1. science
2. mathematics
3. statistics

Extracting Sustainable Earnings from Profit Margins∗
Eli Amir
London Business School
London NW1 4SA, United Kingdom
[email protected]
Eti Einhorn
Tel Aviv University
Tel Aviv 69978, Israel
[email protected]
Itay Kama
Tel Aviv University
Tel Aviv 69978, Israel
[email protected]
November 2011
*We thank Joshua Livnat, Doron Nissim, Terrance Skantz, Florin Vasvari, Amir Ziv and participants
in the 2008 Tel Aviv Accounting Conference, the American Accounting Association Annual Meeting
2009, and in workshops held at the University of New South Wales (Sydney), the University of
Melbourne, the University of Queensland (Brisbane), and Stockholm School of Economics (Sweden)
for many useful comments. Financial support by London Business School’s RAMD Fund and the
Henry Crown Institute of Business Research in Israel is gratefully acknowledged.
Extracting Sustainable Earnings from Profit Margins
Abstract
Revenues and expenses are fundamentally proportional to each other, but are likely to be
disproportionally affected by transitory accounting or economic shocks. We build on this
observation and propose a new measure of sustainable earnings based on deviations from
normal profit margins. While some other sustainable earnings metrics attempt to identify
transitory components on a line-by-line basis, our measure uses a vertical section to extract
the transitory part of earnings from all line items. We show that our measure is positively
associated with earnings persistence, better earnings predictability, and stronger market
reaction to unexpected earnings. We also show that our measure is positively associated with
the magnitude of the post-earnings announcement drift.
Extracting Sustainable Earnings from Profit Margins
1. Introduction
The definition and measurement of the quality of earnings is an issue that has received
much attention in the accounting literature. Lev (1989) argues that the power of earnings in
explaining stock returns is low due primarily to their low quality, prompting researchers to
focus on developing and testing direct and indirect measures of earnings quality. Dechow et
al. (2010) identify three categories of earnings quality proxies: (i) properties of earnings (for
instance, earnings persistence), (ii) investors responsiveness to earnings (for instance,
earnings response coefficient), and (iii) external indicators of earnings misstatement (for
instance, accounting and auditing enforcement release).1 While earnings quality can be
measured from different perspectives, including the measurement perspective and the
earnings management perspective (Francis et al., 2006), the popular measures that emerged
in the literature (Dechow and Schrand, 2004) were associated with the ability of current
earnings to predict future earnings and to explain stock returns. Further research into the
association between equity values and earnings components has yielded the empirical
observation that different components of earnings have different persistence and are
therefore priced differentially by equity investors.2
Transitory earnings components, which may arise from reporting manipulations,
accounting measurement problems, and non-recurring economic events, suppress the
persistence and predictability of reported earnings and introduce a substantial amount of
noise into the process of accounting-based equity valuation; decreasing earnings quality.
1
Dechow et al. (2010) define earnings quality as follows: "Higher quality earnings provide more information
about the features of a firm’s financial performance that are relevant to a specific decision made by a specific
decision-maker."
2
See, for example, Lipe (1986) Wilson (1987), Barth et al. (1992), Ohlson and Penman (1992), Sloan (1996),
Ramakrishnan and Thomas (1998), Ertimur et al. (2003). Another measure of earnings quality from the
perspective of earnings management is the magnitude of discretionary accruals (for example, Jones, 1991;
Dechow et al., 1995; Kothari et al., 2005).
1
Consequently, financial analysts and investors care about the sustainable component of
earnings because equity values are based on expected future earnings rather than current
earnings, as argued by Penman and Zhang (2006). Thus, investors should pay more for
sustainable (more persistent) earnings. This is why financial analysis focuses on extracting
information on the core (or sustainable) component of earnings using time-series and crosssectional techniques, separating it from the non-core (or transitory) component.3
Though investors can identify some transitory components of earnings by looking at
the decomposition of earnings into their reported items (e.g., discontinued operations, writeoffs), there are other transitory components that are hidden and cannot be detected this way,
mostly because of earnings management and the accounting aggregation process. The
empirical literature has been rather silent about the way that investors reveal hidden
transitory components of accounting earnings and the effect of this process on stock prices.
While presenting a comprehensive and descriptive picture of the statistical association
between accounting information and equity values, the extant empirical evidence sheds very
little light on the process whereby investors endeavor to detect hidden transitory components
of earnings and adjust for them when pricing equity, and the extent to which this process is
effective.
Our paper is motivated by this void in the literature and the importance of measuring
the quality of reported earnings, and in particular in distinguishing between the core
(sustainable) and the non-core (transitory) components of earnings. We propose a new
measure for assessing the sustainable component of earnings, based on deviations from
normal profit margins. This measure, referred to as the Intensity of Core Earnings (ICE), is
derived from the observation that revenues and expenses are fundamentally proportional to
each other, but are likely to be disproportionally affected by transitory accounting or
3
Other terms used to describe core earnings are normal, continuing, sustainable, permanent, head-light, and
long-run central earnings.
2
economic shocks. Consequently, transitory revenues or expenses are likely to alter the
fundamental stochastic behavior of profit margins. This measure is also motivated by Schilit
and Perler (2010), who argue that deviations from normal profit margins often indicate
accounting manipulation. Thus, financial statement users can identify deviations of the
reported earnings items from what is implied by their expected normal profit margins,
knowing that these deviations might be partly caused by transitory items. These deviations
of actual earnings from what is implied by normal profit margins might serve investors in
distinguishing between core and non-core earnings, and thus assist them in assessing
earnings persistence and predictability. In particular, we expect the larger the deviation of
earnings from what is implied by the normal profit margin, the lower is the persistence and
predictability of earnings, which should also be reflected by lower market reaction to
unexpected earnings.
The frequent use of both time-series and cross-sectional in financial analysis motivates
us to use two alternative proxies of normal profit margins. The first is the firm-specific
average profit margin over the preceding four years (time-series). The assumption embedded
in this measure is that profit margins revert to their fundamental value over time. The second
measure is the current average profit margin in the industry to which the firm belongs (crosssection). The assumption embedded in this measure is that while each firm may deviate from
its fundamental profit margin, the average profit margin in the industry is an unbiased
measure of the fundamental profit margin.4 Using these proxies of normal profit margins, we
estimate core earnings by multiplying the normal profit margin by current sales. Non-core
earnings are then estimated as the difference between actual and core earnings. Based on
estimates of core and non-core earnings, we construct the ICE measure as the absolute value
4
Fairfield et al. (2009) argue that industry analysis has only marginal incremental information over firmspecific figures in forecasting RNOA, ROCE and growth in NOA, but it is useful in predicting future sales
growth.
3
of the core component of earnings divided by the sum of the absolute values of the core and
the non-core components of earnings. We show that our ICE measure is positively associated
with (i) the persistence of earnings, (ii) better earnings predictability as reflected by the
accuracy of analysts' earnings forecasts, and (iii) stronger contemporaneous and subsequent
market reaction to earnings. That is, our ICE measure is strongly associated with the main
attributes of earnings quality, and thus is a valid indicator for the quality of earnings.
The first advantage of using the ICE as a measure of earnings quality is its simplicity –
it can be calculated for each firm at any point in time regardless of whether the firm is listed
or private. In addition, while some other earnings quality measures attempt to identify
transitory components on a line-by-line basis, our measure use a vertical section to extract
the transitory part of earnings from all line items. It is comprehensive and less dependent on
the quality of accounting disclosure. Moreover, this measure can be applied to any level of
income aggregation – gross profit, operating profit, etc.5
The empirical tests that we carry out here are based on a large sample that covers the
years 1990-2009 and includes all available firm/quarter observations with complete price
and financial data on Compustat and CRSP, excluding financial institutions and public
utilities. We begin our analysis by investigating the persistence of overall earnings, the core
component and the non-core component. Using cross-sectional and time-series regressions
we find that the persistence of core earnings, as measured here, is substantially larger than
that of overall earnings, which in turn is larger than the persistence of non-core earnings.
