Are you sure?
This action might not be possible to undo. Are you sure you want to continue?
washing
ton.edu
Journal of Accounting and Economics 24 (1997) 301—336
Does EVA beat earnings?
Evidence on associations with stock returns and
ﬁrm values
Gary C. Biddleº', Robert M. Bowenº*, James S. Wallace°
º School of Business Administration, University of Washington, Seattle, WA 981953200, USA
' School of Business and Management, Hong Kong University of Science and Technology, Clear Water
Bay, Kowloon, Hong Kong, China
° Graduate School of Management, University of California, Irvine, CA 926973125, USA
Received 1 October 1996
Abstract
This study tests assertions that Economic Value Added (EVA) is more highly
associated with stock returns and ﬁrm values than accrual earnings, and evaluates which
components of EVA, if any, contribute to these associations. Relative information
content tests reveal earnings to be more highly associated with returns and ﬁrm values
than EVA, residual income, or cash ﬂow from operations. Incremental tests suggest that
EVA components add only marginally to information content beyond earnings. Con
sidered together, these results do not support claims that EVA dominates earnings in
relative information content, and suggest rather that earnings generally outperforms
EVA. 1997 Elsevier Science B.V. All rights reserved.
JEL classiﬁcation: M41; G14
Keywords: Valuerelevance; Relative information content; Incremental information con
tent; Firm market value; Economic value added (EVA); Residual income; Economic
proﬁts; Earnings; Cash from operations; Charge for capital
1. Introduction and motivation
For centuries, economists have reasoned that for a ﬁrm to create wealth it
must earn more than its cost of debt and equity capital (Hamilton, 1777;
01654101/97/$17.00 1997 Elsevier Science B.V. All rights reserved.
PII S 0 1 6 5  4 1 0 1 ( 9 8 ) 0 0 0 1 0  X
¹ Residual income is generally deﬁned as aftertax operating proﬁts less a charge for invested
capital. Operating proﬁts are proﬁts before deducting the aftertax cost of interest expense. The
ﬁrm’s weighted average cost of debt and equity capital is deducted in the capital charge. Other labels
include: abnormal earnings (Feltham and Ohlson, 1995); excess earnings (Canning, 1929, Preinreich,
1936, 1937, 1938); excess income (Kay, 1976; Peasnell, 1981, 1982); excess realizable proﬁt (Edwards
and Bell, 1961); and superproﬁts (Edey, 1957).
` Stern Stewart & Company is a New Yorkbased consulting ﬁrm that markets the ‘EVA
Financial Management System’ for internal and external performance measurement and incentive
compensation. Performance measures marketed by competing ﬁrms include cashﬂow return on
investment (CFROI) by Boston Consulting Group’s HOLT Value Associates, discounted cashﬂow
analysis (DCA) by Alcar, discounted economic proﬁts (EP) by Marakon Associates, and economic
value management (EVM) by KPMG Peat Marwick.
Marshall, 1890). In the twentieth century, this concept has been operationalized
under various labels including residual income.¹ Residual income has been
recommended as an internal measure of businessunit performance (Solomons,
1965) and as an external performance measure for ﬁnancial reporting (Anthony,
1973, 1982a,b). General Motors applied this concept in the 1920s and General
Electric coined the term ‘residual income’ in the 1950s and used it to assess the
performance of its decentralized divisions (Stern Stewart EVA Roundtable,
1994).
More recently, Stern Stewart & Company has advocated that a trademarked
variant of residual income, economic value added (EVA), be used instead of
earnings or cash from operations as a measure of both internal and external
performance.` They argue: “Abandon earnings per share” (Stewart, 1991) (p. 2).
“Earnings, earnings per share, and earnings growth are misleading measures of
corporate performance” (Stewart, 1991), (p. 66). “The best practical periodic
performance measure is economic value added (EVA)” (Stewart, 1991 (p. 66).
“Forget EPS, ROE and ROI. EVA is what drives stock prices” (Stern Stewart
advertisement in Harvard Business Review, November—December, 1995, p. 20).
Stewart (1994) cites inhouse research indicating that “EVA stands well out from
the crowd as the single best measure of wealth creation on a contemporaneous
basis” and “EVA is almost 50% better than its closest accountingbased com
petitor in explaining changes in shareholder wealth” (p. 75).
