitions and those that did
not. The earnings of companies that carry goodwill on their balance sheets may include
negative shocks to the value of future rents. This effect will increase the coefficient esti-
mates for b3 because goodwill writedowns induce a stronger relationship between negative
returns and negative earnings. These observations lead to the following predictions:
P3 The AT coefficient is smaller in an estimation of (1) where earnings exclude the
effects of goodwill as compared to an estimation where earnings include the effects
of goodwill.
The effects of economic rents on stock returns
123
P4 If the AT ratio is less than (greater than) one based on estimates of (1) using earnings
that include the effects of goodwill, it is smaller (higher) when estimated using
earnings that exclude the effects of goodwill.8
2.3 Prior studies
This section provides a discussion of conservatism research since Basu (1997) in order to
frame our study’s contribution. Table 1 summarizes the conservatism measures used in
studies that cite Basu (1997) published in the following five accounting journals through
March 2012: Contemporary Accounting Research, Journal of Accounting and Economics,
Journal of Accounting Research, Review of Accounting Studies, and The AccountingReview. Table 1, Panel A shows that 76 of the 156 studies that cite Basu (1997) include
empirical tests that measure conservatism or asymmetric timeliness. Regarding these 76
studies, we highlight three observations about the literature.
First, while many papers have used some version of the Basu (1997) coefficient, other
measures have been developed, and there appears to be no real consensus on the use of
asymmetric timeliness measures. Table 1, Panel B describes the empirical measure(s) used
in these 76 empirical studies. Twelve different measures have emerged with some degree
of regularity in the literature. The first four measures listed in Panel B are all obtained from
estimates of a regression similar to (1). Others, like the incremental persistence of negative
earnings, as in Basu (1997), Ball and Shivakumar (2005), and Brauer and Westermann
(2012) are driven by data requirements such as the lack of available returns data.9 A few
researchers motivate their choice of measures by referring to analyses in Dietrich et al.
(2007) and Givoly et al. (2007) that suggest there may be problems with using a returns/
earnings framework (see Krishnan and Visvanathan 2008; Wang 2006; Ahmed et al. 2002).
However, many researchers do not discuss their choice of conservatism measures. This
lack of consensus on the use of asymmetric timeliness measures hinders the ability of
researchers to compare findings across different studies. However, this is perhaps mitigated
by the fact that most researchers (41 of the 76 studies) use multiple measures or some
combination of measures. Furthermore, most researchers (63 of the 76 studies) use some
measure estimated from a regression similar to (1), and the majority use the estimated
coefficient on negative returns, which we demonstrate leads to biased inferences in cross-
sectional comparisons.
Second, many of the papers test for cross-sectional differences in asymmetric timeli-
ness, which we highlight as problematic in tests based on the AT coefficient when
researchers do not wish to compare the accounting recognition of future rents. Of the 76
studies using some measure of conservatism, 64 of them conduct a cross-sectional
examination. For example, Givoly et al. (2007) find a negative relation between the AT
coefficient and the market-to-book ratio, which is puzzling because market-to-book
is another measure of accounting conservatism.10 Roychowdhury and Watts (2007)
hypothesize that this negative relation is due to a high proportion of returns being attrib-
utable to revisions of expected future rents, which we explicitly derive in our model.
8 Although this prediction is not immediately obvious, the derivation is available from the authors uponrequest.9 Another approach to estimating asymmetric timeliness in the absence of market return data is to measurethe frequency of large negative net income (e.g. Barth et al. 2008; Elbannan 2011).10 The market-to-book ratio reflects both asymmetric timeliness and ex ante conservatism.
J. A. Caskey, K. Peterson
123
Third, of the eight studies that measure conservatism using a ratio of coefficients, only
Pope and Walker (1999) and Guay and Verrecchia (2006) provide an explanation for the
use of this measure. Pope and Walker (1999) derive a ratio-based test of conservatism in a
permanent income model to control for cross-country differences in discount rates. Our
Table 1 Basu (1997) citations and use of conservatism measures
Paper type Citations
Panel A: Basu (1997) citations by type
Citation only 80
Empirical tests 76
Total citations 156
Papers using multiple measures 41
Papers not using Basu (1997) measure 7
Conservatism measure Citations
Panel B: Basu (1997) citations by conservatism measure used
Basu coefficient 39
Basu good news coefficient 7
Basu R2 9
Basu coefficient ratio 8
Sensitivity of returns (Easton and Pae (2004)) 4
Hidden reserves (Penman and Zhang (2002)) 4
Cumulative discretionary accruals (Givoly and Hayn (2000)) 11
Skewness of earnings (Givoly and Hayn (2000)) 6
Beaver and Ryan (2000) measure 4
Persistence of earnings (Basu (1997) and Ball and Shivakumar (2005)) 19
Aggregated Basu (1997) and Roychowdhury and Watts (2007) 6
Market-to-book or book-to-market 6
Other 7
Total 130
This table reports the use of conservatism measures for all Basu (1997) citations in the following 5Accounting Journals: Contemporary Accounting Research, Journal of Accounting and Economics, Journalof Accounting Research, Review of Accounting Studies, and The Accounting Review. Panel A presentsinformation about whether the paper conducted an empirical test using a measure of conservatism, thenumber of studies using multiple measures, and those not using a measure from estimation of a Basu (1997)style regression. Panel B displays the number of papers in which a particular conservatism measure is used.The Basu coefficient (b3), Basu good news coefficient (b2), Basu R2 and Basu coefficient ratio ([b3 1 b2]/b3) are all estimated from the following regression model: EPSit/Pi,t-1 = b0 ? b1 1(rit \ 0) ? b2 rit ? b3
1(rit \ 0) rit ? eit for firm i and year t where the indicator variable 1(rit \ 0) equals one when returns arenegative and zero otherwise. The sensitivity of returns is either conservatism due to the net present value ofinvestments and/or conservatism due to accounting rules, which are estimates obtained from the regressionspecified in Easton and Pae (2004). Hidden reserves are the sum of the LIFO reserve, hypothetical capi-talized R&D expenditures and hypothetical capitalized advertising expenditures, scaled by net operatingassets as defined in Penman and Zhang (2002). Cumulative discretionary accruals and skewness of earningsare measures initially used by Givoly and Hayn (2000). The Beaver and Ryan (2000) measure is thefirm specific intercept of a pooled OLS regression of BTM ratio on current and lagged returns. ThePersistence coefficient is the estimate b3 from the following regression: Earnt = b0 ?b1 1(Earnt-1 \ 0) ? b2 Earnt-1 ? b3 1(Earnt-1 \ 0) Earnt-1 ? eit consistent with Basu (1997) and Balland Shivakumar (2005). The Aggregated Basu is the procedure conducted by Basu (1997) and Roy-chowdhury and Watts (2007) that aggregates the variables in the Basu regression over two or more years
The effects of economic rents on stock returns
123
model differs from theirs in that we distinguish between separable assets and future rents
and have a more general relation between returns and earnings that does not rely on the
permanent income feature used by Pope and Walker (1999) and critiqued by Basu (1999).
In addition, six of the eight studies discuss a ratio where the aggregate coefficients on
negative returns (b2 ? b3) is divided by the coefficient on positive returns (b2). The use of
this ratio provides a salient way to interpret the asymmetric timeliness estimated using the
coefficients (e.g. ‘‘the incorporation of negative news is six times that of positive news’’).
As we discuss in the next section, this particular ratio may cause econometric problems
because the estimated coefficient b2 on positive returns is often near zero.11
In addition to the AT ratio measure of conservatism, the previous section introduced a
coefficient measure from a Basu (1997)-type regression where we interact returns with the
book-to-market ratio. This approach is different from other studies that use book-to-market
as a measure of conservatism. Roychowdhury and Watts (2007; Figure 1) illustrate that the
book-to-market ratio commingles asymmetric loss recognition with expected future eco-
nomic rents included in market value but excluded from accounting measures of separable
assets. In contrast, the regression-based measure in (7), which utilizes returns interacted
with book-to-market, provides an estimate of asymmetric loss recognition.
Khan and Watts’ (2009) firm-year estimate of conservatism more closely resembles the
regression-based measure in (7), but differs both computationally and in terms of theo-
retical motivation. Regarding the computational differences, the Khan and Watts (2009)
C_Score measure provides an estimate of conservatism as a linear function of the firm
characteristics of Size (log market value of equity), market-to-book, and leverage. In other
words, whereas we interact returns with book-to-market, they estimate a regression with
un-interacted returns and returns interacted with size, market-to-book, and leverage.
