Loss Reversals and Earnings-based Valuation*
Peter Joos and George A. Plesko
Draft 1: January 15, 2002
This version: June 10, 2003
Contact information:
Sloan School of Management
Massachusetts Institute of Technology
E52-325, 50 Memorial Drive
Cambridge, MA 02142-1347
Joos: 617-253-9459, [email protected] (corresponding author)
Plesko: 617-253-2668, [email protected]
* We thank Jeff Abarbanell, Sid Balachandran, Karen Teitel, SP Kothari, Gim Seow, Joe
Weber, Mike Willenborg, Peter Wysocki, and the participants of the accounting
workshop at the Massachusetts Institute of Technology, University of Connecticut, the
2002 European Accounting Association, and the Thirteenth Annual Conference on
Financial Economics and Accounting at the University of Maryland for helpful
comments on an earlier draft. All errors are our own.
Loss Reversals and Earnings-based Valuation
Abstract
We study if investors, when valuing firms reporting losses, assess the likelihood the firm’s loss
will persist and price earnings consistent with the loss’ expected persistence. Estimating a model
of loss reversal, we find both earnings and non-earnings information useful in predicting a loss
firm’s return to profitability. Using the estimated probabilities of loss reversal to classify losses
as persistent or transitory, we find investors do not price persistent losses, but paradoxically price
transitory losses positively. To explain the pricing pattern, we study the properties of losses
across the sub-samples and find transitory losses are driven by negative accruals, indicating their
pricing is consistent with accounting conservatism. By contrast, persistent losses are
characterized by large negative cash flows and, specifically, R&D expenditures. Separating out
the R&D component, we show investors do not value the non-R&D component of persistent
losses but reward firms reporting persistent losses with positive returns for their R&D
expenditures. We corroborate our findings with time-series analyses showing the increased
incidence of negative cash flows and R&D expenditure components of persistent losses relate to
changes in the valuation of persistent losses over the sample period. Our results contribute to
the literature by documenting the valuation implications of losses as a function of their
persistence. Further, we extend Givoly and Hayn (2000), who conclude US firms’ decline in
profitability over time is not associated with a contemporaneous decline in cash flows, by
showing cash flows are a significant and growing determinant for a substantial subset of loss
firms, namely those with persistent losses.
Keywords: earnings; persistence; losses; cash flows; accruals
Data Availability: Data are available from sources identified in the text.
I.
Introduction
The frequency of firms reporting losses has markedly increased over the last three
decades. In the 1990s loss observations constitute about 35% of the US firm-years covered by
the Standard & Poor Compustat database compared to approximately 15% of observations in the
1970s. The increased frequency of firms reporting losses poses a potentially important challenge
for financial statement users who rely on accounting earnings in different decision contexts
(Watts and Zimmerman 1986). Focusing on firm valuation in particular, Modigliani and Miller
(1966) discuss in their seminal paper how accounting earnings are a proxy for the expected and
unobservable earning power of the firms’ assets. However, they also note negative earnings
complicate the use of earnings-based valuation models since a loss impairs the ability of
accounting earnings to act as a proxy for the firm’s assets unobservable earnings power. As a
result, the increase in loss frequency over the last decades calls attention to the question of what
other (accounting) proxies for the earning power of assets financial statement users can use to
value the firm.
We investigate if investors, when valuing firms reporting losses, assess the likelihood the
firm’s loss will persist and price earnings consistent with the loss’ expected persistence. We
focus on loss reversals because a loss places the firm in a temporary position: a return to
profitability is the maintained hypothesis of financial reporting, embodied in the going concern
assumption. Similarly, loss persistence forms the basis of the abandonment option view of loss
valuation: shareholders have the option to redeploy or liquidate the assets of the firm if losses are
expected to continue (Hayn 1995; Berger et al. 1996; Wysocki 2001).
We predict and find loss reversals relate to four generic categories of variables that
capture the business environment and operations of the firm. In a sample of loss observations
1
from 1971 through 2000, we document the probability of loss reversal relates to 1) variables that
describe the loss history of the firm; 2) variables that measure the financial and demographic
profile of the firm; 3) variables that capture the dividend paying behavior of the firm; and 4)
macroeconomic indicators.1
Using the estimated probability of loss reversal as a proxy for the persistence of losses,
we next separate transitory losses (those most likely to reverse in the next period) from persistent
losses (those least likely to reverse) and investigate whether investors price losses differently as a
function of expected persistence. We estimate earnings–return relations for loss firms and find
transitory losses on average obtain significantly positive earnings response coefficients (ERCs).
By contrast, we find investors do not price persistent losses: the ERC for persistent losses is
insignificantly different from zero, consistent with persistent losses being a poor proxy for the
earning power of the firms’ assets. In other words, the more transitory the current loss is
expected to be, the more market returns reflect the information in the loss. The finding is
consistent with Hayn’s (1995) argument for finding a more pronounced stock price response
when firm liquidation is less likely to occur, i.e., when the loss is transitory. However, the result
also presents a paradox, as it suggests the market reduces the market value of a loss firm only if
the firm will soon regain profitability, but not if it expects the firm to continue to generate losses.
To explain the pattern of ERCs across our sample, we next carry out three analyses to
explore in detail the characteristics of loss observations as a function of their persistence. First,
we examine our sample firms’ loss history and find transitory losses are predominantly small
first-time losses. Second, we find our proxy for loss persistence ranks the relative magnitude of
1
Whereas previous research identifies variables that help to predict bankruptcy (e.g., Altman 1968), surprisingly
little is known about what variables help to predict loss reversal of firms.
2
the cash flow and accrual components of losses, with persistent losses exhibiting extreme
negative values of cash flows and accruals. By contrast, the cash flow component of transitory
losses is positive on average, implying negative accruals alone drive transitory losses. Our
finding of a positive ERC for transitory losses is therefore consistent with accounting
conservatism, as defined by Basu (1997), determining transitory losses.
Negative accruals
reflect accounting conservatism, leading not only to bad news earnings (in our case transitory
losses) being less persistent than good news earnings, but also to transitory losses reflecting bad
news in a timely manner relative to returns.
Accounting conservatism, however, does not explain the pricing of persistent losses.
Therefore, in a third analysis, we study if the presence of R&D expenditures or Special Items in
losses varies as a function of persistence. We focus on R&D expenditures since recent research
documents their significant increase over the past two decades. Because US accounting rules
require managers to expense R&D expenditures when they occur, R&D expenditures potentially
cause persistent negative cash flows and influence the properties of reported earnings (or losses).
We also focus on Special Items since research by Givoly and Hayn (2000) and Skinner (2003)
highlights their connection to restructurings and write-offs that typically lower reported earnings
through negative accruals.
We find persistent losses consist of relatively large R&D
expenditures compared to transitory losses. Similarly, we find our proxy for loss persistence
ranks the magnitude of the Special Items component of losses. However, relative to the R&D
component, the magnitude of the Special Items component is much smaller.
Separating out the R&D and Special Items components, we re-examine the valuation of
losses to explore why investors do not ‘punish’ firms with persistent losses (i.e., why their ERC
is zero). Our results show investors do not price the (small) R&D component of transitory
3
losses, but by contrast price the R&D component of persistent losses positively. In other words,
investors acknowledge the relative magnitude of R&D expenditures determines the large
negative cash flow component of persistent losses: they see through the accounting treatment
leading to continuous negative earnings and reward the R&D investment. Investors also price
the Special Items in persistent and transitory losses differently: the coefficient on Special Items is
positive for persistent losses and insignificant for transitory losses. However, we attribute the
pricing difference to the nature of Special Items varying considerably across the sub-samples:
whereas the majority of the transitory losses contain negative Special Items, the corresponding
figure for persistent losses is much smaller.
Our time-series analyses corroborate our initial findings by showing how changes in the
nature of losses over our sampling period influence their valuation. We document losses in
general, but especially losses least likely to reverse, become more persistent over time. Focusing
on valuation, we find investors price transitory losses consistently positively over the sample
period but change their valuation of persistent losses over time: whereas they do not price
persistent losses in the early years of the sample period, they price them negatively during the
last decade. We find the change in valuation relates to a change over time in the relative
magnitude of the components of losses, predominantly as a result of the cash flow component
and in particular the R&D component of losses becoming larger. In other words, our results are
consistent with increasing R&D expenditures determining the change in the nature and valuation
of persistent losses. Although negative Special Items also occur more frequently as a loss
component in recent times, their influence on the nature and pricing of losses is less pronounced
than the effect of the R&D expenditures.