This result suggests that our measure of core earnings based on profit margins is useful in
removing transitory components of earnings, and assessing the sustainable components
earnings. In addition, we find that the persistence of earnings increases monotonically with
5
Clearly, a measure of earnings quality may have certain limitations due primarily to its simplicity. For
instance, sudden changes in cost structure may appear as a deviation from normal profit margin in the short run
until the earnings process stabilizes.
4
the ICE measure. These results indicate that our ICE measure is a valid measure of earnings
persistence, which is an important property of earnings quality.
We continue with analyzing the link between our ICE measure and three attributes of
analysts’ earnings forecasts: Accuracy (absolute forecast errors), dispersion (standard
deviation of forecasts) and bias (the magnitude of forecast errors). We show that higher ICE
is associated with more accurate earnings forecasts, less dispersed forecasts and less
optimistic forecasts. We also show that analysts are on average optimistic with respect to
companies with low ICE and pessimistic with respect to companies with high ICE. This
result suggests that our ICE measure is associated with improved earnings predictability, but
also open the door to the possibility that ICE is not fully priced by the market.
Turning to market reaction tests, we show that the market reaction around the
announcement of preliminary quarterly earnings is positively associated with our ICE
measure. In fact, when we sort companies into quintiles based on ICE, the market reaction
increases monotonically with ICE quintiles. This result supports the argument that ICE is a
valid and useful measure of investors' responsiveness to earnings, which is also a significant
attribute of earnings quality. We also find that post-earnings excess stock returns are
associated with ICE, suggesting that ICE is not fully priced by the market.
Overall, the evidence provided here suggests that the ICE provides useful information
in identifying hidden transitory components of earnings, assessing sustainable earnings and
by that improve the accuracy of earnings forecasts and the explanatory of stock returns.
We contribute to the literature by introducing a powerful, yet simple, tool that serves
investors in assessing the quality of earnings in three mutually related dimensions: (i) the
persistence of reported earnings, (ii) the ability of analysts to forecast earnings, and (iii) the
power of earnings in explaining stock returns. Furthermore, besides extending our
understanding of the means by which investors detect low persistence of earnings embedded
5
in reported earnings and adjust for that in their predictions and pricing firm equity, this study
also ties together many previously documented statistical patterns in equity valuation,
placing them all in the same conceptual accounting context. In particular, our analysis
suggests a role for disaggregated current and past accounting data in equity valuation.6 It
also employs and highlights the function of financial ratios in the valuation process.7
We proceed as follows. In section 2, we construct our measure of intensity of core
earnings (ICE) based on firm-specific and industry benchmarks. Section 3 discusses the
sample and data sources, and provides descriptive statistics and simple correlations. Section
4 presents our main analysis. Section 5 offers concluding remarks.
2. The Intensity of Core Earnings
The basic premise of this study is that current and past net profit margins (NPM) can
be used to extract a useful measure of core (sustainable) earnings, separating out the noncore (transitory) component of earnings. For each firm i and quarter t, the net profit margin
(NPMit) is defined as net income (NIit) divided by total sales (Salesit), that is NPMit =
NIit/Salesit. We use two benchmarks for separating the core from the non-core component of
income: a firm-specific benchmark based on previous profit margins and an industry
benchmark. These benchmarks reflect the common practice of using time-series as well as
cross-sectional financial analysis.
Using the firm itself as a benchmark, we define the core component of income
(FCOREit) as firm i's NPM averaged over the same quarter in the previous four years,
multiplied by current sales. That is:
6
See, for example, Lipe (1986), Rayburn (1986), Wilson (1987), Ohlson and Penman (1992), Dechow (1994),
Sloan (1996), Penman (1998), Ramakrishnan and Thomas (1998), Ertimur et al. (2003), Dechow and Schrand
(2004), Jegadeesh and Livnat (2006).
7
See, for example, Ohlson (1980), Freeman et al. (1982), Ou and Penman (1989), Lev and Thiagarajan (1993),
Abarbanell and Bushee (1997), Nissim and Penman (2001), Soliman (2008), Amir et el. (2011).
6
FCOREit = [(NPMi,t-4 + NPMi,t-8 + NPMi,t-12 + NPMi,t-16)/4]* Salesit
The non-core component of earnings (FNCOREit) is simply the difference between net
income and the core component of income: FNCOREit = NIit – FCOREit.
The industry-based core component of earnings, ICOREit, is measured relative to
industry profit margin, where industry affiliation is based on 2-digit SIC codes.8 We initially
measure industry NPM each quarter using all firms in the same industry. Then, we measure
firm i's core earnings by multiplying the industry NPM by firm i's sales, as follows:
⎡
ICORE it = ⎢ ∑ NI kt
⎣ k∈I ( i )
∑ Sales
k∈I ( i )
kt
⎤
⎥ ∗ Sales it
⎦
where I (i ) is the set of indices of all firms that belong to the industry of firm i .
Accordingly, the industry-based non-core component of earnings is the difference between
net income minus the industry-based core component of net income: INCOREit = NIit –
ICOREit
Next, we define two measures of sustainable earnings– firm-specific one and an
industry-based one. These measures, referred to as intensity of core earnings (ICE), are
defined as the proportion of the absolute value of core income divided by the sum of the
absolute values of core and non-core components of income.9 The intensity measures based
on firm-specific prior profit margins (FINT) and industry profit margins (IINT) are,
respectively:
FINTit =
FCORE it
FCORE it + FNCORE it
, and IINTit =
8
ICORE it
ICORE it + INCORE
it
We repeated the analyses using the industry classification suggested by Kenneth French. Results (not
tabulated) are very similar. See http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html.
9
We use absolute values to capture the magnitude of the deviation of actual profits from what is implied by
normal profit margins (rather than the sign of the deviation) because deviations from both sides mean lower
precision.
7
3. Sample and descriptive statistics
The initial sample includes all observations with complete financial data on Compustat
and stock returns on CRSP during 1990-2009, with market value of equity above $10 million
at quarter-end. We delete firm/quarter observations with missing quarterly data on market
value of equity, book value of equity, sales, and net income over the preceding four years.
We exclude financial institutions (1-digit SIC = 6) and public utilities (2-digit SIC = 49)
because the structure of their financial statements is incompatible with those of industrial
companies. To limit the effect of extreme observations, each quarter we rank the sample
according to the variables and remove the extreme 1% of the observations on each side.
Finally, we remove firms with less than eight quarterly observations and 2-digit SIC
industries in quarters with less than five active firms.
The analysts' earnings forecast sample includes observations with available forecast
data on IBES.10 The full sample includes 103,998 usable firm/quarter observations for 3,804
different firms. The analysts' earnings forecast sample includes 72,898 usable firm/quarter
observations for 3,336 different firms. Table 1 presents the number of quarterly observations
for each year in our sample.
(Table 1 about here)
Table 2 presents descriptive statistics (Panel A) and pair-wise correlations (Panel B)
for the main variables. In addition to the variables forming the ICE measures, we provide
descriptive statistics for three abnormal return windows (AR), book-to-market ratio (BM),
firm size (MV), and analysts' forecasts error (FE). AR(SW) is the 3-day excess buy&hold
return around the preliminary earnings announcement date; AR(PREFILE) is the excess
buy&hold return from two days after preliminary announcement through one day after filing;
10
Consistent with Gu and Wu (2003) and Weiss (2011) we require in this analysis that stock price be at least $3
to avoid the small deflator problem. We replicate our analysis using all firms with stock price over $1 obtaining
virtually the same results (not tabulated).
8
AR(POSTFILE) is the excess buy&hold return from two days after filing through one day
after the next preliminary announcement, if available, or plus 90 days if the next preliminary
earnings announcement is unavailable; BM is measured as book value of equity at quarterend divided by market value of common equity; MV is measured as market value of
common equity at quarter-end; FE is measured as reported earnings per share minus mean
consensus analysts' forecasts, deflated by the stock price at the end of the prior quarter.
As Panel A shows, sales and net income (NI) are skewed to the right as reflected by
the medians being smaller than the means, which is consistent with prior studies. In contrast,
net profit margin (NPM) is slightly skewed to the left (mean = 0.03; median = 0.04).