Regarding theoretical motivation, we utilize the book-to-market ratio to proxy for the
fraction of returns associated with separable assets—the model’s k parameter. Market-
to-book, as used by Khan and Watts (2009), cannot serve this role. Also, our model also
has no particular role for size and leverage. Khan and Watts (2009) utilize these variables
in order to assess the cross-sectional drivers of conservatism rather than to correct biases in
estimates from a standard Basu (1997) regression. Thus, both the computation and theo-
retical support differ between our proposed AT measures and those in Khan and Watts
(2009).
Our study provides guidance on the use of conservatism measures in cross-sectional
studies. In the next section, we test our predictions that cross-sectional comparisons based
on regression coefficients, as in (1), are impacted by the composition of firm value to the
extent that book values exclude future rents. However, tests using a ratio of regression
coefficients, as in (6), or coefficients from a regression with returns interacted with book-
to-market, as in (7), allow for cross-sectional comparisons that reflect accounting policy
choices related to the recognition of separable assets. We test the other predictions in order
to assess the empirical validity of our model. Furthermore, we verify the AT ratio’s
econometric properties by comparing asymptotic, delta method confidence intervals to
Fieller method confidence intervals, which addresses the econometric concerns raised by
Basu (1999) in his discussion of Pope and Walker (1999).12 We describe the Fieller
method confidence intervals in the Appendix.
11 Our analysis in the next section shows that the typical Basu (1997) framework produces estimates of b2
near zero. However, including additional controls to better measure unexpected earnings and unexpectedreturns often results in a positive coefficient on positive returns.12 Pope and Walker (1999) do not assess the statistical significance of their ratio measures.
J. A. Caskey, K. Peterson
123
3 Data and sample selection
We conduct our empirical analysis on the sample of firm-years reported in Table 2, Panel
A. We begin with 254,127 firm-years in the Compustat Fundamental Annual table from
1963 to 2007. We restrict the sample to firm-years with matches to CRSP identifiers in the
CRSP/Compustat merged database (205,448 firm-years). Missing variables from Compu-stat or CRSP reduces the sample to 172,260 firm-year observations. We require that firms
have share prices greater than $2 in order to avoid small-denominator problems in the
price-deflated earnings measures and to reduce the impact of the high return variance for
firms with low share prices (Patatoukas and Thomas 2009). This requirement reduces the
sample by 17,058 firm-years. We further require that firms have positive book value of
equity, net of goodwill, for use in tests where we interact returns with the book-to-market
ratio. The final sample is 155,201 firm-years with 17,359 firms.
Panel B of Table 2 reports descriptive statistics for the sample firms. We winsorize all
variables at the 1st and 99th percentile within Compustat fiscal years. We calculate Assets
(Compustat Annual Xpressfeed item AT), Market value (PRCC_F 9 CSHO), Book value
(CEQ), and book-to-market (Book value/Market value) at the beginning of the period.
Earnings refers to earnings per share from continuing operations (EPSPX) scaled by the
beginning-of-fiscal-year price (PRCC_F), which we adjust for stock splits by multiplying
by the ratio of the current-to-prior year price adjustment factors (AJEX).13 We obtain
annual buy-and-hold returns from the CRSP monthly stock file beginning at the start of the
fourth month of the fiscal year. The descriptive statistics show the mean Assets, Market
and Book values of equity to be $2.1 billion, $880.6 million, and $514.8 million for our full
sample of firms, respectively. The mean book-to-market for our sample is 0.78.
4 Empirical results
4.1 Effect of including prior information
Before proceeding with tests of the predictions from Sect. 2, we first investigate how the
inclusion of pre-returns information impacts estimates of asymmetric timeliness. In par-
ticular, firms with high price/earnings ratios tend to have lower future returns (Basu 1977)
while the analysis in Sect. 2 assumes a risk-neutral setting with no discounting where
expected returns are zero. Furthermore, Patatoukas and Thomas (2009) show that estimates
of asymmetric timeliness in regressions similar to Basu (1997) are significantly impacted
by past earnings and size, so that the inclusion of lagged earnings acts as somewhat of a
control for this bias.
Table 3 presents estimates of asymmetric timeliness using both the specification (1) as
in Basu (1997) and a specification that includes proxies for expected earnings and expected
returns. We condition on past earnings as a component of both expected earnings and
expected returns, where the latter derives from the aforementioned relation between returns
and earnings/price ratios (Basu 1977). We also condition on factors noted in Fama and
French (1992). If returns are independent of pre-returns information, a simplifying
13 We use earnings per share from continuing operations, but note there is no consistency in the literature.Of the 36 studies that use returns in their measurement of conservatism, 18 use earnings per share fromcontinuing operations, 10 use earnings per share, 4 use both, and 4 do not provide enough detail to determinewhich measure they used.
The effects of economic rents on stock returns
123
Ta
ble
2S
ample
sele
ctio
nan
dsu
mm
ary
stat
isti
cs
Panel
A:
sam
ple
sele
ctio
n
All
U.S
.fi
rm-y
ears
on
Co
mp
ust
at
25
4,1
27
Les
sfi
rms
wit
hn
om
atch
on
CR
SP
(48
,679
)
Les
sm
issi
ng
Co
mp
ust
at/
CR
SP
dat
a(3
3,1
88
)
Les
sfi
rms
wit
hsh
are
pri
ceB
$2
(17
,058
)
Les
sfi
rms
wit
hn
egat
ive
bo
ok
equ
ity
,n
eto
fg
oo
dw
ill
(1)
Fir
m-y
ear
ob
serv
atio
ns
15
5,2
01
Un
iqu
efi
rms
17
,35
9
Var
iab
leN
Mea
nS
td.
dev
.1
0th
pct
.2
5th
pct
.M
edia
n7
5th
pct
.9
0th
pct
.
Pa
nel
B:
sum
ma
ryst
atis
tics
Beg
tota
las
sets
15
5,2
01
2,1
00
.99
,18
0.1
14
.34
0.0
15
3.2
77
6.4
3,3
72
.9
Beg
mar
ket
val
ue
of
equ
ity
15
5,2
01
88
0.6
3,4
75
.91
0.5
27
.49
7.6
42
3.0
1,5
79
.3
Beg
bo
ok
val
ue
of
equ
ity
15
5,2
01
51
4.8
1,8
68
.67
.11
9.0
64
.12
55
.51
,00
5.3
Beg
bo
ok
val
ue
of
equ
ity
less
go
odw
ill
15
5,2
01
45
5.2
1,5
66
.86
.81
8.0
59
.72
31
.09
08
.0
Beg
bo
ok
-to
-mar
ket
(BT
M)
15
5,2
01
0.7
74
0.6
19
0.1
97
0.3
57
0.6
19
1.0
05
1.5
19
Beg
bo
ok
-to
-mar
ket
(BT
M)
less
go
odw
ill
15
5,2
01
0.7
44
0.6
20
0.1
69
0.3
22
0.5
87
0.9
77
1.4
97
Ret
urn
s1
55
,20
10
.14
20
.582
-0
.442
-0
.187
0.0
70
0.3
56
0.7
41
Ear
nin
gs
15
5,2
01
0.0
40
0.1
68
-0
.100
0.0
14
0.0
62
0.1
07
0.1
72
Ear
nin
gs
less
go
od
wil
l1
55
,20
10
.04
10
.165
-0
.098
0.0
14
0.0
62
0.1
07
0.1
72
Th
ista
ble
pre
sen
tsth
eat
trit
ion
of
the
sam
ple
and
sum
mar
yst
atis
tics
.P
anel
Are
po
rts
ho
wth
esa
mp
lew
aso
bta
ined
afte
rd
ata
req
uir
emen
ts.
Th
eo
bse
rvat
ion
sco
ver
the
yea
rs1
96
3–
200
7.P
anel
Bre
po
rts
sum
mar
yst
atis
tics
for
the
sam
ple
of
firm
s.A
sset
s(C
omp
ust
at
An
nu
alX
pre
ssfe
edit
emA
T),
Bo
ok
val
ue
of
equ
ity
(Co
mp
ust
at
CE
Q),
mar
ket
val
ue
of
equ
ity
(CR
SP
PR
C9
SH
RO
UT
),an
db
oo
k-t
o-m
ark
et(b
oo
kv
alue
of
equ
ity
/m
ark
etv
alue
of
equ
ity
)ar
eca
lcu
late
das
of
the
beg
inn
ing
of
the
fisc
aly
ear.