Our research contributes to the valuation literature by establishing the existence of a
4
relation between the probability of loss reversal, our proxy for loss persistence, and investors’
valuation of losses. We show transitory losses follow predominantly from negative accruals and
their pricing is consistent with the valuation effects of accounting conservatism: bad news
earnings (in our case transitory losses) are not only less persistent, but are also more timely
relative to returns.
Our findings for persistent losses extend the literature on accounting
conservatism: we show accounting conservatism explains neither the existence nor the pricing of
persistent losses, consisting largely of large negative cash flows and large R&D expenditures that
investors price positively. We also extend the literature on the (changing) properties of earnings.
Specifically, we complement the findings in Givoly and Hayn (2000), who conclude the
observed decline in profitability of US firms over time does not follow from a decline of cash
flows but rather from increased conservatism in the aggregate. We show however negative cash
flows increasingly influence the time-series properties of persistent loss observations.
Summarizing, our findings enhance our understanding of how investors use information
beyond aggregate earnings to value the firm in the increasingly common case when the firm
reports a loss. Although we find investors price transitory and persistent losses differently, our
findings illustrate investors look beyond mere mechanical patterns of loss persistence to assess
the valuation implications of a loss. Specifically, investors appear to consider the causes and
nature of the loss to assess its long-term implications for firm value.
In the next section we describe our sample and document the prevalence and duration of
losses. In Section III we describe the financial profile of loss observations and present our model
of loss reversals. In Section IV we evaluate whether investors’ valuation of losses as a function
of the persistence of losses and in Section V we study the changing properties and valuation over
the sample period. The final section summarizes and concludes.
5
II.
The Prevalence and Duration of Losses
We collect our sample of firm-year observations from Compustat’s Industrial and
Research Annual Data Bases for the years 1971-2000. Consistent with Hayn (1995) we define
our earnings variable as income (loss) before extraordinary items and discontinued operations or
IB (annual Compustat data item #18).
Our initial sample contains 217,085 firm-year
observations. Table 1 presents descriptive statistics on the prevalence of losses in our sample.
We report statistics for both our main earnings variable of interest, IB, and for bottom-line
earnings or net income NI (annual Compustat data item #172). Panel A shows the sample
contains 29.63% loss observations.
Similar to Table 1 in Hayn (1995) and to patterns in Givoly and Hayn (2000) and Klein
and Marquardt (2002), we find the number of loss observations, i.e., firms with negative amounts
for IB, increases over time. From 1971 to 1980 loss observations represent less than 20% of the
sample in each year. From 1981 to 1984 loss firms increase their share of the sample from 21.4
to 28.7%. From 1985 through 2000 loss firms constitute more than 30% of all observations, and
more than 40% in 1998 and 2000. We observe a similar pattern when we define a loss as a
negative NI observation.
In Panel B of Table 1, we document the distribution of the number of years with losses
based on a sample of firms with at least seven years of observations.2 Focusing on IB, panel B
shows only 27.21% of the firms in our sample never incur a loss over the period studied. By
contrast, about 10% of firms incur 10 or more losses over this 30-year period. Similar to panel
2
The seven-year criterion allows us to study loss history over a longer window for a subset of firms in a later
analysis. To mitigate possible effects of survivorship bias we code a firm as non-reversing if it is dropped from the
Compustat Annual File due to bankruptcy or liquidation but still appears in the Research File.
6
B, panel C shows the distribution of the number of years with losses, but now based on a sample
of 885 firms with 30 years of observations (i.e., the complete sample period). We find about
one-third of firms never incur a loss during the sample period. However, more than 7.0% of this
sample has 10 or more losses over the entire period. The results in panels B and C are again
similar for NI.
Panels B and C of Table 1 show losses can persist for a considerable time. Therefore, as
a prelude to our detailed focus on loss reversals in the following section, we show in Table 2
how firms’ return to profitability varies as a function of their recent loss history. In panel A, we
document how reversal one year into the future varies as a function of the length of the past
sequence of losses. Of firms experiencing a first loss during the sample (i.e., the loss sequence is
1 year) 45.47% are profitable the next year.
The percentage decreases drastically and
monotonically as a function of the past history of losses. For firms suffering two consecutive
losses, the probability of reversal decreases to 34.76%. The reversal probability monotonically
decreases to 27.55% for firms with 5 sequential losses.
Panel B documents how reversal over a five-year horizon varies as a function of the past
sequence of losses. The analysis reduces the number of observations as it imposes substantial
restrictions on our dataset, requiring 10 consecutive observations for each loss firm (the current
year observation, 4 past observations and 5 future observations). We find for the 6,983 firm-year
observations where the current loss is the first in a (potential) sequence, 46.79% of observations
return to profitability the next year, and 11.6% do not reverse within 5 years. For losses that do
not immediately reverse, the conditional probability of reversing in subsequent years declines
monotonically, from 36.77% after two losses to 31.88% after 5 years. Each column of the table
shows a pattern similar to the rows.
The relative magnitude of the reversal percentages,
7
however, varies considerably as a function of the length of the loss sequence of the firm across
the columns of the table. For example, in the last column, comprised of 621 firms where the
current loss is the fifth in the sequence, less than a third reverse the following year and about a
quarter of the observations do not reverse at all over the 5-year horizon.
Taken together, the descriptive evidence in Table 2 suggests loss reversals follow a
distinct pattern conditional on the number of prior losses. The longer the loss sequence, the
lower the ex ante probability the current loss will eventually reverse, presenting particular
challenges for fundamental analysis and/or valuation of the firm.
III.
Loss Reversals
The increased frequency of losses challenges investors to consider information other than
aggregate accounting earnings when measuring the earning power of the firm’s assets to value
the firm. We hypothesize investors will assess if the firm’s loss will persist or not to value a firm
reporting a loss.3 The descriptive results in Tables 1 and 2 show losses can persist for a number
of years, i.e., loss reversals do not always take place in the immediate future. Note however the
results in panel B of Table 2 show that regardless of the number of losses a firm experiences, the
unconditional probability of reversal is always highest in the next year. In our research design,
we therefore focus on loss reversal one year into the future and estimate a model of loss reversal
based on factors related to the firm’s business environment and operations as follows:
yt+1 = Xt β + εt+1
(1)
3
Our approach is similar to Chambers (1996, p.9), who argues “investors estimate initial-loss persistence using
information available at the time of the initial loss”. However, his definition of persistence is different from ours,
making the results not directly comparable. While we define persistence in reference to the probability of loss
reversal, Chambers defines persistence as “the ratio of ex-post observed present value of total loss-period negative
earnings scaled by initial-year losses” (Chambers 1996, p. 11). Using his definition, firms that reverse after a
different number of years reporting losses potentially exhibit similar persistence.
8
where yt+1 is an indicator variable equal to one if the firm becomes profitable in the subsequent
period, and zero otherwise, Xt represents the information variables of the model, and εt+1 is an
error term.
In the absence of a structural model of loss reversals, we include four different types of
information variables in our model based in part on previous literature. The first set of variables
measure the incidence and the relative magnitude of past losses. Specifically, FIRSTLOSS is an
indicator variable equal to one if the current year's loss is the first in a sequence (i.e., the firm
was profitable the prior year) and zero otherwise; NUMLOSS is an indicator variable equal to
one if the firm incurred more than two losses during the previous five years and zero otherwise;
and finally, MAGNLOSS is an indicator variable equal to one if the sum of the current loss and
the past three earnings numbers is negative and zero otherwise. Based on the patterns in Table 2,
we expect the coefficient on FIRSTLOSS to be positive: if the current loss is the first in a
sequence the probability of loss reversal is higher relative to other loss firms. Similarly, we
expect the coefficient on NUMLOSS to be negative: the more losses the firm incurs the smaller
the probability the loss will reverse in the next period. Finally, we predict a negative coefficient
on MAGNLOSS since MAGNLOSS is one if the current loss is larger than the cumulative
amount of income of the past three years. When MAGNLOSS is one, the firm has experienced a
particularly large decline in current earnings, and potentially experiences a great difficulty in
returning to profitability.
Our second set of variables captures the financial profile of the firm. First, we expect a
positive coefficient on SIZE, measured as the log of current market value (annual Compustat
data item # 199 * annual Compustat data item # 25), consistent with large firms being financially
stronger than small firms and able to return to profitability more easily. We measure the second
9
variable, return-on-assets (ROA) as income before extra-ordinary items (annual Compustat data
item # 18) scaled by lagged total assets (annual Compustat data item # 6). Since most firms in
our sample have negative ROA in the current year, we predict a positive sign for ROA indicating
firms with less negative ROAs will be more likely to return to profitability. Next we define
NEGCEQ, an indicator variable equal to one if the firm has negative equity (annual Compustat
data item # 60) and zero otherwise, to capture cumulative profitability.
