Professional Documents
Culture Documents
Abstract. Prior research has shown that pro-forma (recurring operating) earnings reported by managers
and analysts are more value relevant than GAAP net income. Since GAAP net income contains many non-
operating items that reduce its value relevance compared to operating earnings, comparing the value
relevance of GAAP net income with operating earnings unduly favors operating earnings. We show that
operating earnings reported by managers and analysts are more value relevant than a measure of operating
earnings derived from firms’ financial statements, as reported by Standard and Poor’s. Our evidence is
important because it indicates that operating earnings reported by managers and analysts contain value
relevant information beyond that provided by operating earnings obtained by sophisticated users from
firms’ financial statements.
Keywords: operating income, pro-forma earnings, value relevance, predictive ability, valuation,
information content
*Corresponding author.
562 BROWN AND SIVAKUMAR
Consistent with Abarbanell and Lehavy (2002) and Bradshaw and Sloan (2002),
we use Thomson Financial I/B/E/S data to measure operating income disclosed in a
firm’s earnings release (hereafter STREET).3 We use Standard and Poor’s (S&P)
data to measure EPS from operations obtained from 10-Q and 10-K filings to the
SEC (Compustat item Q177, hereafter EPSOP).4 Since the early 1980s I/B/E/S has
provided operating earnings that it obtains from earnings announcements appearing
on newswires as the actual EPS of firms.5 S&P Compustat began providing operating
earnings to the marketplace in 1988. Compustat documentation does not disclose the
source of its measure of operating earnings (Q177), but S&P personnel informed us
that the 10-Q and 10-K reports are the sole source and that EPSOP is not determined
via a formula such as GAAP adjusted for special items.6
We use the three procedures suggested by Aboody and Lev (1998) to assess value
relevance: (1) ability to predict future earnings [predictive ability], (2) association of
earnings levels with stock price levels [valuation], and (3) correlation of earnings
surprises with abnormal stock returns [information content]. Using quarterly data
from 1989 to 1997, we show that operating income provided by managers and
analysts is more value relevant than operating income obtained from 10-Q and 10-K
filings to the SEC under all three procedures.7 Our results suggest that operating
earnings reported by managers and analysts contain value relevant information
beyond that provided by operating earnings obtained by sophisticated users,
exemplified by S&P Compustat, from firms’ financial statements.
We proceed as follows. Section 1 describes our sample. We discuss our
methodologies and present results in Section 2. We summarize and interpret our
results in Section 3.
1. Sample
Each firm-quarter observation in our sample must have I/B/E/S and Compustat data.
We conduct predictive ability tests using a seasonal random walk (SRW) model (i.e.,
expected earnings in quarter t equal those in quarter t-4) so STREET and EPSOP must
be available in quarter t and t-4. We impose a CRSP data requirement only for
information content tests. Our sample period begins in 1989 because EPSOP is
unavailable prior to 1988. It ends in 1997 because we require our data to be stated in
primary EPS and beginning in 1998, most I/B/E/S data are stated as fully diluted EPS.8
We examine all firm-quarters when STREET and EPSOP differ. We classify
observations into three groups based on comparisons of the operating income
measures with GAAP (Compustat item Q19):
The three groups have simple interpretations. The first represents cases where
I/B/E/S, but not S&P, backs out ‘‘non-recurring items’’ from GAAP. The second
COMPARING THE VALUE RELEVANCE OF TWO OPERATING INCOME MEASURES 563
consists of cases where both I/B/E/S and S&P back out ‘‘non-recurring items’’ from
GAAP, but disagree on their amount. The third contains observations where S&P,
but not I/B/E/S, backs out ‘‘non-recurring items’’ from GAAP. We omit cases where
the two data sources agree because we seek to ascertain which source provides the
more value relevant earnings numbers. Sample sizes vary across partitions and tests
and are noted in the tables.
Each of the three approaches we use to evaluate value relevance has strengths and
weaknesses. A strength of the predictive analysis is that earnings predictions are
important inputs for valuation models (Ohlson, 1995; Frankel and Lee, 1998); a
weakness is that simply reporting the same earnings number each quarter trivially
maximizes predictive validity. A strength of the valuation analysis is the use of stock
price information without requiring an ‘‘event window;’’ a weakness is the
assumption that stock price is in equilibrium with respect to earnings. A strength
of the information content analysis is the use of a short window, making it most
suitable for capturing the contemporaneous information in earnings; a weakness
include the potential for contaminated events (Brown and Kim, 1993). The
advantage of using three approaches is that the validity of our overall findings is
enhanced if we obtain similar results with each method.