This study provides independent empirical evidence on the information
content of EVA, residual income, and two mandated performance measures,
earnings and cash ﬂow from operations. Our inquiry is motivated by: the claims
cited above, interest in EVA in the business press, increasing use of EVA by
ﬁrms, increasing interest in EVA among academics, and potential interest in
EVA among accounting policy makers. Citations of EVA in the business press
have grown exponentially, rising from 1 in 1989 to 294 in 1996 (Lexis/Nexis
‘allnews’ library). Fortune has touted EVA as “The Real Key to Creating
Wealth” (30 September 1993), “A New Way to Find Bargains” (9 December
1996), and has begun augmenting its wellknown ‘500’ ranking with an annual
302 G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
` CFO Basil Anderson of Scott Paper states: ‘‘We used to have diﬀerent ﬁnancial measures for
diﬀerent purposes — discounted cash ﬂow for capital decisions, another measure for rewarding
performance and the like.
2
Now EVA is one measure that integrates all that.
2
it oﬀers an
excellent link to the creation of shareholder value’’ (Walbert, 1994) pp. 111—112. Jim Meenan, CFO
of AT&T’s communications services group expresses a similar view: ‘‘Every decision is now based on
EVA. The motivation of our business units is no longer just to make a proﬁt. The drive is to earn the
cost of capital.
2
when you drive your business units toward EVA, you’re really driving the
correlation with market value’’ (Walbert, 1994) (p. 112). Eugene Vesell, managing director of
Oppenheimer Capital states: ‘‘The ﬁrst thing we look at when we pick companies is, are they
motivated by EVA? We prefer it to measures like EPS or return on equity.’’ (Tully, 1994) (p. 143).
" We emphasize that our results are only an input to the policy making process. Each of the
measures we consider may have value in other decision contexts, e.g., cash from operations may
provide valuable information to lenders and suppliers about liquidity. Questions regarding cost and
best source(s) of data are beyond the scope of this research.
‘Performance 1000’ based on data from Stern Stewart (Tully, 1993, 1994; Fisher,
1995; Lieber, 1996; Teitelbaum, 1997).
Companies that have adopted EVA for performance measurement and/or
incentive compensation include AT&T, Coca Cola, Eli Lilly, Georgia Paciﬁc,
Polaroid, Quaker Oats, Sprint, Teledyne and Tenneco. The ‘EVA Financial
Management System’ is alleged to encourage managers to act more like owners
by helping managers make improved operating, ﬁnancing and investment
decisions.` Evidence provided in Wallace (1997) suggests that managers com
pensated on the basis of EVA (instead of earnings) take actions consistent with
EVAbased incentives.
Recently, academics have shown interest in models of equity valuation that
express ﬁrm value in terms of book value and the expected stream of residual
income or ‘abnormal earnings’ (Ohlson, 1995; Feltham and Ohlson, 1995). Our
study provides empirical evidence on whether current period realizations of
residual income (RI) and EVA are more closely associated with stock returns
than are traditional accounting measures such as earnings and cash from
operations (CFO).
Finally, data on the information content of EVA and RI provide potentially
useful input to the normative policy debate on what performance measure(s)
should be reported in ﬁnancial statements." Financial reporting has been
criticized for lowquality and lack of relevance in today’s informationrich
environment. The AICPA Special Committee on Financial Reporting (1994),
the Jenkins Committee, makes suggestions for improving ﬁnancial reporting
that are consistent with ﬁrms using EVA for internal decision making and
external reporting. A prediction from an April 1995 AICPA workshop on the
future of ﬁnancial management is that EVA will replace EPS in ¹he ¼all Street
Journal’s regular stock and earnings reports (Zarowin, 1995) (p. 48). Widespread
interest in revisiting the quality of ﬁnancial reporting suggests that altern
atives to currently mandated performance measures should be evaluated for
G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336 303
valuerelevance. This study provides evidence that we hope will be useful to
policy makers who may be interested in EVA or RI as replacements (or
complements) to earnings and CFO as key measures of ﬁrm performance.
The ﬁrst (of two) empirical questions we address is
Q1: Do EVA and/or RI dominate currently mandated performance
measures, earnings and operating cash ﬂow, in explaining contempor
aneous annual stock returns?
This relative information content question examines which variables (EVA, RI,
CFO or earnings) have a greater association with contemporaneous stock
returns and provides a direct test of one of Stern Stewart’s claims about the
superiority of EVA. (In Section 5.5 we examine separately another Stern Stewart
claim that EVA outperforms earnings in explaining ﬁrm values.) Using a sample
of 6,174 ﬁrmyears representing both adopters and nonadopters of EVA over
the period 1984—1993, tests of question 1 indicate that earnings (R`"12.8%) is
signiﬁcantly more highly associated with marketadjusted annual returns than
are RI (R`"7.3%) or EVA (R`"6.5%) and that all three of these measures
dominate CFO (R`"2.8%). This ﬁnding is supported across a number of
alternative speciﬁcations.