We interpret the
occurrence of negative equity as indicating the profitability problems of the firm are substantial,
and predict a negative coefficient on the variable. Finally, we include recent growth in sales,
SALESGROWTH, measured as the percentage growth in sales (annual Compustat data item #
12) during the current year. Although we expect sales growth to signal a pending return to
profitability, the effect of sales growth on the probability of loss reversal is weakened if high
sales growth identifies relatively young firms in the sample that have not yet achieved
profitability. Young firms can remain unprofitable for a number of years during the early stages
of their life so that sales growth will not be a good predictor of loss reversals.4
We expect long-term accruals will also influence loss reversal. For example, for firms
with acquisitions accounted for as purchases, goodwill amortization likely influences earnings
downward for a number of years. We therefore include a profitability measure in the model
unaffected by these accruals, namely EBITDA. We measure our variable EBITDA as operating
income before depreciation (annual Compustat data item # 13), scaled by sales (annual
Compustat data item # 12). The predictive power of EBITDA for loss reversals depends on
whether the current loss follows from real operational problems or from accounting choices. We
4
Note we require each observation in the sample to have a history of five years of data. As a result, our sample does
not include recent IPOs.
10
predict higher (or less negative) values of EBITDA will be associated with a higher probability
of loss reversal.5
A third category of explanatory variables measures the dividend paying behavior of the
firm since prior research relates dividend paying behavior to future earnings of the firm. Healy
and Palepu (1988) for example show management signals profitability changes through dividend
changes.
With respect to underperforming firms, DeAngelo and DeAngelo (1990) show
dividend changes relate to persistent losses while DeAngelo et al. (1992) show information about
dividend reductions increases the ability of current earnings to predict future earnings. Recently,
Skinner (2003) relates management’s dividend signals to the persistence of losses. Based on
previous findings, we include two indicator variables to capture the dividend behavior of the
firm. First, we define DIVDUM to be equal to one if the firm is paying dividends (annual
Compustat data item # 21) and zero otherwise. We predict a positive coefficient on DIVDUM,
consistent with firms continuing to pay dividends while incurring losses signaling their
expectation the loss sequence will be brief. Second, we include DIVSTOP, an indicator variable
equal to one if a firm stops paying dividends in the current year and zero otherwise. Based on
the results of Healy and Palepu (1986) and DeAngelo et al. (1992), we predict the coefficient on
DIVSTOP will be negative.
Finally, we include two variables to control for macroeconomic conditions, based on the
work by Klein and Marquardt (2002) who study the determinants of the increased frequency of
firms reporting losses. They examine the time-series relation between the frequency of loss
firms and accounting conservatism, biases in Compustat sampling methodology, and economic
5
We also estimated our model with both a cash flow and an accruals variable included instead of EBITDA. The
results remain qualitatively unchanged.
11
conditions. While they find all three factors help to explain the increase in reported losses,
business cycle effects appear particularly important. Accordingly, we include two variables to
capture the macroeconomic conditions affecting firms: 1) GDP_GROWTH, or the annual growth
rate in real Gross Domestic Product (GDP) to control for economic growth in the year of the
loss; 2) FGDP_GROWTH, the year-ahead growth in GDP to capture economic conditions during
the year the firm’s loss potentially reverses.6 We do not predict a sign for GDP_GROWTH since
our sample includes loss observations during the entire sample period, but expect
FGDP_GROWTH to relate positively to the likelihood of loss reversal as firms benefit from
improved economic conditions.
Table 3 presents descriptive results for the variables included in the logistic regression
model (1). Panel A shows descriptive statistics for the indicator variables in the model. We
observe the occurrence of a first loss is significantly associated with loss reversal: when the
current loss is the first in a sequence (i.e., FIRSTLOSS = 1) 45.13% of losses reverse compared
to 25.64% when the current loss occurs after a previous loss. Focusing on NUMLOSS, we see
the probability of loss reversal is significantly smaller (23.06% vs. 41.98%) if the firm has
experienced more than two losses in the last five years (i.e., NUMLOSS = 1). Finally, if the sum
of the past three years of earnings is less than the current loss (i.e., MAGNLOSS = 1) the
probability of reversal is also smaller than if the sum is larger (27.11% vs. 50.12%). All χ2statistics indicate the loss variables significantly relate to the loss reversals.
In addition to the loss history variables, Panel A shows the indicator variable for negative
equity NEGCEQ also statistically relates to the probability of loss reversal. Current loss firms
with negative equity become profitable in only 21.31% of the cases, compared to 34.95% for
6
We obtain GDP data from the Bureau of Economic Analysis website (http://bea.gov/).
12
positive equity firms. Finally, Panel A shows DIVDUM, a binary variable equal to 1 if the firm
pays a dividend, significantly relates to the probability of loss reversal. Consistent with our
expectation, the probability of loss reversal for a firm paying dividends is 53.16% compared to
29.34% for firms not paying dividends. We also find a relatively small number of sample firms
eliminate their dividends when they incur losses (860 out of 19,454 or 4.42%), with no
statistically significant difference in the probability of reversal between them and other firms.7
Panel B shows descriptive statistics for the continuous variables in model (1). We present
statistics for the full sample of observation and for two sub samples based on the observed loss
reversal. Focusing on the means, observations with losses that reverse are generally larger, have
less negative ROA and less negative EBITDA, than observations that do not reverse.
In
addition, the medians of each group show observations with losses that reverse have less
negative sales growth.
Finally, the macro-economic variables are not directly related to
subsequent loss reversal.
Table 4 reports the coefficients and associated t-statistics computed using the FamaMacBeth procedure (1973). The coefficients on all three loss history variables are statistically
significant and of the predicted sign. The coefficients on FIRSTLOSS and NUMLOSS are
positive and negative, respectively, suggesting first-time losses are more transitory while losses
of firms experiencing more than two losses in the last 5 years more likely persist.
The
coefficient on MAGNLOSS is also negative and significant, indicating large losses less likely
reverse. Continuing with the financial profile variables, we obtain positive and statistically
significant coefficients on SIZE, ROA, and EBITDA indicating losses of larger firms more likely
7
The percentage is less than the 15 percent DeAngelo et al. (1992) report; however, they condition their sample on
identifying dividend paying firms first, a restriction we do not impose.
13
reverse, as are losses of firms with (relatively) smaller losses measured through ROA or through
EBITDA. The result for EBITDA suggests accruals such as goodwill amortization, taken out of
of EBITDA, potentially lead to persistent loss patterns.
The coefficient on NEGCEQ is
significantly negative, suggesting losses of firms with negative equity less likely reverse.
Finally, the coefficient on SALESGROWTH is not significant.
Both dividend history coefficients are statistically significant.
The positive and
significant coefficient on DIVDUM indicates losses of firms that continue to pay dividends more
likely reverse than losses of firms that do not pay dividends (see also Skinner 2003). In addition,
the coefficient on DIVSTOP is negative and statistically significant in a one-tailed test (t=-1.873)
consistent with the finding in previous literature that stoppage of dividend payments signals the
firm loss will not reverse in the immediate future.
Finally, the coefficient on contemporaneous GDP growth is negative and significant
(one-tailed test t=-1.769), whereas one-year ahead GDP growth does not relate to loss reversals.
The relatively weak results for the GDP growth variables suggest accounting or dividend
information, even if purely historical, is more important than knowledge of short-run economic
performance to predict future performance. The results also indicate idiosyncratic characteristics
of the firm relate more strongly to loss reversals than the conditions of the economy as a whole.8
Summarizing, the results of the logistic regression model in Table 4 are consistent with both
earnings and non-earnings information variables being useful in predicting the loss firm’s return
to profitability.
8
Although our results seemingly contrast with the findings in Klein and Marquardt (2002), our results are not
directly comparable, since Klein and Marquardt (2002) address a different question, namely what causes the
increase in the cross-section of the number of firms annually reporting losses.
14
IV.
The Valuation of Losses: Earnings Response Coefficients
Having established a parsimonious model can assess the persistence of a firm’s loss, we
address the question of whether investors differentially price losses as a function of their
expected persistence. Our analysis relates to earlier research by Hayn (1995), who considers the
role of the abandonment option for the value-implications of losses. In her analysis, Hayn
(1995) explicitly uses two proxies for the future prospects of the firm (bond ratings and a
liquidation value) to test how the option affects valuations and concludes the market reaction to a
loss is greater when liquidation of the firm is less likely. We argue our model of loss reversal
captures a particular way for investors to structure financial information to assess the persistence
of a loss, or alternatively the likelihood of abandoning their investment in the firm.9 Therefore,
we extend Hayn (1995) by assessing if investors price losses as a function of their expected
probability of loss reversal. To implement our test, we sort the loss observations annually into
quartiles based on their estimated probability of reversal. Next, we define persistent (transitory)
losses as those with probabilities in the first (fourth) quartile of the distribution: persistent losses
are therefore those least likely to reverse, and transitory losses those most likely to reverse.10
To explore the pricing of losses, we estimate ERCs using the following regression (see
also Hayn 1995):
Rett = α + β IBt + εt
(2)
9
A number of other studies, such as Burgstahler and Dichev (1997), Collins et al. (1997) and Collins et al. (1999),
also study firm valuation in the presence of losses and use book value as a proxy for the abandonment value of the
firm.