Ret
urn
sar
ean
nu
alb
uy
-and
-ho
ldre
turn
so
bta
ined
fro
mth
eC
RS
Pm
on
thly
sto
ckfi
lean
db
egin
atth
est
art
of
the
fou
rth
mon
tho
fth
efi
scal
yea
r.E
arn
ing
sis
earn
ings
per
shar
efr
om
con
tin
uin
goper
atio
ns
(Co
mp
ust
at
EP
SPX
)sc
aled
by
the
stock
pri
ce(P
RC
C_
F)
atth
eb
egin
nin
go
fth
efi
scal
yea
r,w
her
ew
ead
just
the
pri
cefo
rst
ock
spli
tsb
ym
ult
iply
ing
the
rati
oo
fth
ep
rice
adju
stm
ent
fact
or
(AJE
X)
atth
eea
rnin
gs
ann
ou
nce
men
td
ate
toth
efa
cto
rat
the
beg
inn
ing
of
the
fisc
aly
ear.
Th
efo
reca
sto
fcu
rren
tea
rnin
gs
isth
efi
rst
mea
nI/
B/E
/Sea
rnin
gs
per
shar
efo
reca
stfo
llo
win
gth
ean
no
un
cem
ent
of
the
pri
or
yea
r’s
earn
ings.
We
adju
stth
efo
reca
stfo
rst
ock
spli
tsu
sin
gth
era
tio
of
the
CR
SP
shar
ead
just
men
tfa
cto
r(C
FA
CS
HR
)at
the
earn
ings
ann
ou
nce
men
td
ate
toth
efa
cto
rat
the
fore
cast
dat
e.W
esc
ale
the
fore
cast
by
the
spli
tad
just
edb
egin
nin
g-o
f-fi
scal
-yea
rsh
are
pri
ce.
All
var
iab
les
are
win
sori
zed
at1
/99
%w
ith
inC
omp
ust
at
fisc
aly
ears
J. A. Caskey, K. Peterson
123
assumption we made when deriving our predictions, then the addition of these regressors
will not affect the coefficients on returns but will impact the regression’s explanatory
power. If including additional regressors such as lagged earnings impacts the coefficients
on returns, then these regressors are also acting as controls for expected returns.
The results in Table 3 show that estimation of specification (1) [column (1)] generates a
positive and significant coefficient on negative returns consistent with asymmetric time-
liness. The coefficient on positive returns is insignificant which is consistent with prior
Table 3 Asymmetric timeliness tests with prior information
(1) (2) (3) (4)
Negative return indicator -0.003(-0.548)
-0.003(-0.638)
-0.002(-0.360)
-0.001(-0.323)
Returns 0.003(0.281)
0.020*(1.884)
0.021**(2.220)
0.019**(2.031)
Negative indicator 9 returns 0.307***(12.661)
0.204***(10.120)
0.182***(9.324)
0.200***(10.155)
Lag EPS 0.482***(16.907)
0.730***(16.061)
0.879***(31.833)
Negative earnings indicator -0.037***(-6.732)
-0.029***(-5.940)
Neg. earnings indicator 9 earnings -0.387***(-6.887)
-0.559***(-14.117)
Beginning BTM -0.030***(-4.192)
Beta 0.008**(2.574)
MV equity -0.002*(-1.944)
Intercept 0.079***(10.771)
0.046***(8.715)
0.028***(6.981)
0.044***(4.993)
N 155,201 155,201 155,201 155,201
Firm clusters 17,359 17,359 17,359 17,359
Year clusters 45 45 45 45
v2 223.5*** 538.6*** 1,709.1*** 4,301.8***
Adjusted R2 0.147 0.365 0.386 0.396
AT ratio 0.989 0.911 0.897 0.912
95 % confidence intervals
Delta method [0.915, 1.064] [0.817, 1.005] [0.804, 0.989] [0.828, 0.997]
Feiller method [0.915, 1.065] [0.814, 1.004] [0.800, 0.988] [0.825, 0.997]
This table reports coefficient estimates from Basu (1997)-type regressions of earnings on returns with theinclusion of t - 1 information to improve the specification. Table 2 describes our sample of firms andvariable definitions. Earnings are earnings per share from continuing operations (Compustat EPSPX) scaledby the beginning-of-fiscal-year price. Returns are annual buy-and-hold returns. Forecast is the price-scaledbeginning-of-year mean analyst forecast of earnings. Lag EPS is the 1-period lagged Earnings variable(dependent variable). Beginning BTM is the beginning-of-fiscal-year book value of equity scaled by marketvalue of equity. Beta is the market model beta estimated over the prior 5 months. MV Equity is the marketvalue of equity at the beginning of the fiscal year. All variables except indicator variables are winsorized atthe 1st and 99th percentile within Compustat fiscal years. Two-tailed t-statistics are shown below coefficientestimates, with standard errors clustered by firm and fiscal year. *, **, and *** indicate significance at 10, 5,and 1 %
The effects of economic rents on stock returns
123
studies. Column (2) adds a control for expected earnings using lagged EPS while column
(3) splits the lagged EPS into positive and negative EPS to control for the effects docu-
mented in Patatoukas and Thomas (2009). In both of these specifications, the coefficient on
positive returns is now positive and significant (0.020 and 0.021). Although the coefficient
on negative returns remains significantly positive, it is diminished (0.307 vs. 0.204 and
0.182) relative to the original specification. The altered coefficients on returns imply that
prior earnings impact both expected returns and expected earnings.
Finally, we include book-to-market, market model beta, and beginning market value of
equity in column (4) as additional controls for expected returns. We note that these factors
have little effect on the AT coefficients relative to specifications including controls for
expected earnings. Because the control for lagged scaled EPS [column (3)] improves the
regression specification for asymmetric timeliness significantly and produces a positive
coefficient on positive returns, we include this additional control in our subsequent tests.
4.2 Effect of rents on returns
Table 4 presents tests of prediction P1. The regressions examine cross-sectional differ-
ences in conservatism by separating firms along two dimensions: high and low book-
to-market ratios and REIT versus non-REIT. The first cross-sectional split categorizes
firm-years into high and low beginning-of-period book-to-market ratios. In specification 1
(2) of Table 4, firm-years above the median (66th percentile) are identified as high book-
to-market, while those below the median (33rd percentile) are identified as low book-
to-market. Firms with high book-to-market ratios appear to have more conservative
earnings as indicated by coefficients of 0.276 and 0.365 (significant at 1 %) on the
interaction between the high book-to-market indicator and the AT coefficient. This result is
consistent with prior studies and our prediction P1. On the other hand, the AT ratio
difference is insignificant when we partition by high book-to-market, suggesting that firms
with high book-to-market ratios have similar asymmetric timeliness to low book-to-market
firms. These results are consistent with Roychowdhury and Watts’ (2007) conjecture that
the higher AT coefficient for high book-to-market firms is, at least in part, due to the high
proportion of their value associated with separable assets that are more likely to be
reflected in earnings. The AT ratio removes the effects of cross-sectional differences in the
portion of returns associated with separable assets and shows little difference between the
samples.
In addition to reporting the AT ratios, we compare confidence intervals on the ratios
using both the Delta method and the Fieller method as a gauge of how well the Delta
method approximates the confidence interval for the ratio. In all cases, the two confidence
intervals are nearly identical. This suggests that the Delta method provides a reasonable
approximation to the standard errors for the ratio—a necessary precursor to using the ratio
in cross-sectional comparisons because a Fieller-type test is not available for comparing
two coefficient ratios.