Second, we examine whether EVA and/or RI complement currently mandated
performance measures, earnings and CFO:
Q2: Do components unique to EVA and/or RI help explain contempor
aneous stock returns beyond that explained by CFO and earnings?
This is equivalent to asking: Does the market appear to value a given EVA
component beyond the information contained in the other components? To
address this incremental information content question, we decompose EVA into
components (e.g., cash from operations, operating accruals, capital charge, and
accounting ‘adjustments’) and evaluate the contribution of each component
toward explaining contemporaneous stock returns. For the full sample, while
each component is signiﬁcantly associated with marketadjusted returns, the
EVA components do not appear to be economically signiﬁcant. Further, tests
across alternative speciﬁcations indicate that, while cash ﬂow and accrual
components are consistently signiﬁcant, components unique to EVA (capital
charge and accounting adjustments) are typically not signiﬁcant. Considering
the relative and incremental information content results together, neither EVA
nor RI appears to dominate earnings in its association with stock market
returns.
The remainder of the paper is organized as follows. Section 2 provides
a description of EVA and its components, presents hypotheses, and describes
statistical tests for relative and incremental information content. Section 3
304 G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
reports sample selection criteria, variable deﬁnitions, and descriptive statistics.
Section 4 provides empirical results on the relative and incremental information
content of EVA and its components. Section 5 reports various extensions and
sensitivity analyses. We close with a summary and a discussion of potential
factors contributing to the failure of EVA and/or RI to dominate earnings.
2. Components of EVA, hypotheses and statistical tests
2.1. Linkages between operating cash ﬂow, earnings, residual income and E»A
This section describes linkages between operating cash ﬂows (CFO), earnings
before extraordinary items (EBEI), residual income (RI) and economic value
added (EVA). We begin by partitioning earnings into operating cash ﬂows and
accruals:
EBEI"CFO#Accrual,
where
CFO "net cash provided by operating activities.
Accrual "total accruals related to operating (as opposed to investing or
ﬁnancing) activities, e.g., depreciation, amortization, noncash
current assets, current liabilities (other than notes payable and
current portion of longterm debt), and noncurrent portion of
deferred taxes.
Next, we deﬁne net operating proﬁts after tax (NOPAT) as EBEI plus the
aftertax cost of interest expense
NOPAT"EBEI#ATInt,
where
ATInt"the aftertax equivalent of book interest expense.
NOPAT separates operating activities from ﬁnancing activities by adding back
the aftertax eﬀect of debt ﬁnancing charges (interest expense) included in EBEI.
Residual income diﬀers from EBEI in that it measures operating performance
(NOPAT) net of a charge for the cost of all debt and equity capital employed:
RI"NOPAT!(k
*
Capital),
where
k "Stern Stewart’s estimate of the ﬁrm’s weighted average cost of
capital.
G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336 305
` Other adjustments to NOPAT include: adding the change in bad debt allowances; adding the
change in the LIFO reserve; adding goodwill amortization; adding other operating income; and
subtracting an estimate of taxes owed for the period (Stewart, 1991) (pp. 742—743). Stern Stewart do
not disclose complete details about their accounting adjustments, e.g., asset lives and amortization
patterns.
" Other adjustments to Capital include: capitalization and amortization of certain marketing
costs; subtracting marketable securities and construction in progress (because neither contributes to
current operating activities); adding the present value of noncapitalized long term leases; adding
allowances for bad debts, inventory obsolescence, warranties, etc.; adding the LIFO reserve; adding
net capitalized intangibles (including R&D); adding cumulative goodwill amortization; adding
unrecorded goodwill; and adding (subtracting) cumulative unusual losses (gains), net of taxes
(Stewart, 1991) (pp. 112—117). AcctAdj
ºº
and AcctAdj
°
are not examined individually in subsequent
empirical tests because Stern Stewart does not disclose them separately.
Capital "Stern Stewart’s deﬁnition of assets (net of depreciation) invested in
goingconcern operating activities, or equivalently, contributed and
retained debt and equity capital, at the beginning of the period.
Positive RI reﬂects proﬁts in excess of that required by debt and equity capital
suppliers and, thus, is consistent with the ﬁrm creating wealth for the residual
claimants, the shareholders. Negative RI is consistent with decreasing share
holder wealth.
EVA is Stern Stewart’s proprietary version of RI. Stern Stewart attempts to
improve on RI by adjusting NOPAT and Capital for what they view to be
‘distortions’ in the accounting model of performance measurement (Stewart,
1991) (Chapter 2):
EVA"NOPAT#AcctAdj
ºº
!k
*
[Capital#AcctAdj
°
],
where
AcctAdj
ºº
"Stern Stewart adjustments to accounting measures of operating
proﬁts.