10
In contrast to Chambers (1996) who relies on perfect foresight of both the number and magnitude of incurred
losses to establish the persistence of first-time losses, we estimate the ex ante persistence of a loss. We also consider
a more general setting than his by studying the pricing of losses irrespective of whether the loss is a first-time loss or
a recurring loss.
15
where Rett is the return over the 12-month period commencing with the fourth month of fiscal
year t, IBt is the earnings per share variable in year t (annual Compustat data item #18 scaled by
annual Compustat data item #25) scaled by Pt-1 or share price (annual Compustat data item #199)
at the end of year t-1, εt is the error term. Consistent with our annual estimation of the loss
reversal model, we estimate equation (2) in each year of the sample period and assess the
significance of the ERCs using the Fama-Macbeth procedure (1973).
Table 5 presents the results of the analysis. The descriptive statistics in Panel A show the
mean (median) probability of reversal of a firm is 34.3% (32.1%) for the entire sample.
Naturally, we find the mean (median) of persistent losses to be much lower than the
corresponding values for transitory losses: 17.4% (17.8%) versus 54.4% (53.5%).11 Means and
medians of the probability of reversal are of the same order of magnitude in both groups,
indicating a symmetric distribution of the probabilities. The descriptive statistics for IB show all
means and medians are negative (by construction), with the overall sample mean (median) equal
to –0.236 (-0.122).
The panel further shows the probability of reversal ranks the relative
magnitude of IB, with both mean and median IB being far more negative in the persistent loss
sub-sample. Note also means are more negative than their corresponding medians in all samples,
suggesting the presence of influential observations in each group.
Lastly, Panel A shows returns follow a pattern different from IB across the sub-samples:
mean annual returns surprisingly are positive in the entire sample and the persistent sub-sample
(0.042 and 0.081), but negative in the transitory sub-sample (-0.040). However, median returns
of persistent loss firms are more negative than median returns of transitory losses, a pattern
11
By construction, the mean and median probability of reversal is higher for the transitory group. While we do not
report them, we find the underlying quartile classification monotonically ranks all variables we use in the ERC
estimation, as well as other variables we examine in later analyses.
16
distinct from the pattern of mean returns and evidence that extreme observations influence the
return distributions.
To avoid the potential influence of the extreme observations for IB and Ret in the sample,
we estimate equation (2) in ranks in our main analyses and re-estimate the regressions using the
raw data in sensitivity tests. Panel B contains the results of the ERC estimation in the full
sample and in the two sub-samples. We find a positive and statistically significant ERC for the
full sample, implying investors price losses as a function of their magnitude. We find a similar
positive ERC for transitory losses, but a statistically insignificant ERC for persistent losses. That
is, the more transitory we estimate the current loss to be, the more market returns reflect the
information in earnings.12 The finding is consistent with Hayn’s (1995) argument for finding a
more pronounced pricing effect of the loss when investors will least likely consider abandonment
of their investment or, in our case, when the loss is transitory. However, the result also present a
paradox, as it suggests the market will reduce the value of a loss firm only if the firm will soon
regain profitability, but not reduce the value of a loss firm expected to continue to generate
losses.
To resolve the paradox and better understand the pattern of ERCs, we carry out three
analyses to explore the profile of observations across sub-samples. First, we investigate how loss
persistence relates to the loss history variables we define earlier. Panel A of Table 6 shows loss
persistence exhibits a clear relation with the loss history variables. In fact, the loss history
descriptive statistics in Panels A of Tables 5 and 6 suggest transitory losses are different from
12
Using the raw data rather than the ranks of IB and Ret we find the presence of extreme IB and Ret observations
affect our estimation but the general tenor of results remains unaltered. In particular, the ERC of losses in the full
sample and for persistent losses is insignificantly different from zero, but the ERC for transitory losses remains
positive and significant.
17
persistent losses in three ways. First, as Table 5 shows, transitory losses are more likely than not
to reverse in the next year, with a mean and median probability of reversal higher than 50%.
Second, Panel A in Table 6 shows transitory losses are overwhelmingly first-loss firms
(83.51%), with fewer than 6% of the observations experiencing more than two losses during the
preceding 5 years. Third, transitory losses are smaller than persistent losses (the proportion of
transitory observations with MAGNLOSS=1 is 23.10% compared to 97.75% for persistent
observations). In sum, we find transitory losses are relatively small, first-time, and one-time
losses. The distinction between the characteristics of transitory and persistent losses, although
suggesting investors should price them differently, however does not provide a clear intuition for
the observed ERCs in panel B of Table 5.
We therefore continue with a second analysis in Panel B of Table 6 focusing on the cash
flow and accrual components of losses. We base our analysis on findings by Givoly and Hayn
(2000), who observe significant changes in the cash flow and accrual components of earnings
over the last decades. Whereas they document a large decline in firm profitability in the crosssection of firms caused by the increase in large negative accruals over time, we extend their
research by relating the composition of losses into cash flows and accruals to investors’
differential pricing of transitory and persistent losses.
We define CFO as cash flow from operations scaled by sales (annual Compustat data
item # 12). Consistent with previous literature (Hayn 1995), we measure cash flow from
operations as net income (annual Compustat data item # 172) – accruals. We measure accruals or
ACC as (∆Current Assets (data item #4) - ∆Cash (data item #1) - ∆Current Liabilities (data item
#5) + ∆Debt in Current Liabilities (data item #34) + Depreciation and Amortizations (data item
18
#14), scaled by sales (annual Compustat data item # 12). Panel B shows that in the full sample
mean (median) cash flows are negative (positive) whereas both mean and median accruals are
negative. Focusing on the sub-samples, we find a correspondence between the persistence of
losses and the relative magnitude of their cash flow and accrual components. Persistent losses
exhibit extreme negative values for both components relative to transitory losses. In addition, in
contrast to the corresponding statistics for persistent losses, both mean and median cash flow of
transitory losses are positive, implying negative accruals only drive the negative earnings.
Based on the pattern of the cash flow and accruals components of losses across subsamples, we interpret our significant ERC result for transitory losses to be consistent with a
manifestation of accounting conservatism. Basu (1997) defines accounting conservatism as the
asymmetric timeliness of earnings, with earnings reflecting bad news earlier than good news. He
shows conservatism manifests itself through negative accruals, leading to bad news earnings (in
our case losses) being less persistent than good news earnings. Therefore, the pattern of negative
returns and a positive and significant ERC for transitory losses caused by negative accruals
corresponds with Basu’s definition of conservatism.
By contrast, accounting conservatism cannot explain the ERC results for persistent losses
since the relative magnitudes of cash flows and accruals for persistent losses indicate primarily
negative cash flows cause them. Therefore, in a third analysis we study the role of R&D
expenditures and Special Items as determinants of losses and their ERCs across the sub-samples.
We focus on investments in intangibles, and on R&D in particular, since recent research finds
the magnitude of R&D expenditures increases significantly over the past decades, thereby
influencing the properties of reported accounting measures (e.g., Amir and Lev 1996, Collins et
al. 1997, Lev and Zarowin 1999). As US accounting rules require managers to expense R&D
19
outlays when they occur, R&D potentially drives the negative cash flow components in Panel B.
Additionally, since managers often make R&D expenditures as part of an ongoing policy, the
magnitude and occurrence of R&D outlays potentially determine not only the (negative) level of
earnings in the current year, but also the probability of loss reversal.
Other recent studies prompt us to examine the influence of Special Items. Givoly and
Hayn (2000, p. 305) discuss how Special Items related to restructurings and write-offs typically
lower reported earnings through negative accruals. Skinner (2003) singles out the presence of
Special Items in losses as a signal of their persistence: he argues Special Items reflect managers’
accounting discretion more than other components of losses and therefore are more likely to
generate transitory losses. Similarly, earlier work by Dechow (1994) shows Special Items have
only a temporary effect on earnings and reduce the short-term ability of earnings to measure
performance. Finally, Carter (2000) and Burgstahler et al. (2002) both conclude negative Special
Items represent ‘inter-period transfers’ that lead to increased earnings in subsequent periods,
with Carter (2000) showing the post-restructuring performance of firms is greater than the
upward bias caused by the accelerated recognition of Special Item restructuring expenses.