The second regression specification provides additional support for the AT ratio by
examining cross-sectional differences in asymmetric timeliness for REITs (real estate
investment trusts) relative to other firms. We classify a firm as a REIT if the SIC is equal to
6798, but also include a broader definition of real estate firms (including firms with SIC
between 6500 and 6599 or equal to 6798 or 6799). Like high book-to-market firms, we
conjecture that REITs have returns more associated with changes in separable assets than
other firms, and test whether the AT ratio controls for this higher expected k. We do not
have any expectations that accounting rules related to REITs cause more asymmetric
J. A. Caskey, K. Peterson
123
Table 4 Results of cross-sectional estimates of asymmetric timeliness using market-to-book and REITsample partitions
High/low
BTM
(1)
Top/bottom
third BTM
(2)
REIT/
non-REIT
(3)
Alt REIT
(all real estate)
(4)
Negative return indicator 0.003
(0.954)
0.002
(0.826)
-0.003
(-0.718)
-0.003
(-0.776)
Returns 0.006
(1.035)
0.004
(0.718)
0.020**
(2.133)
0.020**
(2.088)
Negative returns indicator 9 returns 0.110***
(9.987)
0.093***
(9.523)
0.175***
(8.998)
0.173***
(9.164)
High BTM/REIT indicator 0.005*
(1.674)
0.007
(1.520)
-0.019***
(-2.921)
-0.023***
(-3.830)
High BTM/REIT 9 neg. indicator 0.013***
(2.616)
0.013*
(1.857)
0.042***
(3.864)
0.029***
(2.608)
High BTM/REIT 9 returns 0.030***
(3.435)
0.039***
(3.172)
0.008
(0.393)
0.031**
(2.017)
High BTM/REIT 9 neg.
indicator 9 returns
0.276***
(9.124)
0.365***
(9.207)
0.206*
(1.928)
0.157**
(2.477)
Lag EPS 0.745***
(16.208)
0.754***
(14.535)
0.724***
(15.998)
0.726***
(16.041)
Negative earnings indicator -0.038***
(-8.606)
-0.040***
(-9.034)
-0.035***
(-7.055)
-0.035***
(-7.059)
Neg. earnings indicator 9 earnings -0.416***
(-7.934)
-0.420***
(-7.278)
-0.374***
(-6.960)
-0.376***
(-7.019)
Intercept 0.023***
(6.969)
0.021***
(6.407)
0.029***
(7.343)
0.029**
(7.359)
N 155,201 103,466 155,201 155,201
Firm clusters 17,359 16,332 17,359 17,359
Year clusters 45 45 45 45
v2 2,420.7*** 1,984.6*** -3,164.8 2,360.8***
Adjusted R2 0.416 0.425 0.383 0.384
Additional tests
AT ratio
High BTM/REIT 0.914 0.915 0.932 0.866
Delta method 95 % interval [0.860, 0.968] [0.859, 0.971] [0.841, 1.023] [0.795, 0.937]
Feiller method 95 % interval [0.858, 0.967] [0.856, 0.970] [0.815, 1.026] [0.776, 0.929]
Control 0.949 0.963 0.897 0.898
Delta method 95 % interval [0.852, 1.046] [0.862, 1.064] [0.802, 0.993] [0.802, 0.995]
Feiller method 95 % interval [0.850, 1.047] [0.860, 1.065] [0.798, 0.992] [0.798, 0.994]
Difference in ratios (F-stat) -0.035
(0.795)
-0.048
(1.099)
0.034
(0.242)
-0.032
(0.242)
This table reports coefficient estimates from Basu (1997)-type regressions of earnings on returns with cross-sectional
comparisons based on book-to-market (columns 1 and 2) and REITs (real estate investment trusts) and non-REITs
(columns 3 and 4). Table 2 describes our sample of firms and variable definitions. For BTM, the X-section indicator is set to
one for column 1 (column 2) if beginning BTM is greater than the median (66th percentile) of BTM for all sample firms and
zero if BTM is below the median (33rd percentile). For REITs, the X-section indicator is set to one if the firm’s SIC code is
equal to 6,798 and zero otherwise. Earnings are earnings per share from continuing operations scaled by the beginning-
of-fiscal-year price. Returns are annual buy-and-hold returns. All variables except indicator variables are winsorized at the
1st and 99th percentile within Compustat fiscal years. Two-tailed t-statistics are shown below coefficient estimates, with
standard errors clustered by firm and fiscal year. Significance levels for the ratio tests are computed using the delta method,
however we present confidence intervals using both the delta method and the Fieller method as explained in the Appendix
(using 5 % significance). *, **, and *** indicate significance at 10, 5, and 1 %
The effects of economic rents on stock returns
123
timeliness relative to other firms. The interaction coefficients for our REIT indicator and
negative returns in the two specifications are 0.206 (significant at 5 %) and 0.157 (sig-
nificant at 5 %), suggesting that REITs tend to exhibit more asymmetric timeliness.
However, the AT ratio, which controls for cross-sectional differences in the amount of
returns due to separable assets, does not indicate that REITs have greater asymmetric
timeliness. To the extent that REITs exhibit no more or less asymmetric loss recognition
than other firms, this suggests that the positive interaction coefficient is due to the high
proportion of returns that are attributable to separable assets, a bias remedied by using
comparisons based on the AT ratio.
4.3 Effect of interacting returns with the book-to-market ratio
Table 5 presents regressions that estimate regression (7) where returns are interacted with
the book-to-market ratio as a proxy for the portion k of returns attributable to separable net
assets rather than future rents. Consistent with prediction P2, the AT coefficients are higher
in the regressions where returns are interacted with the book-to-market ratio (significant at
1 %). The higher coefficient on positive returns is also consistent with the book-to-market
interaction controlling for the dampening effect of returns attributable to future rents.14
We also compare the AT ratios from the two specifications as a check on the degree to
which bias impacts the estimated coefficients in the regression where returns are interacted
with the book-to-market ratio. The ratios are nearly identical (p value for the difference in
ratios is 0.82), suggesting that, in this case, bias does not appear to have a large impact on
the coefficients. This does not suggest that this holds generally, though, and we recommend
that researchers perform this comparison before relying on coefficients estimated from a
regression where returns are interacted with the book-to-market ratio.
4.4 Effects of goodwill
Table 6 presents our tests of predictions P3 and P4 on how the accounting recognition of
goodwill impacts estimates of conservatism with respect to separable assets. We test these
predictions on both the full sample and a limited sample that includes only firms that report
goodwill. Prediction P1 states that the AT coefficient will be larger in a regression where
earnings include the effects of goodwill. While the coefficients are larger when earnings
include goodwill, the difference in coefficients is statistically insignificant.
Prediction P4 pertains to the AT ratio. The full sample regression with unadjusted
earnings yields an AT ratio of 0.897, which is equal to the AT ratio when earnings excludes
goodwill. When we restrict the sample to firms that report goodwill, the ratio is larger when
earnings include the effects of goodwill, but the difference is again insignificant. We
conclude that goodwill does not materially impact the estimation of conservatism in the
broad sample. We also compare confidence intervals on the ratios using both the Delta
method and the Fieller method. In all cases, the two confidence intervals are nearly
14 This evidence must be viewed as consistent with P2, but not as conclusive. This is because multiplyingreturns by any number between zero and one will tend to generate larger coefficients. The interpretation ofthe estimates rests on their derivation from our econometric model. If the book-to-market ratio is unrelatedto the structural construct k, then its interaction with returns when estimating (7) can be viewed as a sourceof bias. For the specification to be totally invalid would require the book-to-market ratio to be totallyunrelated to the portion of firm value due to separable net assets, which seems unlikely; however, the book-to-market ratio is an imperfect measure of k and therefore does yield some estimation bias as we notedearlier.
J. A. Caskey, K. Peterson
123
identical, suggesting the Delta method provides a reasonable approximation to the standard
errors for the ratio.
4.5 Additional settings
The importance of the two prescriptions we outline depends on the nature of the study. For
example, a recent study by Ettredge et al. (2012) compare asymmetric timeliness between
restated periods for a sample of firms and control samples comprised of either other firms,
Table 5 Results of asymmetric timeliness estimates with book-to-market interactions
Returns(1)
Returns 9 BTM(2)
Negative return indicator -0.002(-0.521)
0.005(1.515)
Returns 0.022**(2.204)
0.031***(4.462)
Negative indicator 9 returns 0.184***(9.288)
0.331***(8.397)
Lag EPS 0.610***(7.562)
0.686***(8.081)
Negative earnings indicator -0.260***(-3.397)
-0.378***(-4.909)
Neg. earnings indicator 9 earnings -0.045***(-5.438)
-0.051***(-6.552)
Intercept 0.038***(5.697)
0.030***(4.467)
N 155,201 155,201
Firm clusters 17,359 17,359
Year clusters 45 45
v2 1,337.0*** 3,266.1***
Adjusted R2 0.379 0.432
Additional tests
Difference in AT coefficients (F-stat) 0.148***(14.019)
AT ratio 0.894 0.914
95 % confidence intervals
Delta method [0.799, 0.990] [0.871, 0.956]
Feiller method [0.795, 0.988] [0.866, 0.953]
Difference in ratios (F-stat) 0.020(0.819)
This table reports coefficient and ratio estimates from Basu (1997)-type regressions of earnings on returnsusing the original specification and including interactions with book-to-market. Table 2 describes oursample of firms and variable definitions. Earnings are earnings per share from continuing operations scaledby the beginning-of-fiscal-year price. Returns are annual buy-and-hold returns. All variables except indi-cator variables are winsorized at the 1st and 99th percentile within Compustat fiscal years. Two-tailedt-statistics are shown below coefficient estimates, with standard errors clustered by firm and fiscal year.Significance levels for the ratio tests are computed using the delta method, however we present confidenceintervals using both the delta method and the Fieller method as explained in the Appendix (using 5 %significance). *, **, and *** indicate significance at 10, 5, and 1 %
The effects of economic rents on stock returns
123
or the same firms in the post-restatement period. They find lower asymmetric timeliness in
the periods that were later restated. Given the nature of this study using a matched control
sample and the firm itself as a control, it is unlikely that asset composition is an important
factor that potentially influences the results. However, there are other settings where our
prescriptions may prove useful. Table 7 presents regression estimates for two settings to
highlight cases where the ratio provides different inferences relative to the AT coefficient.