Absolute core and non-core components of NI are also skewed to the right. Furthermore, the
absolute core component of NI is larger than the absolute non-core component of NI for both
the firm-specific and the industry-based measures. The distributions of firm-specific
intensity (FINT) and industry-based intensity (IINT) are similar to each other, as the means
of FINT and IINT are 0.61 and 0.57, and the standard deviations are 0.26 and 0.25,
respectively. The medians of FINT and IINT are about 0.6, with inter-quartile range of
approximately 0.4 to 0.8.
Means and medians buy-and-hold abnormal returns for the contemporaneous and
post-preliminary earnings announcement returns are zero, by construction. As in prior
studies, market values (MV) and book-to-market ratios (BM) are skewed to the right. Mean
and median analysts' forecast errors (FE) are close to zero, which is also consistent with the
existing literature.
Panel B of Table 2 presents pair-wise Pearson (above diagonal) and Spearman (below
diagonal) correlations between the main variables. The correlations between NI and its core
and non-core components are positive, however, the correlation between NI and its core
9
component (FCORE or ICORE) is significantly larger (at the 0.01 level, not reported in a
table) than that between NI and its non-core component (FNCORE or INCORE).
Also, the correlations between firm-specific and industry-based core and non-core
components are positive. For example, the Spearman correlation between FCORE and
ICORE is 0.62 and between FNCORE and INCORE it is 0.32. Furthermore, the Spearman
correlation between firm-specific core intensity (FINT) and industry-based core intensity
(IINT) is surprisingly low, 0.21. These correlations suggest that firm-specific and industrybased profitability analyses are complementary to each other. The correlations between the
core and non-core components of income are negative by construction; for instance, the
Spearman correlation between FCORE and FNCORE is -0.21. In addition, larger firms tend
to report more stable earnings, as reflected by the positive correlation between core intensity
and market value of equity (Spearman correlation between FINT and MV is 0.21). Finally,
companies with larger book-to-market ratios have lower firm-specific ICE measures
(Spearman correlation between FINT and BM is -0.16). This result is interesting because it
indicates that our measure of core earnings intensity is significantly different than a measure
of growth opportunities, which are captured by low book-to-market ratios.
(Table 2 about here)
Figure 1 presents average firm-specific and industry-based intensities (FINT and IINT)
over the sample period (1990-2009). While average intensities are similar to each other
(about 0.6), firm-specific intensity is relatively stable over time, while the industry-based
intensity is more volatile; in fact it is associated with economy-wide declines that occurred in
the early 1990s, 2001, and 2008. This is because firm-specific intensity is measured
relatively to previous years, implicitly smoothing large economic shocks; in contrast,
industry-based intensity is cross-sectional in nature.
(Figure 1 about here)
10
4. Empirical Analysis
4.1 Intensity of core earnings and earnings' persistence
If deviations from normal profit margins assist in extracting sustainable earnings, we
would expect the persistence of the core component of earnings, as measured here, to be
substantially larger than that of the non-core component of earnings. To estimate the
persistence of income, we use the coefficient α 1i obtained from the following regression
model estimated on a firm-by-firm basis:
NIit = α0i + α1i NIi ,t −4 + α2iCV ( NI )i,t + α3i BMi,t + α4i MVi,t + ε i,t
(1a)
Where CV(NI)it is the coefficient of variation of NI in the same quarter over the previous
four years, BM it is the book-to-market ratio, and MVit is the market value of common equity
at quarter-end, for firm i and quarter t . We repeat this process for the firm-specific and
industry-based core component of income:
FCOREi ,t = α 0i + α1i FCOREi ,t −4 + α 2i CV ( NI )i ,t + α 3i BM i ,t + α 4i MVi ,t + ε i ,t
(1b)
ICOREi ,t = α 0i + α1i ICOREi ,t −4 + α 2i CV ( NI ) i ,t + α 3i BM i ,t + α 4i MVi ,t + ε i ,t
(1c)
and also for the firm-specific and industry-based non-core component of income:
FNCOREi ,t = α 0i + α1i FNCOREi ,t −4 + α 2i CV ( NI ) i ,t + α 3i BM i ,t + α 4i MVi ,t + ε i ,t
(1d)
INCORE i ,t = α 0i + α1i INCORE i ,t −4 + α 2i CV ( NI) i ,t + α 3i BM i ,t + α 4i MVi ,t + ε i ,t
(1e)
We expect α1i to be higher for the core component of income (Equations 1b and 1c) than for
the non-core component (Equations 1d and 1e).11
11
We repeated the analyses on a quarter-by-quarter basis (cross-section). In addition, we added accruals as a
control variable and omitted the coefficient of variation from the model. Results (not tabulated) are not
sensitive to these model changes.
11
To validate our measure of core earnings intensity, we expect earnings persistence to
be higher for firms with higher intensity of core income. We estimate equation (2a) on a
quarter by quarter basis (similar to Fama and MacBeth, 1973).
NI it = β 0 t + β1t Di ,t − 4 + β 2 t NI i ,t −4 + β 3t Di ,t − 4 NI i ,t −4 + β 4 t CV ( NI ) it + β 5t BM it + β 6 t MVit + ε it
(2a)
where Di,t-4 is a dummy variable that equals "1" if core earnings intensity (FINT or IINT) for
firm i is above the quarterly median at time t-4 (same quarter in the prior year), and "0"
otherwise. We expect a positive association between the intensity of core earnings and the
persistence of earnings, that is, β3 > 0.
In addition, for each quarter, we sorted all firms according to their ICE measures
(FINT and IINT) in quarter t-4. Then, we assign each firm-quarter into quintile portfolios
based on the intensity of core earnings in quarter t-4. We estimate equation (2b) in each
quarter for each of the five quintile portfolios.
NIit = γ 01t + γ 1t NIit−4 + γ 2t CV ( NI )it + γ 3t BMit + γ 4t MVit +ψ it (2b)
We expect γ1 (the earnings persistence coefficient) to increase with the intensity of core
earnings.
Table 3 presents regression results for equations (1a) – (1e). The average coefficient of
earnings persistence (α1) is 0.18 (significantly larger than zero at the 0.01 level). When we
divide net income into core and non-core components based on prior firm-specific profit
margins (FCORE and FNCORE), the average persistence of core earnings is 0.66 and 0.05,
respectively. In fact, the persistence of non-core earnings is only significant at the 0.10 level.
When core and non-core components of net income are measured relative to industry profit
margins (ICORE and INCORE), the average persistence of core and non-core earnings are
0.35 and 0.20, respectively.
12
Consistent with our prediction, the persistence of core earnings, measured by α1, is
significantly larger than that of non-core earnings. The difference in persistence coefficients
is significant at the 0.01 level for the firm-specific measure, and at the 0.05 level for the
industry-based measure. Notice that the distance between the industry-based core and noncore components (0.35 - 0.20 = 0.15) is significantly smaller (at the 0.01 level) than the
distance between the firm-specific core and non-core components (0.66 - 0.05 = 0.61). This
finding suggests that sustainable earnings extracted from prior firm-specific profit margins
are more informative about future earnings than those extracted from industry-based profit
margins.
Regarding the control variables in equation (1), the coefficients on firm size (MV) are
positive and significant at the 0.01 level in equations (1b), (1c) and (1d). The coefficients on
the book-to-market ratios are positive in all models but one (1c). These coefficients are
generally significant at the 0.05 level or better. This result suggests that larger core earnings
indicate, on average, lower expected growth, while larger non-core earnings are positively
associated with expected growth in earnings. Finally, the coefficients on the earnings'
coefficient of variation are significant only in models (1b) and (1c); this variable is
negatively associated with core earnings and positively associated with non-core earnings.
(Table 3 about here)
Table 4, Panel A, presents the estimation results for model (2a). This equation allows
the persistence of earnings to vary by firm-specific and industry-based intensity of core
earnings (FINT and IINT). For the first specification (the full sample without a Dummy
variable) the average earnings persistence coefficient, measured as β2,, is 0.36 (significantly
different from zero at the 0.01 level). Focusing on the second (third) specification, when
FINT (IINT) is below the quarterly median the persistence of net income is 0.28 (0.31).
13
However, when FINT (IINT) is above the quarterly median, the persistence of earnings
increases (at the 0.01 level) by 0.23 (0.17).