Panel C of Table 6 provides our descriptive statistics on the R&D and Special Items
components of losses. We define R&D and Special Items or SPI as annual Compustat data item
#46 and data item #17, respectively, scaled by sales (annual Compustat data item # 12). We find
mean R&D is 1.145 in the full sample, with the median R&D expenditure being zero. Consistent
with the evidence in Givoly and Hayn (2000) and Skinner (2003), we find average Special Items
to be negative in the full sample, but again with a zero median. Continuing, we observe
significant differences in the relative magnitude of both R&D and SPI across the sub-samples.
The means of both R&D and SPI for persistent losses are several orders of magnitude larger than
20
the values for transitory losses.13 Comparing the two groups, persistent loss observations have
more R&D outlays, consistent with the finding in Panel B that cash flows on average are more
negative, and report more negative Special Items, consistent with their accruals being more
negative as well. Also, measured as a percentage of sales, mean Special Items are much smaller
than R&D expenditures. In addition, the results for the medians indicate more than half of
persistent losses contain an R&D component. Similarly, more than half of the transitory losses
contain a negative SPI component, indicating that, although SPI are on average smaller for
transitory losses relative to persistent losses, they appear much more often in transitory losses.
Summarizing, Panels B and C of Table 6 highlight the magnitude of the cash flow
component and in particular R&D outlays for persistent losses. To investigate if the incidence of
R&D expenditures across the two groups explains the difference between the ERCs of persistent
and transitory losses in Panel B of Table 5, we next re-estimate rank regression (2) but include
two independent variables: 1) the loss variable without its R&D component (IB+R&D), and 2)
the R&D component of the loss:
Rett = α + β (IBt+R&Dt )+ γ (R&Dt )+
εt
(3)
Panel D of Table 6 shows investors price the loss variable without its R&D component
(β=0.065, t-statistic=3.537) similar to what we document earlier for the full sample in Table 5;
however, they do not price the R&D component itself (γ=0.008, t-statistic=0.309). We find,
similar to the pattern of ERCs in Panel B of Table 5, the coefficients on (IB+R&D) increase as
we move from the persistent to the transitory sub-sample, with the ERC in the persistent sub-
13
Extreme observations in the persistent sub-sample influence the large differences in the means reported in Table
6. However, even after excluding the top 1 percent of reported R&D amounts, mean R&D of persistent losses is
more than 50 times greater than the mean of transitory losses. Similar results apply to all of the comparisons of
means in Table 6 and reinforce our decision to use rank regressions for our primary results.
21
sample being insignificant. However, the panel also documents investors price R&D differently
for persistent versus transitory loss observations: the coefficient on R&D for persistent losses is
positive and significant, indicating the more a firm with a persistent loss spends on R&D, the
larger the market returns are. By contrast, investors do not price the R&D component of
transitory losses, potentially because the majority of observations in the sub-sample do not report
R&D (see Panel D) and the average reported amounts are relatively small.
The presence and pricing of R&D therefore partially explain why investors do not punish
firms with persistent losses by reducing their market value, or alternatively, why we find
insignificant ERCs for persistent losses. In our sample, the most persistent losses relate to
negative cash flow effects, particularly R&D expenditures, that investors appear to reward; they
also do not consider the other components of persistent losses when valuing the firm. In sum,
our evidence suggests investors, when confronted with an apparently persistent loss, look beyond
aggregate earnings to value the firm.
Further, since the R&D component is relatively
unimportant for transitory losses, it does not affect their valuation.
In a second unreported decomposition analysis we also separate out the Special Items
component of the losses and find the results in Panel D of Table 6 do not change qualitatively.
Similar to our findings for R&D, we find investors price Special Items in persistent losses
positively, whereas they do not price Special Items in the sub-sample with transitory losses. The
composition of Special Items across both groups varies considerably though: 58% of Special
Items in the transitory loss sample are negative versus 37% for persistent losses. Also, more than
50% of persistent losses contain no Special Items. The result that investors do not price Special
Items in transitory losses where they are predominantly negative is consistent with findings in
Burgstahler et al. (2002) that negative Special Items are predominantly transitory and not priced.
22
Taken together though, the differential pattern of pricing of Special Items in persistent and
transitory losses does not explain the ERC pattern across the sub-samples, also because the R&D
component in persistent losses dwarfs the Special Items component.
Our results for both
components though emphasizes investors look beyond aggregate losses to value the firm. In
particular, the results for persistent losses mirror those for transitory losses: in the case of
transitory losses investors do not price either R&D and Special Items but value the remaining
components of losses whereas the opposite pattern emerges for persistent losses.14
V.
The Valuation of Losses: Time-Series Evidence
Our results in Table 1 show the frequency of loss observations increases substantially
over the sample period, consistent with findings by other researchers (e.g., Hayn 1995, Klein and
Marquardt 2002). Similarly, several recent studies illustrate the properties and valuation of
earnings change in recent decades (e.g., Collins et al. 1997, Givoly and Hayn 2000). To the
extent the changing nature of earnings influences either the increased occurrence or valuation of
losses, our results in Tables 4 through 6 potentially hide time-series trends. We therefore carry
out two descriptive analyses to explore if the increase in loss observations in the annual cross-
14
In sensitivity analyses, we re-estimate regression (3) using the raw data instead of the ranks. The sensitivity
analyses generally confirm our main results, but extreme observations influence the results. In particular, we find
that in the full sample investors do not price (IB+R&D), similar to earlier findings using raw data (see footnote 8),
but they do price the R&D component of losses. Focusing on the sub-samples, we find the coefficient on
(IB+R&D) is larger for transitory losses, whereas the coefficient on R&D is positive and statistically
indistinguishable across sub-samples. However, given their average magnitude, R&D expenditures only influence
the pricing of persistent losses. We also estimate the specification of the regression separating out R&D and Special
Items and find again the presence of Special Items does not influence the results for the R&D component. In fact,
using the raw data, we find no significant coefficients on the Special Items component in either group. Therefore,
although our findings for R&D and Special Items lead to the same conclusion irrespective of whether we use ranks
or raw data, they also indicate the estimation of parameters is highly influenced by extreme observations for both
variables.
23
sections of the sample period relates to a change in the nature of losses and investors’ valuation
of losses.
First, we document the time-trend in the probability of loss reversal and ERCs. In Table
7, we report the coefficients of regressions of the annual median loss reversal probabilities and
the annual estimate of ERCs on year-indicators. The results in Panel A show the median annual
loss reversal probability decreases over the sample period in the full sample and in both subsamples, with a larger decrease for persistent losses. To explain the decrease, we regress in
unreported analyses the annual coefficients in the logistic model on year-indicators and find the
magnitude of most coefficients remains unaltered over the sample period, with three exceptions.
First, the coefficient on NUMLOSS decreases over time suggesting a general increase in
persistence over the sample period. Second, the coefficient on EBITDA decreases, consistent
with a reduced effect of EBITDA improvements on loss reversals and an increase in “cosmetic”
losses over the sample period, perhaps caused by an increase of long-term accruals, such as
goodwill amortization.
Third, the coefficient on DIVDUM increases, consistent with the
dividend paying signal becoming more important over the sample period (see Skinner 2003).
Panel B in Table 7 shows how investors change their valuation of losses over the sample
period. In contrast to the ERCs in the full sample and the transitory subgroup, that show no
time-trend, the ERCs for persistent losses decline over the sample period. Consistent with the
time-trend evidence, unreported analysis shows ERCs for persistent losses in the first 10 years of
the sample period are not significantly different from zero, but become significantly negative in
the last 10 years; by contrast, we do not find a similar change over the sample period for
transitory losses.
24
Summarizing, the results in Table 7 indicate the nature and valuation of losses changes
over the sample period: losses, especially those least likely to reverse, become more persistent
over time. Simultaneously, we observe a growing divergence between returns and earnings for
persistent losses, leading investors to price losses reliably negative in the later years of the
sample period. The time-series results therefore qualify our earlier findings: averaging annual
ERCs over the sample period hides a time-series trend in the valuation of persistent losses. By
contrast, investors price transitory losses consistently over the entire sample period.
Continuing our earlier analysis, we investigate how the change in the nature and
valuation of losses relate to specific loss components. Panel A of Table 8 reports the coefficients
of median annual cash flow and accruals components on year-indicators.