Table 6 Results of asymmetric timeliness estimates including and excluding the effects of goodwill
All firms Only firms with goodwill
As reported(1a)
Excl. goodwill(1b)
As reported(2a)
Excl. goodwill(2b)
Negative return indicator -0.002(-0.360)
-0.003(-0.622)
0.002(0.716)
0.001(0.333)
Returns 0.021**(2.220)
0.020**(2.150)
0.016**(2.085)
0.013*(1.891)
Negative indicator 9 returns 0.182***(9.324)
0.176***(9.060)
0.174***(9.308)
0.162***(8.119)
Lag EPS 0.730***(16.061)
0.724***(15.975)
0.485***(9.779)
0.456***(9.940)
Negative earnings indicator -0.037***(-6.732)
-0.035***(-7.013)
-0.043***(-6.097)
-0.039***(-5.941)
Neg. earnings indicator 9 earnings -0.387***(-6.887)
-0.374***(-6.942)
-0.170**(-2.200)
-0.125*(-1.758)
Intercept 0.028***(6.981)
0.028***(7.280)
0.034***(8.160)
0.036***(10.157)
N 155,201 155,201 35,645 35,645
Firm clusters 17,359 17,359 7,214 7,214
Year clusters 45 45 21 21
v2 1,709.1*** 1,510.4*** 1,191.5*** 955.7***
Adjusted R2 0.386 0.382 0.335 0.322
Additional tests
Difference in AT coefficients (F-stat) 0.006(0.095)
0.012(0.358)
AT Ratio 0.897 0.897 0.918 0.924
95 % confidence intervals
Delta method [0.804, 0.989] [0.802, 0.992] [0.839, 0.997] [0.843, 1.004]
Feiller method [0.800, 0.988] [0.798, 0.991] [0.835, 0.995] [0.838, 1.003]
This table reports coefficient and ratio estimates from Basu (1997)-type regressions of earnings on returnswith both as-reported earnings and earnings excluding the effects of goodwill, such as write-downs andamortization. Table 2 describes our sample of firms and variable definitions. The first set of tests includes allfirms, while the second set includes only those firms with goodwill during the period. Earnings are earningsper share from continuing operations scaled by the beginning-of-fiscal-year price. The effects of theamortization of goodwill and goodwill write-downs on earnings per share are obtained from Compustsat(GDWLIEPS and GDWLAM). Returns are annual buy-and-hold returns. All variables except indicatorvariables are winsorized at the 1st and 99th percentile within Compustat fiscal years. Two-tailed t-statisticsare shown below coefficient estimates, with standard errors clustered by firm and fiscal year. Significancelevels for the ratio tests are computed using the delta method, however we present confidence intervals usingboth the delta method and the Fieller method as explained in the Appendix (using 5 % significance). *, **,and *** indicate significance at 10, 5, and 1 %
J. A. Caskey, K. Peterson
123
Our first setting examines differences in asymmetric timeliness for high-technology and
low-technology firms as defined in Francis and Schipper (1999). Kwon et al. (2006) find
that high-technology firms have higher AT coefficients than low technology firms, while
Chandra et al. (2004) find high technology firms have a lower AT coefficient than non
high-technology firms. The results in Table 7 suggest the AT coefficient is lower for high-
technology firms using the regression specification as in (1) (coefficients negative and
significant at 1 %), consistent with the Chandra et al. (2004) findings.15 However, when we
attempt to control for k by estimating the regression by interacting returns with book-
to-market, the coefficient becomes positive although insignificant at 10 %. The differences
in ratios for both specifications indicate high-technology firms do have greater asymmetric
timeliness relative to low-technology firms (significant at 1 %). These results highlight the
importance of estimating the ratio and not just the coefficient when comparing firms with
dissimilar portions of returns related to separable assets versus future rents.
Our second setting in Table 7 examines differences in asymmetric timeliness between
firms with high versus low probability of litigation. Basu (1997) argues and finds that an
increased probability of litigation is associated with greater asymmetric timeliness. We
estimate the probability of litigation as in Khan and Watts (2009), fitted with the param-
eters and variables as in Table 3 of Shu (2000) and estimated in the prior year.16 Firms are
designated as high probability of litigation if the predicted probability of litigation is
greater than the 90th percentile of all firms in the sample. The AT coefficients for high
probability of litigation interacted with negative returns is positive and significant at 1 % in
both specifications, suggesting that greater probability of litigation is associated with
greater asymmetric timeliness. Unlike some of the prior results, testing the differences in
ratios in this setting produces similar inferences to the AT coefficient in this case. Because
the high litigation firms have low book-to-market ratios and a lower coefficient on positive
returns, this suggests that these firms have a relatively low portion of returns associated
with separable assets, which biases against them having a higher AT coefficient than the
broader population.17 These results highlight the need to examine ratios in conjunction
with coefficients, as researchers may not know ex ante the extent of bias in the AT
coefficient.
5 Empirical properties of the asymmetric timeliness (AT) ratio
There is no theoretically ideal benchmark for measuring conservatism with which we can
test our model. Consequently, the previous section demonstrates that our model’s pre-
dictions perform well with commonly used archival data, which suggests that the model is
reasonably descriptive of actual data. This section provides additional evidence for use of
the AT ratio by showing that it exhibits intuitive empirical properties such as an association
with other empirical proxies for conservatism. Prior research has shown that AT coeffi-
cients are negatively associated with other conservatism proxies, which suggests that they
fail to accurately measure the degree of accounting conservatism (Givoly et al. 2007). We
15 The difference in coefficient results relative to Kwon et al. (2006) may be due to differences in sampleconstruction.16 The number of observations is reduced in these tests due to the variable restrictions necessary to estimatethe Shu (2000) model.17 Untabulated analysis shows that the mean (median) beginning book-to-market ratio of high litigationfirms is 0.560 (0.447) versus 0.781 (0.626) for the non-high litigation firms.