We conduct two predictive ability tests: a parametric (means) test and a non-
parametric (binomial) test. Prediction error is defined as the absolute value of the
difference between actual quarterly earnings and its expectation, divided by
beginning of quarter stock price, where the expectation is the quarter t-4 (SRW)
forecast using STREET and EPSOP respectively.9 The means test identifies the
operating income measure with the smallest prediction error, on average. The
binomial test identifies the operating income measure that most often has the
smallest prediction error.
We conduct predictive ability tests on two samples, each requiring STREET to
differ from EPSOP in quarter t-4. Our first sample requires STREET and EPSOP to
agree in quarter t. It has the advantage of mitigating the ambiguity concerning
measurement of quarter t earnings numbers, so differences in predictive accuracy are
attributable to differences in predictors, not actuals. However, its findings may not
generalize to all of our data so we also examine a second sample, which requires
STREET and EPSOP to disagree in quarter t. We examine the ability of each of the
operating earnings (EPSOP and STREET) in quarter t-4 to predict itself in quarter t
in both samples.10
564 BROWN AND SIVAKUMAR
Panel A of Table 1 provides predictive ability results for cases where STREET and
EPSOP differ in quarter t-4, but agree in quarter t. Panel B shows results for cases
where STREET and EPSOP differ in quarter t-4 and t. Panel A shows that STREET
predicts significantly more accurately than EPSOP in group 1, but that EPSOP
predicts significantly more accurately than STREET in group 3. This result is
consistent with the notion that when either I/B/E/S or S&P (but not both) back out
items they deem to be non-recurring, the ‘‘backing out’’ facilitates predictions of
future earnings. When both sources back out non-operating items but disagree as to
their amount (group 2), STREET predicts better than EPSOP. This result is
consistent with the notion that managers and analysts do a better job than S&P in
identifying the magnitude of non-recurring items when they agree that the GAAP
number needs to be adjusted. For all three groups combined, managers and analysts
report an operating income measure that predicts significantly better than S&P
nearly 56% of the time. This result is consistent with the notion that managers and
analysts, on average, do a better job than does S&P in identifying non-recurring
items.
Table 1. Predictive ability evidence. t-statistics of tests that EPSOP is less accurate than STREET
and percent of time that EPSOP is less accurate than STREET (in parentheses).
EPSOP EPSOP
Sample versus Sample versus
Nature of Disagreement Size STREET Size STREET
2.2. Valuation
Pi,t is market value per share defined as the closing share price three months after the
fiscal quarter end (Compustat monthly item PRCCM); BVi,t is common equity per
share at quarter end (Compustat items Q59/Q15); OPINCi,t equals EPSOPi,t or
STREETi,t; and NI is the ‘‘bottom-line’’ quarterly net income per share number
(Compustat item Q11).
We compare the adjusted R-squares using alternatively STREET or EPSOP in
equation (1), and consider the operating income measure with the most explanatory
power to be the more value relevant one. We use a Vuong test to determine whether
one valuation equation has a significantly larger adjusted-R-square. We also
compare coefficient estimates on the OPINC variable and use a t-test to determine
whether one operating income measure has a larger coefficient in the valuation
equation.12
Table 2 presents valuation results, showing regression slopes (betas) and adjusted R-
squares of the valuation equations for both operating income measures for each
group and for all three groups combined. Based on Vuong tests, STREET is
significantly more value relevant than EPSOP in groups 1 and 2 and for all three
groups combined (Vuong Z ¼ 5.17, 3.11 and 2.79 respectively). In contrast, there are
566 BROWN AND SIVAKUMAR
Table 2. Valuation evidence. Estimated coefficients and adjusted R2s from valuation regressions of stock
price on book value and operating earnings measures.
Model: Pi,t ¼ bV0 þ bV1 BVi,t þ bV2 OPINCi,t þ bV3 ðNIi,t OPINCi,tÞ þ ei;t where
Pi,t: Price per share three months after the fiscal quarter end (Compustat item PRCCM).
BVi,t: Common equity per share at end of quarter t (Compustat item Q59/Compustat item Q15).
OPINC: Operating income is measured either as:
EPSOP: Operating income reported by Compustat (Q177), or
STREET: Operating income reported by Thomson Financial I/B/E/S.