The significant
negative time-series coefficient in the full sample (-0.003, t-statistic=-7.781) indicates the cash
flow component of losses becomes more negative over the sample period. Examining CFO by
sub-sample, we observe the change in the full sample follows entirely from the change observed
for persistent losses since the coefficient for transitory loses is statistically insignificant. For
accruals, the time-series coefficients are significantly negative in the full sample and in each subsample, consistent with accruals overall becoming more negative over the sample period. The
effect appears more pronounced for persistent losses though.
Panel B of Table 8 shows the results of a similar time-series analysis focusing on the
R&D and Special Items components of losses. For R&D the coefficient for the full sample, and
for persistent loss observations, is positive and significant, in contrast to a significant negative,
and much smaller, coefficient for transitory losses. The result suggests R&D expeditures have
become a more (less) important component of persistent (transitory) losses. We also observe a
25
larger presence over time of a negative SPI component of losses in the full sample, as a result of
their increased presence in transitory losses.
Overall, Table 8 shows investors change the pricing of persistent losses over the sample
period. Consistent with the change in their pricing, we observe the nature of persistent losses
changes over the sample period, with the negative cash flow component in general and the R&D
component in particular increasing significantly over time. Although the nature of transitory
losses also changes, the effect of the change is not as pronounced since we observe no change in
investors’ pricing of transitory losses over the sample period. Our time-series results therefore
corroborate and extend our earlier findings on the pricing of persistent and transitory losses. The
time-series results confirm the distinction between the nature and valuation of losses as a
function of their persistence or probability of reversal. In addition, the time-series results
emphasize the importance of the changing role of the negative cash flow and R&D expenditure
components in losses: they determine not only the changing time-series properties of losses, but
also investors’ changing valuation of persistent losses over the sample period.
V.
Concluding Remarks and Future Research
We study how financial statement users address the increasingly important challenge of
finding (accounting) proxies other than earnings to evaluate for the earning power of assets to
value firms reporting losses. Referring to the abandonment option literature, we investigate if
investors use information to assess the likelihood the firm’s loss will persist and price earnings
consistent with the loss’ expected persistence.
We argue investors structure financial
information to assess the persistence of a loss, or alternatively the likelihood of abandoning their
investment in the firm.
26
To evaluate our claim, we first estimate a parsimonious model of loss reversal one-year
into the future and find evidence consistent with both earnings and non-earnings information
variables being useful in predicting a loss firm’s return to profitability. Using the estimated
probabilities of loss reversal to define persistent and transitory losses, we next show investors
price transitory and persistent losses differently. In particular, investors price transitory losses
consistent with the prediction based on Basu’s definition of accounting conservatism: transitory
losses, determined by negative accruals, reflect bad news in a timely fashion and have positive
and significant ERCs.
By contrast, accounting conservatism does not explain the pricing of persistent losses.
Examining the properties of persistent losses we find they largely consist of negative cash flows
in general and R&D expenditures in particular. We re-examine the pricing of persistent losses
by separating out the R&D component and find investors price R&D expenditures positively. In
other words, although investors do not value the non-R&D component of persistent losses, they
reward firms reporting persistent losses with positive returns for their R&D expenditures. We
find investors also price Special Items differently in the persistent and transitory loss subsamples, but the result does not explain the difference in ERCs across both sub-samples since the
nature of Special Items is different for both sets of losses: transitory losses contain
predominantly negative Special Items, whereas persistent losses mostly contain no Special Items.
Taken together though, our findings underline that, when confronted with an apparently
persistent loss, investors consider the distinct components of the loss to value the firm. We
corroborate our findings on the pricing of persistent and transitory losses with time-series results
that emphasize the changing role of the negative cash flow and R&D expenditure components in
losses: they determine not only the changing time-series properties of losses, but also investors’
27
changing valuation of persistent losses over the sample period.
We contribute to the literature by documenting the valuation implications of losses as a
function of their persistence, measured through a model of loss reversal. We show accounting
conservatism, as per Basu (1997), while explaining the pricing of transitory losses, does not
explain the pricing of apparently persistent losses. We relate the pricing difference of transitory
and persistent losses to the distinct effects of accruals and cash flows on the time-series
properties of losses. Our findings extend Givoly and Hayn (2000), who conclude the decline in
profitability over time is not associated with a contemporaneous decline in cash flows. We show
though that cash flows are a significant and growing determinant for a substantial subset of loss
firms, namely those with persistent losses. By illustrating how structural changes in the nature of
business operations, i.e., an increase in R&D expenditures over time, affect the properties and
pricing of losses, our findings emphasize the role of financial statement analysis to assess the
exact nature of losses to establish their valuation effects. For example, a closer look at the
components of losses explains why the proportion of firms reporting persistent losses in the
cross-section increased so dramatically over the last three decades at a time when the stock
market rose to unprecedented heights.
28
References
Altman, E. (1968) “Financial ratios, discriminant analysis and the prediction of corporate
bankruptcy” Journal of Finance 22: 589-610
Amir, E., and B. Lev (1996) “Value-relevance of nonfinancial information: The wireless
communications industry.” Journal of Accounting and Economics 22: 3-30.
Basu, S. (1997) “The conservatism principle and the asymmetric timeliness of earnings.” Journal
of Accounting and Economics 24: 3-37.
Berger, P.G., E. Ofek, and I. Swary. (1996) “Investor valuation of the abandonment option.”
Journal of Financial Economics 42: 257-287.
Burgstahler, D., and I. Dichev. (1997) “Earnings, adaptation, and firm value.” The Accounting
Review, April: 187-215.
Burgstahler, D., J. Jiambalvo, and T. Shevlin. (2002) "Do stock prices fully reflect the
implications of special items for future earnings? Journal of Accounting Research, 40:
585 – 612.
Carter, M.E. (2000) “Operating Performance Following Corporate Restructurings,” working
paper, Columbia University.
Chambers, D.J., (1996) “The information content of negative earnings and its relation with
initial-loss persistence,” working paper, University of Illinois.
Collins, D.W., E.L. Maydew, and I.S. Weiss. (1997). Changes in the value relevance of earnings
and book values over the past forty years. Journal of Accounting and Economics 24: 3967.
Collins, D. W., M. Pincus, and H. Xie. (1999) “Equity valuation and negative earnings: The role
of book value of equity.” The Accounting Review, April: 29-61.
29
DeAngelo, H., and L. DeAngelo. (1990) “Dividend policy and financial distress: an empirical
investigation of troubled NYSE firms,” Journal of Finance 45: 5 1415-1431.
DeAngelo, H., L. DeAngelo, and D. Skinner. (1992) “Dividends and losses” Journal of Finance
47: 5 1837 – 1863.
Dechow, P. M., (1994) “Accounting earnings and cash flows as measures of firm performance:
The role of accounting accruals.” Journal of Accounting and Economics, 18: 3-42.
Fama, E., and J. MacBeth. (1973) “Risk, return, and equilibrium: empirical tests.” Journal of
Political Economy, 81: 607-636.
Givoly, D. and C. Hayn, (2000) “The changing time-series properties of earnings, cash flows,
and accruals:
Has financial reporting become more conservative?” Journal of
Accounting and Economics 29: 287 – 320.
Hayn, C. (1995) “The information content of losses.” Journal of Accounting and Economics 20:
125-153.
Healy, P., and K. Palepu, (1988) “Earnings information conveyed by dividend initiations and
omissions,” Journal of Financial Economics 21: 149-175.
Klein, A., and C. Marquardt (2002) “Can economic factors explain the rise in accounting losses
over time?”, Working paper, New York University.
Lev, B., and P. Zarowin. (1999) “The boundaries of financial reporting and how to extend them.”
Journal of Accounting Research 37 (Autumn): 353-385.
Modigliani F., and M.H. Miller. (1966) “Some estimates of the cost of capital to the electric
utility industry, 1954-57.” The American Economic Review LVI, 3: 333-391.
Skinner, D. (2003) “What do dividends tell us about earnings quality?” Working paper,
University of Michigan Business School.
30
Watts, R., and J. Zimmerman (1986) Positive Accounting Theory. Prentice-Hall, Englewood
Cliffs, NJ.
Wysocki, P.D. (2001) “Real options and the informativeness of segment disclosures.” Working
Paper, MIT.