The effects of economic rents on stock returns
123
Table 7 Results of asymmetric timeliness estimates using high-tech and high probability of litigationsample partitions
Hi-tech/low-tech High/low prob litigation
Returns
(1)
BTM 9 returns
(2)
Returns
(3)
BTM 9 returns
(4)
Negative return indicator 0.007**
(2.197)
0.005*
(1.648)
0.000
(-0.125)
0.001
(0.516)
Returns 0.034***
(4.502)
0.034***
(4.957)
0.020**
(2.127)
0.026***
(3.251)
Negative returns indicator 9 returns 0.201***
(11.170)
0.323***
(8.059)
0.188***
(10.731)
0.306***
(7.944)
Hi-tech/High-PLIT -0.008***
(-2.797)
-0.008***
(-3.588)
-0.012***
(-3.352)
-0.007**
(-2.355)
Tech/PLIT 9 neg. indicator -0.015***
(-3.448)
-0.003
(-0.690)
0.017**
(2.548)
0.004
(1.108)
Tech/PLIT 9 returns -0.029***
(-6.151)
-0.024***
(-5.452)
-0.006
(-1.246)
-0.013*
(-1.700)
Tech/PLIT 9 neg. indicator 9 returns -0.075***
(-3.445)
0.059
(1.467)
0.088***
(3.271)
0.222***
(5.406)
Lag EPS 0.714***
(15.173)
0.793***
(13.381)
0.697***
(13.184)
0.788***
(11.449)
Negative earnings indicator -0.038***
(-6.984)
-0.039***
(-9.288)
-0.040***
(-6.668)
-0.037***
(-6.846)
Neg. earnings indicator 9 earnings -0.372***
(-6.473)
-0.481***
(-7.517)
-0.361***
(-5.344)
-0.475***
(-6.325)
Intercept 0.027***
(8.175)
0.022***
(6.019)
0.030***
(7.354)
0.022***
(5.362)
N 155,201 155,201 100,765 100,765
Firm clusters 17,359 17,359 11,384 11,384
Year clusters 45 45 44 44
v2 2,428.1*** 3,836.6*** 2,991.8*** 3,753.6***
Adjusted R2 0.391 0.434 0.374 0.416
Additional tests
AT ratio
Hi-tech/high-PLIT 0.956 0.974 0.952 0.976
Delta method 95 % interval [0.860, 1.051] [0.927, 1.022] [0.877, 1.026] [0.930, 1.022]
Feiller method 95 % interval [0.852, 1.054] [0.923, 1.021] [0.873, 1.026] [0.928, 1.022]
Control 0.854 0.905 0.905 0.921
Delta method 95 % interval [0.789, 0.919] [0.862, 0.948] [0.817, 0.992] [0.870, 0.972]
Feiller method 95 % interval [0.786, 0.917] [0.856, 0.945] [0.816, 0.992] [0.864, 0.969]
Difference in ratios (F-stat) 0.102***
(10.187)
0.069***
(21.360)
0.047**
(5.424)
0.055***
(9.100)
This table reports coefficient and ratio estimates from Basu (1997)-type regressions of earnings on returns using sample
partitions of high-tech/low-tech firms and high/low probability of litigation. Each partition is estimated using a specifi-
cation using returns and returns interacted with book-to-market. Table 2 describes our sample of firms and variable
definitions. High-tech indicator is set to one if the 3-digit SIC is one of the following (283, 357, 360, 361, 362, 363, 364,
365, 366, 367, 368, 481, 737, 873) and zero otherwise. The high probability of litigation indicator equals one if the
estimated probability of litigation is greater than the 90th percentile of all sample firms. The probability of litigation is
estimated as in Khan and Watts (2009), and fitted using the parameters and variables in Table 3 of Shu (2000). All variables
except indicator variables are winsorized at the 1st and 99th percentile within Compustat fiscal years. Two-tailed t-statistics
are shown below coefficient estimates, with standard errors clustered by firm and fiscal year. Significance levels for the
ratio tests are computed using the delta method, however we present confidence intervals using both the delta method
and the Fieller method as explained in the Appendix (using 5 % significance). *, **, and *** indicate significance at 10, 5,
and 1 %
J. A. Caskey, K. Peterson
123
therefore test cross-industry associations between the AT ratio and other conservatism
measures. Furthermore, Givoly and Hayn (2000) and Givoly et al. (2007) find that the AT
coefficient is not stable over time, which is inconsistent with the stability of accounting
rules.18 We therefore test the time-series stability of the AT ratio.
5.1 AT measures across industries
We compare a number of accounting conservatism measures, including the AT coefficient
and ratio, across industries in order to measure the AT ratio’s association with other
measures of accounting conservatism. We estimate the AT coefficients and ratios using
both (1), the traditional Basu (1997) regression, and (7), where book-to-market is inter-
acted with returns. When comparing measures of conservatism across industries, we focus
on measures that reflect ex post conservatism in the form of recognizing bad news, which
should be correlated with the AT ratio.19 Ex post conservatism impacts both the book-
to-market ratio and Givoly and Hayn’s (2000) cumulative discretionary accrual (CDA)
measure because the recognition of bad news reduces both book values and cumulative
discretionary accruals.20 Basu (1997) and Ball and Shivakumar (2005) measure conser-
vatism based on the incremental persistence of negative earnings changes relative to
positive earnings changes, which reflects ex post conservatism. We therefore expect a
negative relation between the ratio and book-to-market and CDA and a positive relation
between the ratio and the relative persistence of negative earnings changes.
We also compare our ratio to a measure based on a conservatism ratio (CR) from Callen
et al. (2010) that utilize the Vuolteenaho (2002) return decomposition to infer earnings
news. Callen et al. (2010) define their conservatism ratio as the shock to current earnings
divided by the total news about earnings during the period. In other words, it represents the
amount of information about all current and future earnings recognized in the current
period earnings. We construct the CR rank as in Callen et al. (2010) by ranking the
conservatism ratio into quintiles within quintiles of negative unexpected returns
(rit � Et�1½rit�\0). We use the average rank within industry as an industry-level measure
of conservatism, and expect a positive relation with other measures of asymmetric
timeliness.
We test the relation between measures of conservatism by computing them for each of
the 48 Fama and French (1997) industry groups. Table 8 compares Pearson and Spearman
rank correlation measures across these groups. Below the correlations we have listed the
number and percentage of industries where the conservatism estimate is significant.21 The
18 We also recognize that economic circumstances affect the implementation of standards and cause var-iation in conservatism measures across time. Our model explicitly allows this. However, as Givoly et al.(2007) point out, the variation across time may partially be due to poor estimation in years with few negativenews observations.19 Measures of ex ante conservatism, including Easton and Pae’s (2004) measures of conservatism due topositive NPV projects and rules and Penman and Zhang’s (2002) ‘hidden reserves’ measure, should notnecessarily be correlated with the AT ratio.20 The market-to-book ratio is also impacted by ex ante conservatism since ex ante conservatism excludesrents from book values and can also understate separable assets via, for example, accelerated depreciation.21 For the AT coefficients, discretionary accruals, and the persistence coefficient the null to test significanceis zero. For the ratios and book-to-market the null to test significance is one, and for the CR rank significanceis tested relative to a null of three.
The effects of economic rents on stock returns
123
sign of the AT coefficient correlations are opposite of the predictions in several cases,
consistent with cross-sectional comparisons of the coefficient being impacted by the
portion of returns due to separable assets. Consistent with Givoly et al. (2007), the AT
coefficient is positively associated with book-to-market although the correlation is not
significant at 10 %. Moreover, the AT coefficient is positively correlated with cumulative
discretionary accruals and negatively correlated with the persistence of negative earnings
coefficient (although insignificant). However, we note that the persistence coefficient is not
estimated very precisely within industries as indicated by only 18 of 48 industries
exhibiting a significant persistence coefficient estimate. As expected, the AT coefficient is
positively associated with the CR from Callen et al. (2010).
Table 8 shows that the AT ratio exhibits the predicted correlations for all of the
reported measures of conservatism except for the Spearman correlation with the per-
sistence coefficient, which, may also be affected by the imprecise estimates of persis-
tence. The AT coefficient estimated with returns interacted with BTM has correlations
more consistent with expectations than the AT coefficient estimated with non-interacted
returns; however, the Pearson correlation with discretionary accruals is insignificant.
Both the Pearson and Spearman correlations with the persistent coefficient are negative,
inconsistent with predictions, but insignificant. This may again be due to the fact that
the persistence coefficient is fairly noisy. The ratio computed from the regression with
returns interacted with the book-to-market ratio has insignificant correlations with the
AT coefficient, the persistence coefficient and the CR Rank measure. Otherwise, it
exhibits the predicted correlations.
5.2 Time series properties of the AT ratio
In untabulated analysis, we also examine the variation across time for the AT coefficient
and ratio estimated with both returns and returns interacted with BTM. The coefficients of
variation (standard deviation divided by the mean) for the coefficient, coefficient with
BTM interaction, ratio, and ratio with BTM interaction for our sample are 0.518, 0.424,
0.172 and 0.103. The ratios have less variability over time than the coefficients, a desirable
property since asymmetric timeliness, as required by accounting standards, does not
change dramatically from year to year. Estimating using returns interacted with BTM
reduces the variation over time, although not as much as the reduction achieved by the ratio
measures.
The time series data also indicate that there was a fundamental shift in both the AT
coefficient and the ratio in the mid-1970 s. The coefficients of variation for years prior
to 1975 are 1.175, 0.738, 1.082, and 1.165 for the four measures, substantially higher
than years following 1975. This shift could be explained by the emergence of the
Financial Accounting Standards Board (FASB) in 1973 and the early standards written
by the board. In 1975, the FASB issue Statement No. 5, Accounting for Contingencies,
which required the recognition of loss contingencies but not gain contingencies (FASB
1975). Statement No. 15, Accounting by Debtors and Creditors for Troubled DebtRestructuring, was issued in 1977, requiring creditors to reduce yields for adjustments
of payment terms of troubled debt if the aggregate payments were not less than the
carrying amount; however, if the aggregate payments were less than the carrying
amount, the creditor must reduce the carrying amount and record a loss. Therefore,
smaller losses adjusted future gains, while larger losses required current period recog-
nition consistent with asymmetric timeliness.