AcctAdj
°
"Stern Stewart adjustments to accounting measures of capital.
As an example of a common accounting adjustment, Stewart (1991) (pp. 28—30)
argues that research and development costs should be capitalized (if material)
and amortized. This requires adjustments to both NOPAT (via AcctAdj
ºº
)
and to Capital (via AcctAdj
°
). NOPAT is adjusted by adding back the period’s
R&D expense and deducting amortization of the R&D asset.` In any given
year, the net eﬀect is an increase (decrease) in NOPAT if R&D expense is greater
(less) than R&D amortization. AcctAdj
°
reﬂect the cumulative eﬀect on Capital
of the capitalization and amortization of current and past R&D expenditures.
At any point in time, Capital is higher by the amount of the net capitalized
R&D asset."
306 G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
Fig. 1. Components of economic value added (EVA).
Relying on the above deﬁnitions, EVAcan be decomposed into its component
parts:
EVA"CFO#Accrual#ATInt!CapChg#AcctAdj,
where
CapChg"k
*
Capital
AcctAdj"AcctAdj
ºº
!(k
*
AcctAdj
°
).
Fig. 1 summarizes these relations by showing how EVA components combine
into other performance measures, i.e., CFO, EBEI and RI. We use this
decomposition to examine the incremental information content of EVA
components.
2.2. Hypotheses
By assuming that equity markets are (semistrong) eﬃcient, forwardlooking
and can form estimates of performance measures, we use stock market returns
to compare the information content, or valuerelevance, of CFO, EBEI, RI
and EVA. Following Biddle et al., 1995, we draw a distinction between
relative and incremental information content. Relative information content
comparisons are appropriate when one desires a ranking of performance
measures by information content or when making mutually exclusive choices
among performance measures, i.e., when only one measure can be chosen. In
contrast, incremental information content comparisons assess whether one
measure provides valuerelevant data beyond that provided by another
measure and apply when assessing the information content of a supplemental
disclosure or the information of a component measure (e.g., Bowen et al.,
1987).
G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336 307
` The MVEdeﬂator is measured 3 months after the prior year end to be consistent with the start of
the returns period measured by the dependent variable.
Despite claims by Stern Stewart and others that EVA and RI are more
valuerelevant to market participants than EBEI and CFO, we take a neutral
position and conduct twotail tests of the null hypotheses that CFO, EBEI, RI
and EVA have equal relative information content:
H
"
: The information content of measure X
¹
is equal to that of X
`
where X
¹
and X
`
represent pairwise combinations from the set of performance
measures: CFO, EBEI, RI and EVA. Rejection of H
"
is viewed as evidence of
a signiﬁcant diﬀerence in relative information content.
We examine the incremental value relevance of EVA components sum
marized in Fig. 1 by testing the null hypotheses that individual components of
EVA do not provide incremental information content beyond other compo
nents that also comprise CFO and EBEI:
H
'
: Component X
¹
does not provide information content beyond that pro
vided by the remaining components X
`
—X
`
where X
¹
—X
`
are components of EVA (i.e., CFO, Accrual, ATInt, CapChg and
AcctAdj). Rejection of H
'
is viewed as evidence of incremental information
content.
2.3. Statistical tests
A standard approach for assessing information content is to examine the
statistical signiﬁcance of the slope coeﬃcient, b
¹
, in the following ordinary
leastsquares regression (that omits ﬁrm subscripts):
D
R
"b
"
#b
¹
FE
6R
/MVE
R¹
#e
R
(1)
where, D
R
is the dependent variable, a measure of (abnormal or unexpected)
returns for time period t; FE
6R
/MVE
R¹
is the unexpected realization (or forecast
error) for a given accounting measure, X (e.g., CFO, EBEI, RI or EVA), scaled
by the beginningofperiod market value of the ﬁrm’s equity, MVE
R¹
,` and e
R
is
a random disturbance term.
Because little is known about suitable proxies for market expectations for
performance measures other than earnings, we use an approach from Biddle and
308 G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
` We also consider a speciﬁcation that allows each information variable to be predicted by lagged
observations of all of the information variables. Thus each information variable, say EVA, is
predicted by lagged values of each of the other variables — CFO, EBEI, RI and EVA. This is one way
of addressing the potential concern that (say) EVA is less well predicted by past observations of EVA
than (say) EBEI is predicted by past values of EBEI. Results based on these speciﬁcations are
qualitatively similar to those reported and are available from the authors.