31
Table 1
Frequency of Losses
Panel A: Total Sample (all firm-years available)a
All firms
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
No. Firm-years
IB
% of Loss Firm-years
NI
% of Loss Firm-years
217,085
29.63
29.74
3,842
4,015
4,403
6,019
6,094
6,137
6,176
6,088
6,000
6,101
6,159
6,484
6,709
6,781
7,203
7,496
7,582
7,477
7,379
7,415
7,568
7,989
9,062
9,452
10,293
10,434
10,158
10,044
8,495
8,050
11.87
7.97
8.09
17.64
19.92
16.52
15.90
14.75
15.28
17.88
21.42
27.80
27.84
28.67
34.32
36.31
36.42
35.84
37.20
37.99
38.65
36.07
33.07
30.61
32.73
33.22
35.05
40.01
39.88
42.04
14.08
9.14
8.97
18.62
20.27
16.56
15.98
14.55
15.20
18.23
21.40
27.80
27.93
28.82
34.49
35.89
35.94
35.27
36.70
37.57
38.45
36.79
33.52
30.45
32.97
33.30
35.23
40.13
39.80
42.12
32
Panel B: Distribution of the number of years with losses (based on a sub-sample of firms with at
least 7 years of data)
IB
NI
No. Firms
% of Firms
No. Firms
% of Firms
Number of losses
0
1
2
3
4
5
6
7
8
9
10
10 < and < 20
20 or more
11,435
100.00
11,435
100.00
3,112
1,396
1,102
934
855
715
647
653
522
377
291
806
25
27.21
12.21
9.64
8.18
7.48
6.25
5.66
5.71
4.56
3.30
2.54
7.05
0.22
2,851
1,446
1,197
980
868
791
668
662
532
377
275
769
20
24.93
12.65
10.47
8.57
7.59
6.92
5.84
5.79
4.65
3.30
2.40
6.73
0.17
33
Panel C: Distribution of the number of years with losses (based on a sub-sample of firms with 30
years of data)
IB
NI
No. Firms
% of Firms
No. Firms
% of Firms
885
100.00
885
100.00
Number of losses
0
297
33.56
263
29.72
1
159
17.97
155
17.51
2
89
10.06
92
10.40
3
62
7.01
75
8.47
4
49
5.54
51
5.76
5
36
4.07
47
5.31
6
35
3.95
34
3.84
7
31
3.50
33
3.73
8
22
2.49
25
2.82
9
15
1.69
15
1.69
10
22
2.49
19
2.15
10 < and < 20
61
6.89
68
7.68
20 or more
7
0.79
8
0.90
a
The data are collected from Compustat’s Industrial and Research Annual Data Bases and covers
the period 1971-2000. IB is defined as income (loss) before extra-ordinary items and
discontinued operations (annual Compustat data item #18). NI is net income (annual Compustat
data item #172).
34
Table 2
Loss Reversals as a Function of the Loss History of the Firma
Panel A: Relation between length of loss sequence and reversal one year into the future
Loss Sequence
Obs.
Reversal (%)
1 year
2 years
3 years
4 years
5 years
10,234
5,055
2,968
1,787
1,118
45.47
34.76
31.17
27.98
27.55
Panel B: Loss reversal in the current loss sample as a function of the string of past lossesb
Future
Loss Sequence (number of years)
Reversal
1
2
3
4
5
Obs.
6,983
3,356
1,882
1,096
621
1 year
2 years
3 years
4 years
5 years
46.79
19.32
11.28
6.60
4.41
36.77
21.31
13.02
8.34
4.95
33.63
20.94
13.71
7.86
5.42
32.66
21.35
12.04
7.57
4.93
31.88
17.71
12.08
7.73
5.48
>5 years
11.60
15.61
18.44
21.44
25.12
The data are collected from Compustat’s Industrial and Research Annual Data Bases and covers
the period 1971-2000. Losses are based on IB, defined as income (loss) before extra-ordinary
items and discontinued operations (annual Compustat data item #18).
b
Loss sequence refers to an uninterrupted sequence of annual losses. Reversal indicates the loss
firm becomes profitable.
a
35
Table 3
Logistic Regression Model of Loss Reversal: Descriptive Statisticsa
Panel A: Indicator Variables
Value
Obs.
Reversal (%)
χ2
FIRSTLOSS
0
1
12,212
7,242
25.64
45.13
.0001
NUMLOSS
0
1
10,113
9,341
41.98
23.06
.0001
MAGNLOSS
0
1
4,886
14,568
50.12
27.11
.0001
NEGCEQ
0
1
16,521
2,933
34.95
21.31
.0001
DIVDUM
0
1
16,589
2,865
29.34
53.16
.0001
DIVSTOP
0
1
18,594
860
32.80
35.00
.1785
36
Panel B: Continuous Variablesb
Obs.
Mean
Std. Dev.
Median
SIZE
Full Sample
No Reversal
Reversal
19,454
13,055
6,399
2.945
2.763
3.314***
2.075
1.975
2.220
2.807
2.699
3.098***
ROA
Full Sample
No Reversal
Reversal
19,454
13,055
6,399
-0.300
-0.359
-0.179***
2.221
2.044
2.537
-0.092
-0.125
-0.049***
SALESGROWTH
Full Sample
No Reversal
Reversal
19,454
13,055
6,399
-0.002
-0.013
0.022
2.012
2.288
1.277
-0.032
-0.047
-0.010***
EBITDA
Full Sample
No Reversal
Reversal
19,454
13,055
6,399
-2.668
-3.813
-0.331***
54.185
65.647
11.223
-0.004
-0.035
0.031***
GDP_GROWTH
Full Sample
No Reversal
Reversal
19,454
13,055
6,399
0.032
0.032
0.031
0.019
0.019
0.020
0.037
0.038
0.037
FGDP_GROWTH
Full Sample
No Reversal
Reversal
19,454
13,055
6,399
0.033
0.033
0.034
0.018
0.018
0.019
0.038
0.038
0.039
a
The data are collected from Compustat’s Industrial and Research Annual Data Bases and covers the period 19712000. Losses are based on IB, defined as income (loss) before extra-ordinary items and discontinued operations
(annual Compustat data item #18). FIRSTLOSS is an indicator variable equal to one if the current year’s loss is the
first in a sequence (i.e., the firm was profitable last year) and zero otherwise; NUMLOSS is an indicator variable
equal to one if the firm incurred more than two losses in the past five years and zero otherwise; MAGNLOSS is an
indicator variable equal to one if the sum of the current loss and the past three earnings numbers is negative and zero
otherwise. NEGCEQ is an indicator variable equal to one if the firm has negative equity (annual Compustat data
item # 60) and zero otherwise. DIVDUM is an indicator variable equal to one if the firm is paying dividends (annual
Compustat data item # 21) and zero otherwise. DIVSTOP is an indicator variable equal to one if the firm stopped
paying dividends in the current year and zero otherwise. SIZE is log of current market value (annual Compustat data
item # 199 * annual Compustat data item # 25). ROA or return-on-assets is income before extra-ordinary items
(annual Compustat data item # 18) scaled by lagged total assets (annual Compustat data item # 6). SALESGROWTH
is percentage growth in sales (annual Compustat data item # 12) over the current year. EBITDA is operating income
before depreciation (annual Compustat data item # 13), scaled by sales (annual Compustat data item # 12).
GDP_GROWTH is the current year percent change in Gross Domestic Product from the prior year and
FGDP_GROWTH is the percent change for the subsequent year (contemporaneous with the year of the dependent
variable).
b
***, **, * indicate the difference between the No Reversal and Reversal sub-sample means or medians is
significant at the 1%, 5%, or 10% level.
37
Table 4
Logistic Regression Models of Loss Reversala
Predicted Sign
Avg. Coefficient
t-statistic
FIRSTLOSS
NUMLOSS
MAGNLOSS
+
-
0.257
-0.194
-0.366
3.862
-3.028
-5.819
SIZE
ROA
NEGCEQ
SALESGROWTH
EBITDA
+
+
?
+
0.056
0.479
-0.187
0.023
0.374
3.293
2.061
-2.633
0.550
2.920
DIVDUM
DIVSTOP
+
-
0.292
-0.129
3.834
-1.873
GDP_GROWTH
FGDP_GROWTH
?
?
-17.902
3.968
-1.769
0.362
811
8.520%
Observations
Pseudo-R2
Percent
Concordantb
68.829%
a
The data are collected from Compustat’s Industrial and Research Annual Data Bases and covers the
period 1971-2000. Losses are based on IB, defined as income (loss) before extra-ordinary items and
discontinued operations (annual Compustat data item #18). The table presents the results of the annual
estimation of logistic regressions where loss reversal is the dependent variable, i.e., a variable that takes
the value of one when the firm becomes profitable one-year into the future, and zero otherwise. For the
other variable definitions, see the notes to Table 3. The table shows the average coefficient over the
estimation period and associated t-statistic derived using the Fama-Macbeth (1973) procedure.
b
Percent Concordant indicates the within-sample percentage of observations correctly classified by the
model. Pseudo R2 measures the increase in log likelihood of a model containing all variables relative to a
model containing only the intercept. Both statistics are averaged over the number of years in the
estimation period.