J. A. Caskey, K. Peterson
123
Ta
ble
8C
orr
elat
ion
sb
etw
een
con
serv
atis
mm
easu
res
Pre
dic
tion
(coef
f&
rati
o)
AT
coef
f
pos
Rat
io
pos
AT
coef
fw
/BT
M
pos
Rat
iow
/BT
M
pos
Book-t
o-m
arket
neg
Dis
cac
crual
s
neg
Per
sist
coef
f
pos
CR
rank
pos
AT
coef
fici
ent
pos
0.1
82
(0.2
36)
0.4
72
(0.0
01)
0.0
52
(0.7
38)
0.2
27
(0.1
38)
0.1
61
(0.2
97)
-0.2
42
(0.1
14)
0.3
75
(0.0
12)
Rat
iopos
0.2
22
(0.1
48)
0.6
31
(0.0
00)
0.8
05
(0.0
00)
-0.6
13
(0.0
00)
-0.4
64
(0.0
02)
0.1
25
(0.4
19)
0.2
83
(0.0
63)
AT
coef
fw
/BT
Mpos
0.6
80
(0.0
00)
0.3
27
(0.0
30)
0.5
25
(0.0
00)
-0.4
30
(0.0
04)
-0.4
39
(0.0
03)
-0.0
85
(0.5
83)
0.2
83
(0.0
63)
Rat
iow
/BT
Mpos
0.0
40
(0.7
99)
0.7
58
(0.0
00)
0.3
36
(0.0
26)
-0.3
69
(0.0
14)
-0.3
47
(0.0
21)
-0.0
04
(0.9
81)
0.1
62
(0.2
93)
Book-t
o-m
arket
neg
0.2
26
(0.1
40)
-0.5
06
(0.0
01)
-0.2
76
(0.0
70)
-0.4
02
(0.0
07)
0.7
12
(0.0
00)
-0.3
49
(0.0
20)
0.0
63
(0.6
84)
Dis
cret
ionar
yac
crual
sneg
0.2
79
(0.0
67)
-0.2
85
(0.0
61)
-0.1
57
(0.3
10)
-0.3
28
(0.0
30)
0.7
21
(0.0
00)
-0.2
30
(0.1
33)
-0.0
20
(0.8
95)
Per
sist
ence
coef
fici
ent
pos
-0.1
91
(0.2
14)
0.3
08
(0.0
42)
-0.1
15
(0.4
59)
0.0
72
(0.6
41)
-0.3
44
(0.0
22)
-0.2
14
(0.1
64)
-0.0
45
(0.7
70)
CR
rank
pos
0.4
12
(0.0
06)
0.3
23
(0.0
32)
0.3
08
(0.0
42)
0.1
04
(0.5
02)
0.0
68
(0.6
59)
-0.0
49
(0.7
51)
0.1
14
(0.4
60)
Num
ber
signifi
cant
46
34
46
38
36
43
18
30
Per
cent
signifi
cant
95.8
%70.8
%95.8
%79.2
%75.0
%89.6
%37.5
%68.2
%
Null
for
signifi
cance
test
s0.0
00
1.0
00
0.0
00
1.0
00
1.0
00
0.0
00
0.0
00
3.0
00
This
table
report
sth
eco
rrel
atio
ns
bet
wee
ndif
fere
nt
mea
sure
sof
acco
unti
ng
conse
rvat
ism
bas
edon
esti
mat
esfo
rth
e48
Fam
aan
dF
rench
(1997
)in
dust
rygro
ups.
The
table
dis
pla
ys
Pea
rson
corr
elat
ions
bel
ow
the
dia
gonal
and
Spea
rman
corr
elat
ions
above,
wit
hp
val
ues
report
edin
par
enth
esis
bel
ow
the
corr
elat
ion
esti
mat
es.
The
asym
met
ric
tim
elin
ess
(AT
)
coef
fici
ent
isth
ees
tim
ate
of
b3
ina
regre
ssio
nof
pri
ce-s
cale
dea
rnin
gs
per
shar
eon
retu
rns:
EP
Sit/P
i,t2
1=
b0
?b
11(r
it\
0)
?b
2r i
t?
b3
1(r
it\
0)
r it
?e i
tfo
rfi
rmi
and
yea
rt
wher
eth
ein
dic
ator
var
iable
1(r
it\
0)
equal
sone
when
retu
rns
are
neg
ativ
ean
dze
rooth
erw
ise.
Rat
ioeq
ual
sth
ees
tim
ates
b3/(b
2?
b3).
Cum
ula
tive
dis
cret
ionar
yac
crual
sar
eth
esu
mof
tota
lac
crual
s(n
etin
com
e?
dep
reci
atio
n-
cash
from
oper
atio
ns)
less
oper
atin
gac
crual
s(c
han
ge
inac
counts
rece
ivab
le?
chan
ge
inin
ven
tory
-ch
ange
inac
counts
pay
able
-
chan
ge
inta
xes
pay
able
)sc
aled
by
tota
las
sets
inth
efi
rm’s
firs
tyea
rw
ith
avai
lable
dat
a(G
ivoly
and
Hay
n2000
).T
he
Per
sist
ence
coef
fici
ent
isth
ees
tim
ate
b3
from
the
foll
ow
ing
regre
ssio
n:
Ear
nt
=b
0?
b1
1(E
arn
t21\
0)
?b 2
Ear
nt2
1?
b3
1(E
arn
t21\
0)
Ear
nt2
1?
e it
consi
sten
tw
ith
Bal
lan
dS
hiv
akum
ar(2
005
).W
eco
nst
ruct
the
conse
rvat
ism
rati
ora
nk
(CR
rank)
asin
Cal
len
etal
.(2
010
)by
rankin
gth
eco
nse
rvat
ism
rati
o(f
rom
Cal
len
etal
.2010
)in
toquin
tile
s,w
ithin
quin
tile
sof
neg
ativ
eunex
pec
ted
retu
rns.
We
use
the
aver
age
rank
wit
hin
indust
ryas
anin
dust
ry-l
evel
mea
sure
of
conse
rvat
ism
.W
eli
stour
pre
dic
tions
for
corr
elat
ions
wit
hth
eco
effi
cien
tan
dra
tio
mea
sure
sonly
list
edas
‘pos’
for
posi
tive
and
‘neg
’fo
r
neg
ativ
e.*,
**,
and
***
indic
ate
signifi
cance
at10,
5,
and
1%
The effects of economic rents on stock returns
123
6 Conclusion
Many empirical studies on accounting conservatism employ regression-based measures of
earnings’ asymmetric recognition of bad news as in Basu (1997). Recent studies have
shown that such tests may be unsuitable for some comparisons of conservatism across
groups of firms or across time (Givoly, Hayn and Natarajan 2007). We develop an
econometric model that shows that coefficient estimates appear to provide valid indications
of the presence of conservatism but also reflect the portion of firm value attributable to
future rents. This occurs because book values primarily represent a firm’s separable net
assets while market values and returns incorporate the value of both separable assets and
the value of non-separable assets and expected future economic rents. Two groups of firms
with identical accounting policies can therefore have different asymmetric timeliness
coefficients simply because one group’s returns are more associated with future rents that
are not included in book values. We show that a simple test using a ratio of regression
coefficients controls for the effects of value composition. This allows for comparisons of
conservatism across groups.
We derive several predictions from our empirical model and test them on a large sample
from the CRSP and Compustat databases often used in tests of accounting conservatism.
We find support for our predictions, which suggests that the model provides a reasonable
description of actual data. Furthermore, we verify that the ratio tests are well-behaved, in a
statistical sense, by demonstrating that delta method confidence intervals for the ratios
provide very close approximations to more robust confidence intervals based on Fieller’s
Theorem (Fieller 1954).
We conclude by noting that the usefulness of coefficient ratios extends beyond studies
of accounting conservatism. The test design in this study compared the incremental
response of an event, bad news, across groups of firms. A variety of studies in accounting,
such as those that compare the effects of regulation across groups of firms, utilize a similar
test design. We conjecture that inferences based on coefficient ratios may be more
appropriate than inferences based on the coefficients themselves, in cases where coefficient
biases may be associated with the sample partition.
Acknowledgments We thank an anonymous referee, David Aboody, Sudipta Basu, Peter Demerjian,Carla Hayn, Jack Hughes, Steve Matsunaga, Karl Muller, Eddie Riedl and workshop participants at AAAFARS 2009 Mid-Year Meeeting, Pennsylvania State University, UCLA and University of Southern Cali-fornia for their helpful comments. This paper was previously titled ‘‘On the estimation of the asymmetrictimeliness of earnings: Inference and bias corrections.’’