Seow (1991) and Biddle et al., 1995 that estimates market expectations ‘jointly’
with slope coeﬃcients. This is accomplished by ﬁrst expressing the forecast error
as the diﬀerence between the realized value of a performance measure and the
market’s expectation: FE
R
"X
R
!E(X
R
). It is then assumed that market expecta
tions are formed according to a discrete linear stochastic process (in autoregres
sive form):
E(X
R
)"#
¹
X
R¹
#
`
X
R`
#
`
X
R`
#
2
(2)
where the is a constant and ’s are autoregressive parameters. Substituting
Eq. (2) into Eq. (1) yields:
D
R
"b
"
#b
¹
(X
R
!(#
¹
X
R¹
#
`
X
R`
#
`
X
R`
#
2
))/MVE
R¹
#e
R
"b
"
#b
¹
X
R
/MVE
R¹
#b
`
X
R¹
/MVE
R¹
#b
`
X
R`
/MVE
R¹
#b
"
X
R`
/MVE
R¹
#
2
#e
R
. (3)
Eq. (3) relates abnormal returns and (scaled) lagged measures of accounting
performance, where E(b
"
)"b
"
!b
¹
, E(b
¹
)"b
¹
, and E(b
G
)"!b
G
G¹
for
i'1. In Eq. (3), the proxy for market expectations is estimated jointly with the
slope coeﬃcient (b
G
) using the same data and optimization criterion (minimum
mean squared errors).
Eq. (3) encompasses a range of alternative speciﬁcations for market expecta
tions, including randomwalk, ARIMA, constant stock price multiple, and
combined ‘levels and changes’ speciﬁcations. Although Eq. (3) is ﬂexible in terms
of allowing any number of lagged observations to be included as explanatory
variables, in the presence of possible structural change across time, we limit
Eq. (3) to one lag:`
D
R
"b
"
#b
¹
X
R
/MVE
R¹
#b
`
X
R¹
/MVE
R¹
#e
R
. (4)
This ‘onelag’ version is equivalent to the ‘levels and changes’ speciﬁcation
proposed by Easton and Harris (1991), but it is motivated diﬀerently. It also is in
a more convenient form that allows the slope or ‘response’ coeﬃcient (b
¹
) to be
G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336 309
" The relation between the two speciﬁcations can be illustrated by starting with the levelschanges
speciﬁcation (in Eq. (4a)) and deriving the onelag speciﬁcation (in Eq. (4c)):
D
R
"a
"
#a
¹
X
R
/MVE
R¹
#a
`
(X
R
!X
R¹
)/MVE
R¹
#e
R
(4a)
"a
"
#a
¹
X
R
/MVE
R¹
#a
`
X
R
/MVE
R¹
!a
`
X
R¹
/MVE
R¹
#e
R
(4b)
"a
"
#(a
¹
#a
`
)X
R
/MVE
R¹
!a
`
X
R¹
/MVE
R¹
#e
R
. (4c)
Eq. (4c) corresponds to Eq. (4) where b
¹
"a
¹
#a
`
and b
`
"!a
`
. Since a
¹
and a
`
are both
expected to be positive, b
¹
(b
`
) is predicted to be positive (negative). The coeﬃcient(s) on the nonlag
term(s) can be interpreted directly as ‘response’ coeﬃcient(s), e.g., in Eq. (4c) the response coeﬃcient
is (a
¹
#a
`
).
¹" The Biddle—Seow—Siegel test derives from Hotelling (1940). By using a lackofﬁt measure
deﬁned as the average of the sum of squared residuals and the sum of squared prediction errors,
a nonlinear null hypothesis is obtained that involves quadratic forms of regression coeﬃcients. It is
tested using a Wald test (Kennedy, 1985) of estimated coeﬃcients and their heteroskedasticity
adjusted variance—covariance matrix. As discussed in Biddle et al. (1995), this method for assessing
relative information content compares favorably with alternative tests provided in Davidson
and MacKinnon (1981) and Vuong (1989). Davidson and MacKinnon’s nonnested ‘Jtest’ and
Vuong’s likelihood ratio test are ‘pairwise tests for model selection’ designed to assess which of
two competing models is closer (in terms of KullbackLiebler distance) to the ‘truth’. Both are
valid only asymptotically and may have poor ﬁnite sample properties. The Jtest also can yield
ambiguous results, which is problematic in applications assessing relative information content.