We also estimated equation (2b), each quarter, for five quintiles of ICE, where
quintiles are formed in quarter t-4. Table 4, Panel B, shows that the average persistence
coefficients, γ 1 , increases monotonically with the intensity quintile for both firm-specific and
industry-based measures of core intensity. The difference in γ 1 between the lower and higher
quintiles is significant at the 0.01 level. Overall, the results in Table 4 show a strong positive
association between earnings persistence and the intensity of core earnings, by that
validating our measure. These results also support the view that analyzing deviations from
normal profit margins is a useful method for extracting information on sustainable earnings.
(Table 4 about here)
4.2 Intensity of Core Earnings and the predictability of earnings
A useful measure of sustainable earnings should be associated with improved earnings
predictability, and in particular, the quality of analysts' earnings forecasts. We therefore
examine the association between the intensity of core earnings in period t-4 (a year before
the forecasts) and three analysts' earnings forecast attributes: (i) forecast accuracy in quarter
t, measured as absolute forecast errors; (ii) forecast dispersion in quarter t, measured as the
standard deviation of forecasts; and (iii) forecast bias in quarter t, measured as the magnitude
of forecast error. Consistent with prior studies, forecast errors for firm i in quarter t (FEit) is
computed as reported earnings per share minus average analysts' forecasts announced in the
month immediately preceding that of the earnings announcement (as reported in IBES),
deflated by the stock price at the end of the previous quarter. We expect the intensity of core
14
earnings to be negatively associated with absolute forecast errors and with the standard
deviation of forecasts.12
To test our prediction regarding the positive association between the intensity of core
earnings and the quality of analysts' earnings forecasts, we form quintile portfolios according
to intensity of core earnings. Specifically, in each quarter, we sorted all observation
according to their FINT or IINT in quarter t-4 and assigned the firm into quintiles. Then, for
each quintile, we measure mean analysts' forecast accuracy, mean forecast dispersion, and
mean forecast bias. Note that the intensity of core earnings is determined in quarter t-4,
whereas the quality of the forecasts is measured in quarter t (a year later). Therefore, this
analysis could yield powerful insights as to the usefulness of our intensity measure.
Tables 5 presents, for each intensity quintile, mean analysts' forecast accuracy, mean
forecast dispersion, and mean forecast bias in quarter t (accuracy, dispersion, and bias values
are multiply by 1,000). Additionally, for each quintile, we compute the percentage of lossreporting firms in period t. Panel A provides results for quintiles assigned according to firmspecific core earnings intensity (FINT), and Panel B provides results for quintiles assigned
according to industry-based intensity (IINT).
Focusing on analysts' accuracy in period t, we observe a monotonic decrease in mean
absolute forecast errors [ABS(FE)] as we go up to a higher level of intensity quintile. The
difference in [ABS(FE)] between extreme quintiles of core intensity is 1.72 and 0.81 for
firm-specific intensity (FINT) and industry-based intensity (IINT), respectively, significantly
different from zero at the 0.01 level. This evidence shows a strong association between
current core earnings intensity and the accuracy of subsequent earnings forecasts. We also
12
We replicate our analysis using actual earnings as a deflator instead of stock price in the prior quarter.
Results (not tabulated) are qualitatively the same. Also, in measuring analysts' dispersion we limit our sample
to firm/quarter observations with minimum of three different analysts' earnings forecasts. Limiting the
dispersion analysis to firm/quarter observations with minimum of two different analysts' earnings forecasts
does not change the qualitative results. Also, limiting the analysis of analysts' accuracy and analysts' bias to a
minimum of two or three analysts' earnings forecasts does not change the qualitative results.
15
observe a monotonic decline in forecast dispersion as we go up to a higher level of intensity
quintile. The difference in forecast dispersion between the extreme intensity quintiles is 0.82
(0.41) for FINT (IINT), significant at the 0.01 level.
Interestingly, analysts' bias (FE) in period t is also associated with firm-specific core
intensity in period t-4. As Panel A shows, we find a monotonic increase in mean forecast
errors as we go up to a higher level of firm-specific intensity quintile. While mean forecast
error is -0.14 (significantly different from zero at the 0.05 level) in the lower quintile, it is
+0.06 in the upper quintile (significantly different from zero at the 0.10 level). The
difference in FE between these extreme quintiles is 0.20, which is significant at the 0.01
level. This result suggests that financial analysts tend to be optimistic when the intensity of
core earnings is low, but tend to be rather pessimistic about earnings forecasts of companies
with high intensity of core earnings. Since the intensity of core earnings is positively
associated with earnings persistence, it implies that analysts' bias is associated with their
misperception of earnings' persistence. We note there is no significant association between
analysts' bias and industry-based core intensity.
To get further insight on the association between core intensity and the predictability
of earnings, we also examine whether ICE in period t-4 is associated with the probability of
losses in the current quarter. Specifically, we present that information for the entire sample
and for those companies that reported positive earnings in quarter t-4 (that is, NIt-4 > 0). For
the full sample, both panels indicate a monotonic decrease in the percentage of loss-reporting
firms in period t, as we go up to a higher level of intensity quintile in t-4. Specifically, the
percentage of loss-firms in the lower quintile of FINT (IINT) is 26% (32%), whereas the
percentage of loss-firms in the upper quintile of FINT (IINT) is only 8% (10%). As for the
sub-sample of firms with reported profits in quarter t-4, the monotonic decline in the
16
frequency of losses holds only for the firm-specific intensity measure; much less for the
industry-based intensity. 13
To summarize, the results in Table 5 indicate a strong association between ICE and
forecast attributes, suggesting that ICE is positively associated with better earnings
predictability. Specifically, high intensity of core earnings measured in quarter t-4 is
associated with more accurate forecasts, less dispersed forecasts and less optimistic forecasts
in quarter t. This association could be related to the frequency of losses, as the intensity level
in period t-4 is negatively associated with the probability of reporting a loss in period t.
(Table 5 about here)
4.3 Contemporaneous Market Reaction to Earnings and the Intensity of Core Earnings
We have established a strong link between ICE and two important properties of
earnings quality - persistence and predictability. A useful measure of earnings quality should
also be associated with a stronger market reaction to unexpected earnings. To examine
whether higher intensity of core earnings is indeed associated with a stronger market
reaction we estimate equation (3a) on quarter-by-quarter basis for firm-specific intensity
(FINT) and industry-based intensity (IINT).
AR(SW )it = δ 0t + δ1t Dit + δ 2t FEit + δ3t Dit ∗ FEit +ηit
(3a)
The dependent variable, AR(SW)it, is the 3-day excess buy&hold return around firm i's
preliminary earnings announcement date in quarter t (calculated as the buy&hold return on
the security minus the average buy&hold return on a portfolio of firms with similar size and
B/M). The explanatory variables are unexpected earnings, measured as analysts' earnings
forecast error for firm i in quarter t (FEit), and a dummy variable (Dit) that obtains the value
of "1" if firm i's intensity (either FINT or IINT) is above the median in quarter t, and "0"
13
We also replicate the analysis of analysts' accuracy, analysts' dispersion, and analysts' bias using a subsample of firms that reports positive earnings. Results (not tabulated for brevity) are qualitatively the same.
17
otherwise.14 We expect δ3 to be positive if a positive association exists between the intensity
of core earnings and the market reaction to unexpected earnings.
We also assign firms each quarter to quintiles based on ICE in quarter t (FINT and
IINT) and estimate the following equation: AR(SW )it = λ0t + λ1t FEit + ηit . We expect the
earnings response coefficient (λ1) to increase with the intensity of core earnings.
Table 6, Panel A, presents average coefficients and corresponding t-statistics for
equation (3a), which is estimated on a quarter-by-quarter basis. The first specification is
estimated without the intensity dummy, and is provided as a benchmark. The average
earnings response coefficient for that specification (δ2) is 3.60 (significant at the 0.01 level).
Focusing on the second specification, when the intensity of firm-specific core earnings is
below the median, the earnings response coefficient is 3.21 (significant at the 0.01 level);
this coefficient increases by 1.73 (significant at the 0.01 level) when the firm-specific
intensity is above the quarterly median. Similarly, when the industry-based intensity of core
earnings is below the quarterly median, the earnings response coefficient is 2.90, increasing
by 2.51 (significant at the 0.01 level) when the industry-based intensity is above the quarter
median.