NI-OPINC: Net income reported by Compustat (Q11) – Operating income measure EPSOP or STREET.
Period: 1989–1997.
Groups: The three mutually exclusive and collectively exhaustive groups are defined based on the
agreement or disagreement of the operating income measures (EPSOP and STREET) with GAAP
(earnings from continuing operations reported by Compustat Q19).
*Significant at the 1% level (two-tailed tests).
no significant differences for group 3 (Vuong Z ¼ 0.28). The t-tests provide similar
results. STREET is significantly more value relevant than EPSOP in groups 1 and 2
and for all three groups combined (t-values are 4.81, 4.92 and 5.14 respectively). In
contrast, there are no significant differences for group 3 ðt-value ¼ 0.72Þ. The
coefficient estimates for OPINC using STREET are 12.8%, 22.9% and 11.4% larger
than the coefficients using EPSOP for groups 1 and 2 and for all three groups
combined. With the exception that the results for group 3 are insignificant, the
valuation results mirror the predictive ability results.
COMPARING THE VALUE RELEVANCE OF TWO OPERATING INCOME MEASURES 567
We examine information content using both a three-day and a 63-day window. Our
regression model is:
CARi;t ¼ bIC0 þ bIC1 UOPINCi;t þ ei;t ; ð2Þ
CARi,t is the three-trading days or 63-trading days market-adjusted return for firm i
for days 1 to þ 1 and 1 to þ 61 respectively (source: CRSP), where day 0 is the
earnings announcement day (source: Compustat) and UOPINCit is the unexpected
operating income for firm i in quarter t, defined as operating income minus expected
operating income divided by the beginning of quarter share price. We alternatively
use STREET or EPSOP as our operating income measure. The last mean consensus
estimate in the I/B/E/S Summary file prior to the quarterly earnings announcement is
our proxy for expected operating income. The measure with the most explanatory
power in the OLS regression is considered to be the most value relevant. We use a
Vuong test to determine whether one specification has a significantly larger adjusted
R-square and a t-test to determine whether one slope coefficient is significantly larger
than the other.
than is the S&P number. To mitigate this problem, we replicate our analysis using a
63-trading-day window, 1 to þ 61. This larger window should include the period
when firms’ 10-Qs and 10-Ks are filed. The results appear in Table 4.
The Vuong test results for the 63-trading day window differ somewhat from the
three-day window results. The relative magnitudes of the adjusted R2s for all groups
and for the three groups combined are similar to the three-day tests, but the Vuong
COMPARING THE VALUE RELEVANCE OF TWO OPERATING INCOME MEASURES 569
Table 4. Information content evidence (63-days). Estimated coefficients and adjusted R2s
from information-content regressions of 63-day CARs around quarterly earnings
announcements (1 to þ61) on unexpected operating earnings.
test statistics are insignificant for group 2 and for all three groups combined (Vuong
Z ¼ 0.28 and 0.54). However, using the t-test for comparison of response
coefficients, the 63-day window results mirror the three-day window results.
STREET is significantly more value relevant than EPSOP in groups 1 and 2 and
for all three groups combined (t-values are 5.55, 1.72 and 3.83, respectively). Once
570 BROWN AND SIVAKUMAR
again, there are no significant differences for group 3 (t-value ¼ 1.45). The
coefficient estimates using STREET are 74%, 59% and 43% larger than coefficients
using EPSOP for groups 1 and 2 and for all three groups combined, respectively.
Overall, our information content results are similar to the valuation results.
We compare the value relevance of two operating income measures, one provided by
managers and analysts (source: Thomson Financial I/B/E/S), and one obtained from
official filings with the SEC (source: Standard and Poor’s Compustat). We use three
techniques to assess value relevance: predictive ability, valuation and information
content. We find that under all three methods the operating income reported by
managers and analysts is more value relevant than the one obtained by Standard and
Poor’s Compustat, a sophisticated user of firms’ financial statements.
One interpretation of our results is that the operating income reported by
managers and analysts has fewer transitory components than operating income
obtainable from firms’ financial statements, placing our paper in the tradition of
research showing that permanent earnings are more value relevant than transitory
earnings (Lipe, 1986; Elliott and Shaw, 1988; Elliott and Hanna, 1996; Francis et al.,
1996; Ali and Zarowin, 1992; Ramakrishnan and Thomas, 1998). A second
interpretation, which is not mutually exclusive of the first, is that managers and
analysts seek to provide value relevant information to the marketplace through their
operating earnings measures.