Dechow, Lys and Sabino (1996) provide evidence that Vuong’s test outperforms the Jtest. Biddle
and Siegel (1996) provide evidence that the Biddle—Seow—Siegel test performs as well as or better
than Vuong’s test in calibration and power. As conﬁrming evidence, we replicated our relative
information content tests using Vuong’s test with qualitatively similar results as discussed brieﬂy
in Section 4 below.
observed directly (rather than being derived from separate coeﬃcients on levels
and changes)."
2.3.1. Tests for relative information content
To assess relative information content, we employ a statistical test from
Biddle et al. (1995) that allows a test of the null hypothesis of no diﬀerence in the
ability of two competing sets of independent variables to explain variation in the
dependent variable. Using this test, we make six pairwise comparisons of
regressions among the accounting performance measures CFO, EBEI, RI and
EVA, as speciﬁed in Eq. (4). The test is constructed as a comparison of R`s.
Under usual regularity conditions (uncorrelated homoskedastic errors), it is
ﬁnite sample exact, generalizes to any number of predictor variables, and can be
used in conjunction with White’s (1980) correction for heteroskedastic errors. As
a result, it is well suited to evaluate the signiﬁcance of relative information
content comparisons in accounting contexts.¹"
310 G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
¹¹ For their publicly available database used in this study, Stern Stewart make ‘a handful’ of
standard adjustments. For their corporate clients, Stern Stewart make additional custom adjust
ments (not available to the public).
¹` In other words, data greater (less) than 4 standard deviations from the median of the ﬁrmyear
observations are assigned a value equal to the median plus (minus) 4 standard deviations. The total
number of observations reset for any variable range from 51 to 97, or 0.83% to 1.57% of the 6,174
sample ﬁrm years.
2.3.2. Tests for incremental information content
Following standard methodology (e.g., Bowen et al. (1987)), incremental
information content is assessed by examining the statistical signiﬁcance of
regression slope coeﬃcients. Speciﬁcally, for the onelag speciﬁcation in Eq. (4)
generalized to two accounting performance measures X and ½, incremental
information content is assessed using ttests on individual coeﬃcients and
Ftests of the joint null hypotheses:
H
"6
: b
¹
"b
`
"0,
H
"7
: b
`
"b
"
"0,
where b
¹
, b
`
, b
`
and b
"
are from Eq. (5) below:
D
R
"b
"
#b
¹
X
R
/MVE
R¹
#b
`
X
R¹
/MVE
R¹
#b
`
½
R
/MVE
R¹
#b
"
½
R¹
/MVE
R¹
#e
R
. (5)
To control for the potential eﬀects of heteroskedastic errors, White’s (1980)
correction is employed in both the relative and incremental information content
tests.
3. Sample selection, variable deﬁnitions and descriptive statistics
3.1. Sample selection
Data used in this study were purchased directly from Stern Stewart & Co.¹¹
These data include up to eleven annual observations for economic value added
(EVA), capital, and cost of capital for ﬁrms with ﬁscal years ending June 1983 to
May 1994 (see variable deﬁnitions below). The initial sample of 1000 ﬁrms (8,524
ﬁrmyear observations) is reduced by 219 ﬁrms (2,271 observations) due to
either missing Compustat or CRSP (Center for Research in Securities Prices)
data or to provide a lagged observation for each variable. We also delete 79
extreme outlier observations deﬁned as more than 8 standard deviations from
the median. Next, both the dependent and independent variables are winsorized
to $4 standard deviations from the median.¹` The resulting sample has 6,174
ﬁrmyear observations for 773 ﬁrms.
G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336 311
¹` Consistent with the possibility that pre1988 measures of CFO are noisy, R`s in twoyear
subperiods from 1988 onward are slightly higher than for the two, twoyear subperiods before
1988.
These data are compiled by Stern Stewart & Company from Business ¼eek’s
listing of the 1,000 largest ﬁrms in market capitalization. Stern Stewart modiﬁes
this list by ﬁrst removing utilities and ﬁnancial institutions, and then adding
ﬁrms from prior Business ¼eek 1000 listings to bring the sample back to 1,000
ﬁrms. Stern Stewart introduced its ﬁrst 1000 ranking for the calendar year ended
1988. The listing is published annually.
3.2. Dependent variable
Our dependent variable, market adjusted returns, is commonly used in
information content studies to measure unexpected returns (e.g., Biddle et al.,
1995; Bowen et al., 1989).
MktAdjRet Market adjusted return computed from CRSP data as a ﬁrm’s
12month compounded stock return less the 12month com
pounded valueweighted marketwide return. A 12month non
overlapping period ending three months following the ﬁrm’s ﬁscal
yearend is chosen to allow time for information contained in the
ﬁrm’s annual report to be impounded in stock market prices.