38
Table 5
Earnings Response Coefficientsa
Panel A: Descriptive Statistics
Sample
Obs.
Mean
Std.
Median
Prob(Reversal)
Full Sample
Persistent
Transitory
14,399
3,608
3,589
0.343
0.174
0.544
0.159
0.083
0.098
0.321
0.178
0.535
IB
Full Sample
Persistent
Transitory
14,399
3,608
3,589
-0.236
-0.369
-0.105
0.295
0.369
0.145
-0.122
-0.227
-0.056
Return
Full Sample
Persistent
14,399
3,608
0.042
0.081
0.728
0.868
-0.140
-0.174
Transitory
3,589
-0.040
0.532
-0.141
Panel B: Annual ERC estimation: Rett = α + β IBt + εt
Sample
No. Obs.
β
Full
Persistent
Transitory
598
148
148
0.041
-0.004
0.060
a
b
t-statistic
Adj. R2
2.406
-0.114
2.825
0.011
0.022
0.016
The data are collected from Compustat’s Industrial and Research Annual Data Bases and covers the
period 1971-2000. Losses are based on IB, defined as income (loss) before extra-ordinary items and
discontinued operations (annual Compustat data item #18).
Panel A shows descriptive statistics for three variables in the full sample and in sub-samples based on the
quartiles of Prob(Reversal). Prob(Reversal) is the predicted probability from the annual logistic
regressions reported in Table 4. Persistent and Transitory refer to sub-samples based on the distribution
of Prob(Reversal): the Persistent (Transitory) sub-sample contains the loss observations in the first
(fourth) quartile of the distribution of Prob(Reversal).
Rett is the return over the 12-month period commencing with the fourth month of fiscal year t; IBt is the
earnings per share variable in year t (annual Compustat data item #18) scaled by Pt-1 or share price
(annual Compustat data item #199) at the end of year t-1.
b
Panel B reports the results from annual rank regressions and shows the average coefficient over the
estimation period and associated t-statistic derived using the Fama-Macbeth (1973) procedure.
39
Table 6
Profile of Loss Observations by Quartile of the Probability of Reversala
Panel A: Loss History Variablesb
Sample
Obs.
Full
Persistent
Transitory
FIRSTLOSS
NUMLOSS
MAGNLOSS
40.37%
9.59%
83.51%
44.73%
77.66%
5.57%
72.02%
97.75%
23.10%
14,399
3,608
3,589
Panel B: Cash Flow and Accrualsc
Sample
Obs.
Mean
Std. Dev.
Median
CFO
Full
Persistent
Transitory
13,418
3,426
3,299
-2.908
-10.620
0.037
72.266
140.587
1.299
0.019
-0.239
0.017
ACC
Full
Persistent
Transitory
13,418
3,426
3,299
-0.252
-0.572
-0.098
19.142
37.773
1.078
-0.076
-0.130
-0.058
Panel C: R&D Expenditures and Special Itemsd
Sample
Obs.
Mean
Std. Dev.
Median
R&D
Full
Persistent
Transitory
14,399
3,608
3,589
1.145
4.097
0.027
42.244
83.481
0.184
0.000
0.005
0.000
SPI
Full
Persistent
Transitory
14,399
3,608
3,589
-0.164
-0.495
-0.048
4.699
9.309
0.246
0.000
0.000
-0.011
Panel D: Annual ERC estimation: Rett = α + β (IBt+R&Dt)+ γ R&Dt + εt
Sample
Full
Persistent
Transitory
Average
No. Obs.
597
148
147
IB+R&D
t-stat.
β
0.065
0.034
0.084
3.537
0.998
3.772
40
γ
0.008
0.056
-0.012
e
Adj. R2
R&D
t-stat
0.309
2.074
-0.428
0.025
0.039
0.038
a
The data are collected from Compustat’s Industrial and Research Annual Data Bases and covers the
period 1971-2000. Losses are based on IB, defined as income (loss) before extra-ordinary items and
discontinued operations (annual Compustat data item #18).
Persistent and Transitory refer to sub-samples based on the distribution of Prob(Reversal), the predicted
probability from the annual logistic regressions reported in Table 4: the Persistent (Transitory) subsample contains the loss observations in the first (fourth) quartile of the distribution of Prob(Reversal).
b
Panel A reports the proportion of observations reporting a 1 for the following indicator variables:
FIRSTLOSS is an indicator variable equal to one if the current year’s loss is the first in a sequence (i.e.,
the firm was profitable last year) and zero otherwise; NUMLOSS is an indicator variable equal to one if
the firm incurred more than two losses in the past five years and zero otherwise; MAGNLOSS is an
indicator variable equal to one if the sum of the current loss and the past three earnings numbers is
negative and zero otherwise.
c
CFO is cash flow from operations scaled by sales (annual Compustat data item # 12). We measure cash
flow from operations as net income (annual Compustat data item # 172) – accruals, where we measure
accruals as (∆Current Assets (data item #4) - ∆Cash (data item #1) - ∆Current Liabilities (data item #5) +
∆Debt in Current Liabilities (data item #34) + Depreciation and Amortizations (data item #14). ACC is
accruals (as defined before) scaled by sales (annual Compustat data item # 12).
d
R&D is R&D expenditures (annual Compustat data item # 46) scaled by sales (annual Compustat data
item # 12). SPI is Special Items (annual Compustat data item # 17) scaled by sales (annual Compustat
data item # 12).
e
Panel D reports the results from annual rank regressions and shows the average coefficient over the
estimation period and associated t-statistic derived using the Fama-Macbeth (1973) procedure.
41
Table 7
Loss Reversals: Time-Series Propertiesa
Panel A: Prob(Reversal): Regression on Year-indicator
Sample
Coeff.
Full
Persistent
Transitory
-0.009
-0.009
-0.005
Panel B: ERC: Regression on Year-indicator
Sample
Coeff.
a
Full
Persistent
Transitory
-0.000
-0.001
0.004
t-statistic
-5.252
-9.091
-2.083
t-statistic
-0.243
-2.122
1.284
The data are collected from Compustat’s Industrial and Research Annual Data Bases and covers the
period 1971-2000. Losses are based on IB, defined as income (loss) before extra-ordinary items and
discontinued operations (annual Compustat data item #18).
Persistent and Transitory refer to sub-samples based on the distribution of Prob(Reversal), the predicted
probability from the annual logistic regressions reported in Table 4: the Persistent (Transitory) subsample contains the loss observations in the first (fourth) quartile of the distribution of Prob(Reversal).
The table reports the coefficients and associated t-statistics of regressions of the annual medians of
Prob(Reversal) and the annual estimates of ERC (see panel B in Table 5) on a year variable.
42
Table 8
Time-series Properties of Loss Observations Componentsa
Panel A: Cash Flow and Accruals: Regression on Year-indicators
CFO
Sample
Coeff.
t-statistic
Coeff.
Full
Persistent
Transitory
-0.003
-0.047
-0.000
-7.781
-5.673
-0.674
a
0.001
0.009
-0.001
2.983
3.761
-2.273
t-statistic
-0.001
-0.005
-0.001
Panel B: R&D and Special Items: Regression on Year-indicators
R&D
Sample
Coeff.
t-statistic
Coeff.
Full
Persistent
Transitory
ACC
-0.001
-0.001
-0.001
-4.861
-4.361
-2.042
SPI
t-statistic
-2.163
-0.913
-4.824
The data are collected from Compustat’s Industrial and Research Annual Data Bases and covers the
period 1971-2000. Losses are based on IB, defined as income (loss) before extra-ordinary items and
discontinued operations (annual Compustat data item #18).
Persistent and Transitory refer to sub-samples based on the distribution of Prob(Reversal), the predicted
probability from the annual logistic regressions reported in Table 4: the Persistent (Transitory) subsample contains the loss observations in the first (fourth) quartile of the distribution of Prob(Reversal).
The table reports the coefficients and associated t-statistics of regressions of the annual medians of the
cash flow, accruals, R&D and Special Item components of losses on a year variable, where CFO is cash
flow from operations scaled by sales (annual Compustat data item # 12). We measure cash flow from
operations as net income (annual Compustat data item # 172) – accruals, where we measure accruals as
(∆Current Assets (data item #4) - ∆Cash (data item #1) - ∆Current Liabilities (data item #5) + ∆Debt in
Current Liabilities (data item #34) + Depreciation and Amortizations (data item #14); ACC is accruals (as
defined before) scaled by sales (annual Compustat data item # 12); R&D is R&D expenditures (annual
Compustat data item # 46) scaled by sales (annual Compustat data item # 12); SPI is Special Items
(annual Compustat data item # 17) scaled by sales (annual Compustat data item # 12).
43