The quintile analysis in Panel B shows a monotonic increase in earnings response
coefficients as we go up to a higher level of core earnings intensity. Specifically, the
coefficient λ1 obtained from equation (3b) increases monotonically with the intensity
quintiles. The difference in λ1 between the lower and the upper quintiles of core intensity is
3.28 (3.17) for FINT (IINT), significantly different from zero at the 0.01 level.
The results in Table 6 suggest that ICE is useful in explaining contemporaneous
market reaction to unexpected earnings. In particular, the earnings response coefficient,
14
We replicate the analysis of contemporaneous market reaction using standardized unexpected earnings
(SUE), and standardized unexpected revenues (SURG) instead of analysts' forecasts error. Results (not
tabulated) regarding the effect of ICE on market reaction to unexpected earnings are qualitatively the same.
18
which is an important attribute of earnings quality, increases with ICE. This finding validates
ICE as a measure of sustainable earnings.
(Table 6 about here)
4.4 Post Earnings Announcement Drift and the Intensity of Core Earnings
The intensity of core earnings (ICE) is a measure that could be calculated for each
firm/quarter based only on profit margins. Also, results in section 4.2 show that analysts'
earnings forecast bias is systematically associated with ICE. An obvious question that arises
is whether ICE is fully reflected in stock prices. Moreover, the post-earnings announcement
drift is often attributed to incorrect estimation of earnings persistence (Bernard and Thomas,
1989; 1990; and Chan et al., 1996), which is linked to our measure. We therefore examine
the effect of firm-specific (FINT) and industry-based (IINT) intensities of core earnings on
post-preliminary earnings announcement stock returns. We use two return windows: The
first one starts two days after the preliminary announcement and ends one day after the SEC
filing (denoted as PREFILE), and the second one starts two days after the SEC filing and
ends one day after the subsequent preliminary earnings announcement, if available, or
otherwise plus 90 days (denoted as POSTFILE). Specifically, we estimate two regression
models as follows:
AR( PREFILE)it = κ 0t + κ1t Dit + κ 2t FEit + κ3t Dit ∗ FEit +ηit
(4a)
AR( POSTFILE) it = κ 0t + κ1t Dit + κ 2t FEit + κ 3t Dit ∗ FEit + ηit
(5a)
The dependent variables are excess buy&hold stock returns for the post-preliminary
announcement windows, and all other variables are as described in the previous section.
We also form quintiles each quarter according to the intensity of core earnings (both
FINT and IINT) and estimate the following equations for each quintile using the postpreliminary earnings announcement windows as the dependent variables:
19
AR( PREFILE)it = ν 0t +ν1t FEit +ηit
(4b)
AR( POSTFILE) it = ν 0t +ν 1t FEit + ηit
(5b)
If the market does not fully incorporate the intensity of core earnings into stock prices, the
coefficients ν1 will be non-zero. However, it is impossible to form ex-ante predictions as to
the sign of ν1.
Table 7 provides the results of the post-earnings announcement drift in a format similar
to that used in analyzing the contemporaneous market reaction. For the benchmark
specifications (1a and 2a), the coefficient κ 2 is 0.914 and 0.617 for the PREFILE and
POSTFILE windows, respectively (significantly different from zero at the 0.01 level). The
positive drift is consistent with prior studies and confirms the existence of a significant postannouncement drift in our samples.15
Specifications 1b and 2b show a stronger post-announcement drift for companies with
above-median firm-specific intensity (FINT). Specifically, the coefficient κ 3 is positive
(0.372) and significant at the 0.05 level for the PREFILE window; this coefficient is positive
(0.313) and significant at the 0.10 level for the POSTFILE window. Specifications 1c and 2c
repeat the analysis for industry-based intensity. The results are very similar: companies with
above median industry-based intensity have larger drifts, as reflected by the positive and
significant coefficients κ 3 .
Interestingly, when core earnings intensity is based on firm-specific profit margins
(FINT), the coefficients κ 1 are also positive and significantly different from zero at the 0.05
level for the PREFILE window and at the 0.10 level for the POSTFILE window. This
finding suggests that firms with above-median firm-specific intensity have stronger drifts
regardless of the magnitude of unexpected earnings.
15
For evidence on the post-earnings announcement drift, see, for example, Bernard and Thomas (1989, 1990).
20
Panel B of Table 8 presents a quintile analysis of post-announcement drifts. We report
the coefficientsν 1t for each quintile of firm-specific (FINT) and industry-based (IINT)
intensity. We find that for both windows (PREFILE and POSTFILE), and for both FINT and
IINT, ν1 increases almost monotonically as we go up to a higher level of intensity quintile
(with the exception of moving from Quintile 4 to Quintile 5). The difference in ν1 between
the lower and upper quintiles of core intensity is positive in all four cases, but the difference
is significant at the 0.05 level only for quintiles assigned according to FINT in the PREFILE
window.
To summarize, the analysis in Table 7 indicates that ICE is positively associated with
the magnitude of the post-earnings announcement drift. In particular, the drift is significantly
higher for firms with above-median intensity, consistent with the argument that the market
does not fully price the effect of core earnings intensity on earnings persistence. This finding
is also consistent with our analysts’ forecast analysis, where we find that earnings forecasts
tend to be optimistic for firms with low intensity, and pessimistic for firms with high
intensity.
(Table 7 about here)
4.5 Additional tests
The proposed earnings quality measure can be computed for any definition of core
versus non-core earnings components. In particular, it is possible to compare it to an
intensity measure based on cash from operations (CFO), as in Sloan (1996). We compute the
intensity of current cash flows (absolute value of current CFO divided by absolute value of
current CFO plus absolute value of current accruals). We show that our measure of the
intensity of core profit is better in assessing earnings quality than a measure based on the
intensity of current operating cash flows. We also compute a cash-based intensity measure
21
based on deviations from average operating-cash-to-sales ratios. We find that our measures
of the intensity of core earnings and the intensity of core operating cash flows provide
distinct information over each other in explaining future earnings and annual stock returns.
5. Summary and Conclusions
We introduce a simple, yet powerful, method that serves investors to imperfectly clear
reported earnings of transitory components, and thus extract sustainable earnings. This
method, which is based on extracting information from profit margins, facilitates the
estimation of the core (sustainable) and non-core (transitory) components of earnings, as
well as the construction of a new measure associated with the main attributes of earnings
quality – the intensity of core earnings.
As financial ratios are made up of two economically related measures, a deviation of a
ratio from its normal value is more likely to reflect a transitory shock. Thus, for instance, if
net income increases, one would expect sales to increase as well, and vise versa. An increase
of income (sales) without a corresponding increase in sales (income) is likely to cast doubts
on the sustainability of these increases. Using this argument, we propose a new methodology
for assessing sustainable earnings based on deviations of earnings from what is implied by
normal ratios. Our proposed measure is simple and applicable to any firm, public or private,
at any point in time and for any level of information aggregation. We show that our measure
is positively related to the main attributes of earnings quality: earnings persistence, the
quality of analysts’ earnings forecast, and the earnings response coefficient, and thus is a
valid indicator of the quality of earnings.
22
References
Abarbanell, J.S., and B.J. Bushee. 1997. Fundamental analysis, future earnings, and stock
returns. Journal of Accounting Research, vol. 35, pp. 1-24.
Amir, E., I. Kama, and J. Livnat. 2011. Conditional versus unconditional persistence of
RNOA components: Implications for valuation. Review of Accounting Studies, vol. 16, pp.
302-327.
Barth, M.E., W.H. Beaver, and W. Landsman. 1992. The market valuation implications of
net periodic pension cost components. Journal of Accounting and Economics, vol. 18, pp.
27-62.
Bernard, V.L., and J.K. Thomas. 1989. Post-earnings-announcement drift: Delayed price
response or risk premium? Journal of Accounting Research, vol. 27, pp. 1-36.
Bernard, V.L., and J.K. Thomas. 1990. Evidence that stock prices do not fully reflect the
implications of current earnings for future earnings. Journal of Accounting and Economics,
vol. 13, pp. 305-340.