3.3. Independent variables and descriptive data — relative information content tests
The four measures of accounting performance in the relative information
content tests, CFO, EBEI, RI and EVA, are deﬁned below:
CFO Cash ﬂow from operations obtained from the statement of cash
ﬂows or the statement of changes in ﬁnancial position, depending
upon the year of the observation. For years after 1987 Compustat
data item D308, operating activities — net cash ﬂow, is used. For
years prior to 1988, data item D110, funds from operations
— total, is used if the ﬁrm used the cash deﬁnition of funds for the
statement of changes in ﬁnancial position. If the ﬁrm used the
working capital deﬁnition of funds in any year prior to 1988, cash
ﬂow from operations is estimated similar to Bowen et al. (1986,
1987) as funds from operations (D110) plus the change in current
liabilities (D5) less the change in debt in current liabilities (D34)
less the change in current assets (D4) plus the change in cash and
cash equivalents (D1).¹`
312 G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
EBEI Earnings deﬁned as Compustat data item D18, net income before
extraordinary items.
RI Residual income equals earnings plus aftertax interest expense
less a charge on all capital (RI"EBEI#ATInt!CapChg). See
Section 3.4 below for deﬁnitions of ATInt and CapChg.
EVA Economic value added obtained from the Stern Stewart 1000
database.
In order to reduce heteroscedasticity in the data, we deﬂate all independent
variables by the market value of equity three months after the beginning of the
ﬁscal year (MVE
R¹
). Descriptive data on these deﬂated, winsorized variables
pooled across time are provided in Panel A of Table 1. EBEI has the lowest
standard deviation among the four performance measures consistent with the
smoothing eﬀects of accruals. CFO has the largest ﬁrmyear mean and median
followed by EBEI, EVA and RI. Undeﬂated median values of each performance
measure are plotted across time in Fig. 2. Despite a survivorship bias in the
data, median RI is negative in every year and median EVA is negative in 7 out of
10 years. Near zero EVA and RI is consistent with a competitive economy where
even the typical large ﬁrm has diﬃculty earning more than its cost of capital.
Low values of EVA and RI are also consistent with a potential upward bias in
Stern Stewart’s cost of capital estimates.
Correlations among these measures are provided in Panel A of Table 1.
Correlations between the independent variables are all positive and signiﬁcant
except that EVA and RI are negatively correlated with CFO. EBEI has the
highest correlation with market adjusted returns.
3.4. Independent variables and descriptive data — incremental information
content tests
The independent variables in the incremental information content tests are
the ﬁve components of EVA described in Section 2.1 and summarized in Fig. 1:
CFO (deﬁned above), operating accruals, aftertax interest expense, capital
charge and accounting adjustments:
Accrual Operating accruals deﬁned as earnings less cash ﬂow from opera
tions (Accruals"EBEI!CFO). Accruals can be positive or nega
tive but are more likely to be negative (reﬂecting noncash expenses
such as depreciation and amortization).
ATInt After tax interest expense computed as 1 minus the ﬁrm’s tax rate
multiplied by interest expense (D15). The ﬁrm’s tax rate is assumed
to be zero if net operating losses are present. Otherwise the max
imumstatutory corporate tax rate is used for the given year. ATInt is
nonnegative.
G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336 313
Table 1
Descriptive statistics for pooled data
Panel A: Descriptive statistics on the dependent and independent variable in relative information
content testsº
Dependent
variable Independent variables
MktAdjRet
'
EBEI
R
EVA
R
RI
R
CFO
R
Descriptive statistics
Mean 0.059 0.057 !0.049 !0.056 0.142
Median 0.011 !0.065 !0.007 !0.017 0.118
Std. Dev. 0.362 0.082 0.134 0.127 0.133
Correlations'
MktAdjRet
'
1.00
EBEI
R
0.247 1.00
EVA
R
0.153 0.592 1.00
RI
R
0.155 0.652 0.900 0.900
CFO
R
0.138 0.307 !0.125 !0.122 1.00
Panel B: Descriptive statistics on the dependent and independent variable in incremental information
content testsº
Dependent
variable Independent variables
MktAdjRet
'
CFO
R
Accrual
R
ATInt
R
CapChg
R
AccAdj
R
Descriptive statistics
Mean 0.059 0.142 !0.086 0.034 0.149 !0.007
Median 0.011 0.118 !0.055 0.016 0.111 !0.007
Std. Dev. 0.362 0.133 0.137 0.054 0.129 0.055
Correlations'
MktAdjRet
'
1.00
CFO
R
0.138 1.00
Accrual
R
0.021 !0.782 1.00
ATInt
R
!0.026 0.363 !0.501 1.00
CapChg
R
!0.018 0.469 !0.580 0.751 1.00
AccAdj
R
!0.011 0.004 0.039 0.210 0.057 1.00
ºThe sample has 6,174 ﬁrmyear observations. All variables are winsorized $4 standard deviations
from the median. All independent variables are deﬂated by the market value of equity three months
after the beginning of the ﬁscal year.