We close with some caveats. First, we interpret our findings based on an
assumption that capital markets are efficient. If capital markets are inefficient, our
results indicate that investors may erroneously focus too much on operating earnings
numbers reported by managers and analysts. Second, our sample period ends in
1997, so our results may not pertain to a more recent time period. Third, we examine
the value relevance of an operating income number that is measured by only one
sophisticated user of 10-Q and 10-K reports, Standard and Poor’s. Other
sophisticated users of financial statements may measure operating income
differently, yielding numbers that may be more value relevant than those reported
by managers and analysts.
Acknowledgments
Notes
1. When he was chief accountant of the SEC, Lynn Turner referred to operating earnings reported by
managers as ‘‘EBS’’ or ‘‘Everything but Bad Stuff’’ (Turner, 2000), counseling investors to be wary of
operating numbers disclosed in firms’ earnings releases, and suggesting they peruse Form 10-Qs,
‘‘where hopefully all the facts are presented in a complete and balanced fashion.’’ The financial press
concurs: ‘‘Companies report ‘pro forma’ earnings that are deceptive, unwarranted, and downright
dangerous to the financial system . . . GAAP [is] the most consistent and objective way to compare
results across companies and industries [and that] can’t be said of pro forma’’ (Henry, 2001).
2. See http://www.sec.gov/news/headlines/proforma-fin.htm.
3. I/B/E/S (2001, p. 7) states: ‘‘Earnings from operations means diluted earnings excluding all
extraordinary items (specifically, those items defined by the accountants as extraordinary such as
cumulative effect of an accounting change, early debt redemption, etc.), and excluding certain non-
recurring or non-operating items (but not extraordinary by accounting definition) that a majority of
the contributing analysts want to exclude (usually footnote items such as most restructuring charges,
acquisition charges, or asset sales gains or losses).’’
4. According to Compustat, EPSOP is: ‘‘Earnings per Share (Primary) adjusted to remove the effect of
all special items from the calculation. This earnings per share item excludes the effect of all
nonrecurring events. This item excludes: 1. Cumulative effect of accounting change, 2. Discontinued
operations, 3. Extraordinary items, 4. Special items.’’
5. I/B/E/S (1999) states: ‘‘I/B/E/S strives to report actual earnings as soon as they are released into the
marketplace. For the US and Canada, earnings reports are culled directly from the newswires,
adjusted for comparability with estimates (page 4). With very few exceptions [I/B/E/S] receives an
analyst’s forecast after discontinued operations, extraordinary charges and other non-operating items
have been backed out. While this is far and away the best method for valuing a company, it often
causes a discrepancy when a company reports earnings. I/B/E/S adjusts reported earnings to match
analysts’ forecasts on both an annual and quarterly basis. This is why I/B/E/S actuals may not agree
with other published actuals; i.e. Compustat (pp. 5–6).’’
6. While both I/B/E/S and S&P maintain that they adjust for non-recurring items, they sometimes adjust
for recurring items (Johnson and Schwartz, 2001). We use the terms non-operating and non-recurring
in the same way that I/B/E/S and S&P do even though the numbers ‘‘backed-out’’ may be recurring
items. We use the terms non-recurring and non-operating synonymously.
7. We use GAAP as a partitioning variable but we do not compare either operating income measure with
GAAP. In a previous version of our paper, we find that both operating income measures are more
value relevant than GAAP (Compustat item Q19).
8. Compustat provides only primary EPS for EPSOP. To avoid problems of converting fully diluted to
primary EPS numbers (or vice versa), we require the I/B/E/S numbers to be stated in primary form.
Imposition of this requirement results in the loss of, at most, 5% of our data in any year in our sample.
I/B/E/S changed its procedures in 1998 when SFAS 128 became effective. We obtain similar results
when we include 1998 data.
9. We obtain similar results using actual earnings as a deflator.
572 BROWN AND SIVAKUMAR
10. An alternative sample is one in which STREET and EPSOP agree in quarter t 4, but disagree in
quarter t. We obtain qualitatively similar results for this sample.
11. We focus our attention on V2 because our interest is the coefficient on the operating earnings
measures (STREET or EPSOP). We do not discuss the coefficients on book value ðV1Þ or V3 because
we view them as control variables.
12. We use the more conservative approximation technique based on unequal variances. More specifically,
we estimate the t-value by dividing the difference between the two coefficient estimates by the square
root of the summation of the variances of the two coefficient estimates.
References