Dechow, P.M. 1994. Accounting earnings and cash flows as measures of firm performance:
The role of accounting accruals. Journal of Accounting and Economics, vol. 18, pp. 3-42.
Dechow, P.M., W. Ge, and C.M. Schrand. 2010. Understanding earnings quality: A review
of the proxies, their determinants and their consequences. Journal of Accounting and
Economics, vol.50 , pp. 344-401.
Dechow, P.M., and C.M. Schrand. 2004. Earnings Quality. Charlottesville, VA: The
Research Foundation of the CFA Institute.
Dechow, P.M., R.G. Sloan, and A. Sweeny. 1995. Detecting earnings management. The
Accounting Review, vol. 70, pp. 193-225.
Ertimur, Y., J. Livnat, and M. Martikainen. 2003. Differential market reaction to revenue
and expense surprise. Review of Accounting Studies, vol. 8, pp. 185-211.
Fairfield, P.M., S. Ramnath and T.L. Yohn. 2009 “Do industry-level analyses improve
forecasts of financial performance?” Journal of Accounting Research, forthcoming.
Fama, E. F., and J. MacBeth. 1973. Risk, return, and equilibrium: Empirical tests. Journal of
Political Economy, vol. 81, pp. 607-636.
Francis, J., P. Olsson, and K. Schipper. 2006. Earnings quality. Foundations and Trends in
Accounting, vol. 1, no. 4, pp. 259-340.
Freeman, R., J.A. Ohlson, and S.H. Penman. 1982. Book rate of returns and the prediction of
earnings changes. Journal of Accounting Research, vol. 20, pp. 639-653.
Gu, Z., and J. Wu. 2003. Earnings skewness and analyst forecast bias. Journal of Accounting
and Economics, vol. 35, pp. 5–29.
23
Jegadeesh, N., and J. Livnat. 2006. Revenue surprises and stock returns. Journal of
Accounting and Economics, vol. 41, pp. 147-171.
Jones, J.J. 1991. Earnings management during import relief investigations. Journal of
Accounting Research, vol. 29, pp. 193-228.
Kothari, S.P., J.L. Leone, and C.E. Wasley. 2005. Performance matched discretionary
accrual measures. Journal of Accounting and Economics, vol. 39, pp. 193-197.
Lev, B. 1989. On the usefulness of earnings and earnings research: Lessons and directions
from two decades of empirical research. Journal of Accounting Research, vol. 27, pp. 153192.
Lev, B., and R. Thiagarajan. 1993. Fundamental information analysis. Journal of Accounting
Research, vol. 31, pp. 190-215.
Lipe, R.C. 1986. The information contained in the components of earnings. Journal of
Accounting Research, vol. 24, pp. 37-64.
Nissim, D., and S.H. Penman. 2001. Ratio analysis and equity valuation: From research to
practice. Review of Accounting Studies, vol. 6, pp. 109-154.
Ohlson, J.A.1980. Financial ratios and the probabilistic prediction of bankruptcy. Journal of
Accounting Research, vol. 18, pp. 109-131.
Ohlson, J.A., and S.H. Penman.1992. Disaggregated accounting data as explanatory
variables for returns. Journal of Accounting, Auditing and Finance, vol. 7, pp. 553-573.
Ou, J.A., and S.H. Penman. 1989. Financial statement analysis and the prediction of stock
returns. Journal of Accounting and Economics, vol. 11, pp. 295-329.
Penman, S.H. 1998. Combining earnings and book value in equity valuation. Contemporary
Accounting Research, vol. 15, pp. 291-324.
Penman, S.H. and X. Zhang. (2006). Modeling sustainable earnings and P/E ratios using
financial statement information. Working Paper, Columbia University and University of
California, Berkeley.
Ramakrishnan, R.T.S., and J.K. Thomas. 1998. Valuation of permanent, transitory, and
price-irrelevant components of reported earnings. Journal of Accounting, Auditing and
Finance, vol. 13, pp. 301-336.
Rayburn, J. 1986. The association of operating cash flow and accruals with security returns.
Journal of Accounting Research, vol. 24, pp. 112-133.
Schilit, H., and J. Perler. 2010. Financial Shenanigans: How to Detect Accounting Gimmicks
& Fraud in Financial Reports, McGraw Hill (3rd Edition).
24
Soliman, M.T. 2008. The use of DuPont analysis by market participants. The Accounting
Review, vol. 83, pp. 823-853.
Sloan, R.G. 1996. Do stock prices fully reflect information in accruals and cash flows about
future earnings? The Accounting Review, vol. 71, pp. 289-315.
Weiss, D. 2010. Cost behavior and analysts’ earnings forecasts. The Accounting Review,
vol. 85, pp. 1441–1471.
Wilson, G.P. 1987. The incremental information content of the accrual and funds
components of earnings after controlling for earnings. The Accounting Review, vol. 62, pp.
293-322.
25
Table 1
Sample Selection*
Year
Full
Sample
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
3,148
3,880
4,242
4,491
4,612
4,714
4,955
5,252
5,229
5,298
5,480
5,562
5,737
6,069
6,192
6,021
5,701
5,224
5,988
6,203
Analysts'
Earnings
Forecasts
1,763
2,279
2,452
2,738
2,907
3,012
3,231
3,611
3,695
3,731
3,789
3,733
3,966
4,433
4,813
4,693
4,457
4,127
4,653
4,815
,
Observations
Companies
103,998
3,804
72,898
3,336
*Note: The table presents the number of quarterly observations for each year in our sample.
The initial sample includes all observations with complete financial data on Compustat and
stock returns on CRSP, with market value of equity above $10 million at quarter-end. We
exclude financial institutions (1-digit SIC = 6) and public utilities (2-digit SIC = 49). We
also remove the extreme 1% of observations (on each side) for each variable. Finally, we
remove firms with less than eight quarterly observations and 2-digit SIC industries in
quarters with less than five active firms. The analysts' earnings forecast sample includes
observations with available forecast data on IBES.
26
Table 2
Descriptive Statistics and Correlations*
Panel A: Descriptive statistics
Variable
N
Mean
Std. Dev.
25th Pctl.
Median
75th Pctl.
Sales
NI
NPM
ABS(FCORE)
ABS(ICORE)
ABS(FNCORE)
ABS(INCORE)
FINT
IINT
AR (SW)
AR (PRFILE)
AR (POSTFILE)
MV
BM
FE
103,998
103,998
103,998
103,998
103,998
103,998
103,998
103,998
103,998
103,998
103,998
103,998
103,998
103,998
72,898
415.64
22.31
0.03
25.75
22.93
14.68
16.13
0.61
0.57
0.00
0.00
0.00
1,890.74
0.62
-0.00003
1,057.66
80.78
0.18
72.71
64.11
42.45
45.35
0.26
0.25
0.07
0.05
0.09
5,336.81
0.44
0.00598
32.00
0.46
0.01
1.53
1.44
0.84
1.05
0.42
0.39
-0.04
-0.02
-0.05
99.99
0.32
-0.00051
102.23
3.41
0.04
5.33
4.77
2.92
3.63
0.64
0.59
0.00
0.00
0.00
368.28
0.51
0.00020
331.84
16.07
0.08
19.03
16.65
10.74
12.44
0.83
0.76
0.04
0.02
0.04
1,342.42
0.79
0.00134
Panel B: Pair-wise correlations
NI
NI
FINT
IINT
FCORE
ICORE
FNCORE
INCORE
MV
BM
FINT
0.18
0.39
0.37
0.67
0.69
0.36
0.50
0.72
-0.34
0.21
0.45
0.19
-0.10
0.26
0.21
-0.16
IINT
0.13
0.23
0.18
0.34
0.16
0.01
0.11
-0.03
FCORE
ICORE
0.84
0.21
0.08
0.80
0.12
0.17
0.74
0.62
-0.21
0.25
0.68
-0.20
0.15
-0.06
0.77
-0.19
FNCORE
0.42
-0.01
0.11
-0.15
0.22
0.32
0.14
-0.25
INCORE
0.60
0.15
-0.01
0.41
0.01
0.42
0.18
-0.30
MV
0.82
0.13
0.05
0.81
0.72
0.13
0.41
BM
-0.17
-0.16
-0.08
-0.12
-0.11
-0.10
-0.13
-0.21
-0.47
*Note: Panel A presents descriptive statistics for key variables. Panel B presents Pearson
(above diagonal) and Spearman (below diagonal) correlations between selected variables.