Appendix
This appendix describes the construction of confidence intervals for ratios of regression
coefficients based on Fieller’s Theorem (Fieller 1954) as implemented in Zerbe (1978).22
The confidence intervals are based on transforming a hypothesis for the ratio to an
equivalent linear hypothesis. Given a value k, a vector b of p estimated coefficients and
vectors n and d, the ratio n0b=d0b [ k if and only if n0b=kd0b [ 0.23 Given estimated
coefficients b and covariance matrix R , a t-statistic for the inequality is:
22 Also see Staiger, Stock and Watson (1997) for an example application of this approach.23 If d0b\ 0 the second inequality is reversed, but (A2) still determines the confidence interval.
J. A. Caskey, K. Peterson
123
T ¼ ðn0b� kd0bÞ=ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
n0Rn� 2kn0Rd þ k2d0Rdp
: ð9ÞGiven a confidence level a, sample size N and critical t-statistic ta/2 with N - p degrees
of freedom such that P(-ta/2 \ T \ ta/2) = 1 - a, the confidence intervals on the ratio
level k solve:
a0 þ 2a1k þ a2k2� 0; ð10Þ
a0 ¼ ðn0bÞ2 � t2a=2n0Rn a1 ¼ t2
a=2n0Rd � ðn0bÞðd0bÞ a2 ¼ ðd0bÞ2 � t2a=2d0Rd ð11Þ
If a2 [ 0, then the 1 - a confidence interval for the ratio is ð�a1 �ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
a21 � a0a2
p
Þ=a2.24
If a2 \ 0 then the confidence interval is unbounded.25 In this case, if a21 � a0a2\0, then
the confidence interval is the entire real line. If a2 \ 0 and a21 � a0a2 [ 0, then the con-
fidence interval is the complement of ð�a1 �ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
a21 � a0a2
p
Þ=a2.26
References
Ahmed A, Billings B, Morton R, Stanford-Harris M (2002) The role of accounting conservatism in miti-gating bondholder-shareholder conflicts over dividend policy and in reducing debt costs. Acc Rev77(4):867–890
Ball R, Shivakumar L (2005) Earnings quality in UK private firms: comparative loss recognition timeliness.J Acc Econ 39(1):83–128
Ball R, Kothari S, Nikolaev V (2009) Econometrics of the Basu asymmetric timeliness coefficient andaccounting conservatism. University of Chicago and MIT, Working paper
Barth M, Landsman W, Lang M (2008) International accounting standards and accounting quality. J AccRes 46(3):467–498
Basu S (1977) Investment performance of common stocks in relation to their price-earnings ratios: a test ofthe efficient market hypothesis. J Financ 32(3):663–682
Basu S (1997) The conservatism principle and the asymmetric timeliness of earnings. J Acc Econ24(1):3–37
Basu S (1999) Discussion of International differences in the timeliness, conservatism, and classification ofearnings. J Acc Res 37:89–99
Beatty A (2007) Discussion of ‘Asymmetric timeliness of earnings, market-to-book and conservatism infinancial reporting’. J Acc Econ 44(1–2):32–35
Beaver W, Ryan S (2000) Biases and lags in book value and their effects on the ability of the book-to-market ratio to predict return on equity. J Acc Res 38(1):127–148
Brauer S, Westermann F (2012) On the time series measure of conservatism: a threshold autoregressivemodel. Rev Quant, Finance Account (Forthcoming)
Buonaccorsi J (1979) Letters to the editor. Am Stat 33(3):162Callen J, Hope O, Segal D (2010) The pricing of conservative accounting and the measurement of con-
servatism at the firm-year level. Rev Acc Stud 15:145–178Chandra U, Wasley C, Waymire G (2004) Income conservatism in the U.S. technology sector. Working
paper, University of Kentucky, University of Rochester and Emory UniversityDietrich J, Muller K, Riedl E (2007) Asymmetric timeliness tests of accounting conservatism. Rev Acc Stud
12(1):95–125
24 See Buonaccorsi (1979) and Zerbe’s (1978) response for a discussion that a2 [ 0 implies a21 � a0a2 [ 0,
and a discussion of the confidence interval when a2 \ 0.
25 For example, if the ratio’s denominator is the jth regressor bj, then a2 ¼ ðb2j =r
2jj � t2
a=2Þr2jj so that
a2\0and the confidence interval is unbounded if bj is insignificant based on a two-tailed test with sig-nificance a.26 We computed the confidence intervals using the regFieller.ado Stata program available on JudsonCaskey’s website: http://webspace.utexas.edu/jc2279/www/data.html.
The effects of economic rents on stock returns
123
Easton P, Pae J (2004) Accounting conservatism and the relation between returns and accounting data. RevAcc Stud 9(4):495–521
Elbannan M (2011) Accounting and stock market effects of international accounting standards adoption inan emerging economy. Rev Quant Financ Acc 36(2):207–245
Eng L, Lin Y (2012) Accounting quality, earnings management and cross-listings: evidence from China.Rev Pac Basin Financ Mark Policies 15(2):1250009
Ettredge M, Huang Y, Zhang W (2012) Earnings restatements and differential timeliness of accountingconservatism. J Acc Econ 53:489–503
Fama E, French K (1992) The cross-section of expected stock returns. J Financ 47(2):427–465Fama E, French K (1997) Industry costs of equity. J Financ Econ 43(2):153–193Fieller E (1954) Some problems in interval estimation. J R Stat Soc B Met 16(2):175–185Financial Accounting Standards Board (FASB) (1975) Accounting for contingencies. Statement of Financial
Accounting Standards No. 5. FASB, Norwalk, CTFrancis J, Schipper K (1999) Have financial statements lost their relevance? J Acc Res 37(2):319–352Givoly D, Hayn C (2000) The changing time-series properties of earnings, cash flows and accruals: has
financial reporting become more conservative? J Acc Econ 29(3):287–320Givoly D, Hayn C, Natarajan A (2007) Measuring reporting conservatism. Acc Rev 82(1):65–106Guay W, Verrecchia R (2006) Discussion of an economic framework for conservative accounting and
Bushman and Piotroski (2006). J Acc Econ 42(1–2):149–165Khan M, Watts R (2009) Estimation and empirical properties of a firm-year measure of accounting con-
servatism. J Acc Econ 48:132–150Krishnan G, Visvanathan G (2008) Does the SOX definition of an accounting expert matter? The association
between audit committee directors’ accounting expertise and accounting conservatism. Contemp AccRes 25(3):827–857
Kwon S, Yin Q, Han J (2006) The effect of differential accounting conservatism on the ‘‘over-valuation’’ ofhigh-tech firms relative to low-tech firms. Rev Quant Financ Acc 27:143–173
Pae J, Thornton D, Welker M (2005) The link between earnings conservatism and the price-to-book ratio.Contemp Acc Res 22(3):693–717
Patatoukas P, Thomas J (2009) Revisiting the Basu (1997) estimate of conditional conservatism. YaleUniversity, Working paper
Penman S, Zhang X (2002) Accounting conservatism, the quality of earnings, and stock returns. Acc Rev77(2):237–264
Pope P, Walker M (1999) International differences in the timeliness, conservatism, and classification ofearnings. J Acc Res 37:53–87
Roychowdhury S, Watts R (2007) Asymmetric timeliness of earnings, market-to-book and conservatism infinancial reporting. J Acc Econ 44(1–2):2–31
Ryan S, Zarowin P (2003) Why has the contemporaneous linear returns-earnings relation declined? Acc Rev78(2):523–553
Shu S (2000) Auditor resignations: clientele effects and legal liability. J Acc Econ 29:173–205Staiger D, Stock J, Watson M (1997) The NAIRU, unemployment and monetary policy. J Econ Perspect
11(1):33–49Vuolteenaho T (2002) What drives firm-level stock returns. J Financ 57(1):233–264Wang D (2006) Founding family ownership and earnings quality. J Acc Res 44(3):619–656Watts R (2003a) Conservatism in accounting part I: explanations and implications. Acc Horiz
17(3):207–221Watts R (2003b) Conservatism in accounting part II: evidence and research opportunities. Acc Horiz
17(4):287–301Zerbe G (1978) On Fieller’s theorem and the general linear model. Am Stat 32(3):103–105
J. A. Caskey, K. Peterson
123