Variables are defined as follows (for firm i in period t):
• Sales – Sales revenue;
• NI – Net income;
• NPM – Net profit margin, measured as NI divided by sales;
• ABS(FCORE) – Absolute value of firm-specific core earnings (FCORE). FCORE is
measured as the average NPM in the same quarter over the previous four years, multiplied
by current sales: FCOREit = [(NPMi,t-4 + NPMi,t-8 + NPMi,t-12 + NPMi,t-16)/4]* Salesit;
• ABS(FNCORE)it – Absolute value of firm-specific non-core earnings (FNCORE).
FNCORE = NI-FCORE;
27
• ABS(ICORE)it - Absolute value of industry-based core earnings (ICORE), where industry
is defined as 2-digit SIC codes. For each quarter, we measure the average NPM in each
industry. Then, we measure firm's i core earnings by multiplying the industry profit
⎡
margin by firm i's sales. ICOREit = ⎢ ∑ NI kt
⎣k∈I ( i )
•
•
•
•
•
•
•
•
∑ Sales
k∈I ( i )
kt
⎤
⎥ ∗ Salesit , where I(i) is the set of
⎦
all firms that belongs to industry of firm i.
ABS (INCORE) – Absolute value of industry-based non-core earnings (INCORE);
INCORE = NI – ICORE;
FINT – Firm-specific intensity of core earnings. FINTit =ABS(FCORE)it/[ABS(FCORE)it
+ ABS(FNCORE)it];
IINTit
–
Industry-based
intensity
of
core
earnings.
IINTit
=
ABS(ICORE)it/[ABS(ICORE)it + ABS(INCORE)it];
AR(SW) - 3-day excess buy&hold return around the preliminary earnings announcement
date, calculated as the buy&hold return on the security minus the average buy&hold
return on a portfolio of firms with similar size and B/M;
AR(PREFILE) - excess buy&hold return from two days after preliminary announcement
through one day after filing, calculated as the buy&hold return on the security minus the
average buy&hold return on a portfolio of firms with similar size and B/M;
AR(POSTFILE) –excess buy&hold return from two days after filing through one day
after next preliminary announcement if available, or plus 90 days if next preliminary
announcement is not available. Calculated as the buy&hold return on the security minus
the average buy&hold return on a portfolio of firms with similar size and B/M;
BM – Book-to-market ratio, measured as book value of common equity at quarter-end
divided by market value of common equity;
MV – Market value of common equity at quarter-end (in millions of dollars);
• FE – Analysts' forecast error, measured as reported earnings per share minus mean
consensus analysts' forecasts, deflated by the stock price at the end of the prior quarter.
28
Table 3
The Persistence of Core and Non-Core Components of Earnings*
Reg. Dependent
Model Variable
1a
NI
Adj-R2
N
0.34
103,998
-3.03
(-0.98)
persistence
coefficient (α1)
0.18
(5.95)***
-4.09
(-2.23)**
0.66
(57.45)***
-2.35
16.06
(-5.42)*** (5.08)***
0.01
0.68
(5.11)*** 103,998
3.48
-17.82
(2.87)*** (-5.10)***
0.01
0.28
(2.85)*** 103,998
Intercept
CV(NI)
BM
MV
-1.04
(-1.25)
12.90
(2.82)***
0.01
(1.23)
1b
FCORE
1c
FNCORE
-0.55
(-0.20)
0.05
(1.84)*
1d
ICORE
-0.64
(-0.35)
0.35
(30.59)***
0.09
(0.14)
7.01
(2.33)**
1e
INCORE
-0.11
(-0.37)
0.20
(2.38)**
-0.83
(-1.03)
6.93
(1.55)
0.01
0.46
(7.27)*** 103,998
0.00
(0.21)
0.24
103,998
*Notes:
1.
The table presents average coefficients (and corresponding t-statistics in parentheses) of
earnings persistence (α1) for net income (NI), the core and non-core components of net
income. To obtain coefficients of persistence for each firm, we estimate the following
regression models on a firm-by-firm basis, and compute average coefficients:
NIit = α 0i + α1i NI i ,t −4 + α 2iCV ( NI )i ,t + α3i BMi ,t + α 4i MVi ,t + ε i ,t
(1a)
FCORE i ,t = α 0i + α1i FCORE i ,t −4 + α 2i CV ( NI ) i ,t + α 3i BM i ,t + α 4i MVi ,t + ε i ,t
(1b)
ICORE i ,t = α 0 i + α 1i ICORE i ,t − 4 + α 2 i CV ( NI ) i ,t + α 3i BM i ,t + α 4 i MVi ,t + ε i ,t (1c)
FNCOREi ,t = α 0i + α1i FNCOREi ,t −4 + α 2i CV ( NI ) i ,t + α 3i BM i ,t + α 4i MVi ,t + ε i ,t (1d)
INCOREi ,t = α 0i + α1i INCOREi ,t −4 + α 2i CV ( NI )i ,t + α 3i BM i ,t + α 4i MVi ,t + ε i ,t (1e)
2.
See Table 2 for definitions of variables.
3.
***,**,* – Denote significant levels from zero at the 0.01, 0.05, and 0.10 levels,
respectively.
29
Table 4
The Effect of Core Intensity on the Persistence of Earnings*
Panel A: Full sample regression analysis
Spec.
Intercept
Dt-4
NIt-4
-0.11
0.36
1
(-0.37)
(16.91)***
Dt-4*NI t-4 CV(NI)
BM
-0.10
-0.26
(-1.45) (-0.56)
MV
Adj-R2
0.01
0.76
(23.75)*** 103,998
2 FINT
1.16
(3.06)***
-2.00
0.28
0.23
-0.07
(-5.80)*** (13.28)*** (13.25)*** (-1.38)
-0.84
0.01
0.77
(-1.85)* (20.60)*** 103,998
3 IINT
0.46
1.62
-1.42
0.31
0.17
-0.10
(-5.08)*** (14.72)*** (11.25)*** (-1.59)
-0.83
(-1.89)*
0.01
23.90***
0.77
103,998
Panel B: Quintile Analysis
Average persistence
coefficient ( γ 1 )
Quintiles based on
core intensity in t-4
Avg N = 20,800
FINTt-4
IINTt-4
All
1
2
3
4
5
5-1
0.36***
0.14***
0.38***
0.57***
0.64***
0.73***
0.59***
0.36***
0.17***
0.43***
0.53***
0.59***
0.59***
0.42***
*Notes:
1. The table presents results for the effect of core earnings intensity on the persistence of
NI. We estimate quarterly regressions models for firm-specific intensity (FINT) and
industry-based intensity (IINT).
2. In Panel A we estimate regression model (2a) with the full sample, and present average
coefficients and t-statistics (in brackets) as in Fama and MacBeth (1973).
NIit = β0t + β1t Di ,t −4 + β 2t NIi ,t −4 + β3t Di ,t −4 NIi ,t −4 + β 4t CV ( NI )it + β5t BMit + β6t MVit + ε it (2a)
3.
4.
5.
6.
In Panel B we estimate regression model (2b), each quarter, in five quintiles, and present
average persistence coefficients ( γ 1 ). Quintiles are assigned according to core earnings
intensity in the same quarter last year (t-4). For each quarter, we sorted all observations
according to their FINT or IINT and assigned the sample observations to quintiles.
NI it = γ 01t + γ 1t NI it −4 + γ 2t CV ( NI ) it + γ 3t BM it + γ 4t MVit + ψ it
(2b)
Dit is a dummy variable that equals “1” if core earnings intensity (FINT or IINT) for
firm i is above the quarterly median at time t, and “0” otherwise.
See Table 2 for definitions of other variables.
***, **, * – Significance levels at the 0.01, 0.05 and 0.10 levels, respectively.
30
Table 5
The Intensity of Core Earnings and the Predictability of Earnings
Panel A: Firm core intensity in period t-4 (FINTt-4)
Quintiles
based on
FINT in t-4
All
1
2
3
4
5
5-1
Accuracy in
period t
ABS(FEt)
Dispersion in
period t
STD(forecasts)t
Bias in
period t
FEt
2.57***
3.40***
3.15***
2.55***
2.04***
1.68***
-1.72***
1.19***
1.63***
1.43***
1.14***
0.93***
0.81***
-0.82***
-0.03
-0.14**
-0.09
-0.03
0.03
0.06*
0.20***
Percentage of loss firms in
period t (NIt < 0)
Full sample
NIt-4 > 0
18.01%
26.37%
24.04%
19.46%
11.65%
8.49%
-17.88%
11.26%
15.24%
17.28%
10.34%
7.45%
5.97%
-9.27%
Panel B: Industry core intensity in period t-4 (IINTt-4)
Accuracy in
Dispersion in
Bias in
Percentage of loss firms
Quintiles
period t
period t
period t
in period t (NIt < 0)
based on
ABS
(FE
)
STD(forecasts)
FE
Full
sample
NIt-4 > 0
t
t
t
IINT in t-4
2.57***
1.19***
-0.03
18.01%
11.26%
All
3.01***
1.44***
-0.03
32.46%
9.88%
1
2.73***
1.25***
-0.10*
21.07%
14.54%
2
2.52***
1.15***
-0.05
14.99%
12.55%
3
2.37***
1.07***
-0.02
11.30%
9.78%
4
2.20***
1.03***
0.00
10.21%
9.56%
5
-0.81***
-0.41***
0.03
-22.25%
-0.32%
5-1
*Notes:
1. The table presents mean forecast accuracy (absolute forecast error), mean forecast
dispersion (standard deviation of forecasts), mean forecast Bias (forecast error), and
percentage of loss-reporting firms in period t. Forecast attributes are multiply by 1,000.
2. Intensity quintiles are assigned according to core earnings intensity in the same quarter
last year (t-4). Firms are sorted each quarter according to their firm-specific (Panel A) or
industry-based (Panel B) intensity of core earnings.
3. The accuracy and the bias samples include 72,898 quarterly observations (14,580 per
quintile). The dispersion sample includes 54,125 quarterly observations due to the
requirement for minimum of three different analysts' forecasts (10,825 per quintile).
4. The sample for computing the percentage of loss firms is 103,998 quarterly observations
for the full sample, and 85,985 quarterly observations for the sample of profitable firms
in quarter t-4 (20,800 and 17,197 per quintile, respectively).
5. See Table 2 for definitions of core earnings intensity.
6. ***, **, * – Significantly different from zero at the 0.01, 0.05 and 0.10 levels,
respectively.
31
Table 6
The Intensity of Core Earnings and the Market Reaction to Earnings*
Panel A: Full sample regression analysis
Specification
1
Intercept
D
FE
0.01
(9.18)***
D × FE
3.60
(16.84)***
Adj-R2
N
0.05
72,898
2
FINT
0.01
(6.40)***
-0.00
(-0.25)
3.21
(15.59)***
1.73
(9.49)***
0.05
72,898
3
IINT
0.01
(6.35)***
0.00
(0.13)
2.90
(14.96)***
2.51
(12.08)***
0.06
72,898
Panel B: Regression analysis over quintiles
Quintiles based on
intensity in period t
Avg N = 14,580
Average market reaction
coefficient ( λ1 )
FINTt
IINTt
3.60***
2.82***
3.49***
4.61***
5.38***
6.10***
3.28***
All
1
2
3
4
5
5-1
3.60***
2.41***
3.28***
4.84***
5.69***
5.58***
3.17***
*Note:
1. The table presents results for the effect of core earnings intensity on market reaction to
unexpected earnings. We estimate quarterly regressions models for firm-specific
intensity (FINT) and industry-based intensity (IINT).
2. In Panel A we estimate regression model (3a) in the full sample, and present average
coefficients and t-statistics as in Fama and MacBeth, 1973.
AR(SW )it = δ 0t + δ1t Dit + δ 2t FEit + δ3t Dit ∗ FEit +ηit
(3a)
3. In Panel B we estimate regression model (3b), each quarter, in five quintiles, and present
average coefficients (as in Fama and MacBeth, 1973). Quintiles are assigned according
to core earnings intensity in the current quarter. For each quarter, we sorted all
observation according to their FINT or IINT and assigned the sample observations to
quintiles.
AR(SW )it = λ0t + λ1t FEit +ηit
(3b)
4. Dit is a dummy variable that equals to 1 if core earnings intensity (FINT or IINT) for
firm i is above the quarterly median at time t, and 0 otherwise. See Table 2 for
definitions of other variables.
5. ***, **, * – Significantly different from zero at the 0.01, 0.05 and 0.10 levels,
respectively
32
Table 7
The Intensity of Core Earnings and the Post Earnings Announcement
Drift*
Panel A: Full sample regression analysis
Dependent
Variable
Specification
Intercept
D
0.001
(0.64)
FE
D × FE
0.914
(11.05)***
Adj-R2
AR (PREFILE)
1a
AR (PREFILE)
1b
FINT
-0.001
(-0.47)
0.002
(2.24)**
0.804
(8.35)***
0.372
(2.10)**
0.004
AR (PREFILE)
1c
IINT
0.000
(0.43)
0.000
(0.12)
0.758
(9.10)***
0.433
(2.38)**
0.004
AR(POSTFILE)
2a
AR(POSTFILE)
2b
FINT
-0.000
(-0.02)
0.002
(1.83)*
0.522
(5.67)***
0.313
(1.77)*
0.003
AR(POSTFILE)
2c
IINT
0.001
(1.03)
-0.000
(-0.63)
0.492
(5.46)***
0.352
(2.07)**
0.003
0.001
(1.01)
0.003
0.617
(7.36)***
0.002
Panel B: Regression analysis over quintiles
Average post earnings announcement drift coefficient (ν 1 )
Quintiles based on
intensity in period t
Avg N = 14,580
All
1
2
3
4
5
AR(PREFILE)
0.914***
0.679***
0.957***
1.196***
1.403***
1.326***
0.914***
0.775***
0.819***
0.973***
1.242***
1.162***
0.617***
0.428***
0.689***
0.804***
1.174***
0.934***
0.617***
0.503***
0.546***
0.677***
0.876***
0.924***
5-1
0.647**
0.387
0.506
0.421
FINT
IINT
AR(POSTFILE)
FINT
IINT
*Notes:
1.
The table presents results for the effect of core earnings intensity on post earnings
announcement drift (PEAD). PEAD is measured as excess buy&hold return from two
days after preliminary announcement through one day after filing [AR(PRFILE)] in
regression models (4a), 4(b), and as excess buy&hold return from two days after filing
33
2.
through one day after next preliminary announcement if available or plus 90 days if next
preliminary announcement is not available [AR(POSTFILE)] in regression models (5a),
(5b).
In Panel A we estimate regression models (4a) and (5a) in the full sample, and present
average coefficients and t-statistics as in Fama and MacBeth, 1973.
AR( PREFILE)it = κ 0t + κ1t Dit + κ 2t FEit + κ3t Dit ∗ FEit +ηit
(4a)
AR( POSTFILE) it = κ 0t + κ1t Dit + κ 2t FEit + κ 3t Dit ∗ FEit + ηit
3.
(5a)
In Panel B we estimate regression models (4b) and (5b), each quarter, in five quintiles,
and present average coefficients (as in Fama and MacBeth, 1973). Quintiles are assigned
according to core earnings intensity in the current quarter. For each quarter, we sorted
all observation according to their FINT or IINT and assigned the sample observations to
quintiles.
AR( PREFILE)it = ν 0t +ν1t FEit +ηit
(4b)
AR( POSTFILE) it = ν 0t +ν 1t FEit + ηit
4.
5.
(5b)
Dit is a dummy variable that equals to 1 if core earnings intensity (FINT or IINT) for
firm i is above the quarterly median at time t, and 0 otherwise. See Table 2 for
definitions of other variables.
***, **, * – Significantly different from zero at the 0.01, 0.05 and 0.10 levels,
respectively.
34
Figure 1
Core intensity over 1990-2009*
Note: The figure presents firm-specific core intensity (FINT), and industry-based core
intensity over 1990-2009
35