'Pearson correlation coeﬃcients '0.0204 are signiﬁcant at (0.10
'0.0319 are signiﬁcant at (0.01
'0.0407 are signiﬁcant at (0.001
314 G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
Fig. 2. Median values of performance measures, 1984—93.
¹" According to Stewart (1994), Stern Stewart estimate the cost of capital by weighting the cost of
equity (applying the capital asset pricing model) and the aftertax cost of debt. Capital is a proxy for all
cash invested in the business since a company’s inception. See Stewart (1991), especially pp. 741—745.
¹` Our deﬁnition of RI incorporates Stern Stewart adjustments to capital. Data were not available
from Stern Stewart to calculate capital before accounting adjustments.
CapChg Capital charge deﬁned as the ﬁrm’s weighted average cost of debt
and equity capital times its beginning of year capital.¹" Both of these
items are obtained from Stern Stewart. CapChg is positive since both
the cost of capital and capital are positive.
AcctAdj Accounting adjustments reﬂect Stern Stewart’s net annual adjust
ments to earnings and capital, and are deﬁned as economic value
added less residual income (AcctAdj"EVA!RI).¹` AcctAdj can be
positive or negative.
G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336 315
¹" On average, we predict a positive (negative) slope coeﬃcient on contemporaneous (lagged)
observations of each performance measure. The negative coeﬃcient on the lagged term follows from
the prediction that changes in the performance measures also are positively associated with stock
returns (see Section 2.3, especially footnote 9). In results available from the authors, coeﬃcient
b
¹
(b
`
) is positive (negative) and signiﬁcant (at (0.00001) for each performance measure based on
the full sample of 6,174 ﬁrmyears.
¹` Nearly identical inferences are obtained using the Vuong (1989) test. For example, for the
relative comparisons in Panel A of Table 2 and Table 4, identical inferences are obtained at
conventional signiﬁcance levels for all pairwise comparisons. In general, the Vuong test provides
greater statistical signiﬁcance, consistent with its asymptotic nature and tendency to reject the null
observed in simulation tests (Biddle and Siegel, 1996).
Descriptive data on these deﬂated, winsorized EVA components are provided
in panel B of Table 1. CFO has by far the largest correlation with marketadjusted
returns. Both mean and median Accrual and AcctAdj are negative, consistent
with some smoothing of the underlying operating cash ﬂows. Correlations be
tween CFO, ATInt and CapChg are positive and signiﬁcant, consistent with ﬁrms
with higher operating cash ﬂows also having higher debt and equity costs. The
negative correlation between CFO and Accrual is again consistent with the
accrual process smoothing earnings relative to the underlying operating cash
ﬂows. The correlation between CFO and AcctAdj is insigniﬁcant.
4. Empirical results
4.1. Relative information content tests
Relative information content is assessed by comparing adjusted R`s from four
separate regressions, one for each performance measure, CFO, EBEI, RI and
EVA.¹" Adjusted R`s from these regressions are provided in Table 2. The
highest R` is shown on the left (which in both panels is from the EBEI
regression) and the lowest is shown on the far right (which in both panels is from
the CFO regression). pvalues from twotailed statistical tests of relative in
formation content are shown centered in parentheses for each of the six possible
pairwise comparisons.
Results in Panel A of Table 2 are based on Eq. (4) and each of the six pairwise
diﬀerences in R` are signiﬁcant at conventional levels, with EBEI having
a signiﬁcantly larger adjusted R` (9%) than each of the other three performance
measures. The RI regression has a signiﬁcantly larger adjusted R` (6.2%) than
does the EVA regression (5.1%), and both have a signiﬁcantly larger adjusted
R`s than CFO (2.4%). These results suggest that, in terms of relative informa
tion content, earnings signiﬁcantly outperforms RI, RI signiﬁcantly outperforms
EVA (although the gap here is smaller), and all three outperform CFO.¹`
316 G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
Table 2
Tests of the relative information content of EVA, residual income, earnings and operating cash ﬂow
(H
"
)
Relative information content
Rank order Observa
of R` tions (1) (2) (3) (4)
Panel A: Coeﬃcient of positive and negative values of each performance measure constrained to be
equalº
All ﬁrms 6,174 EBEI ' RI ' EVA ' CFO
Adj. R` 0.0904 0.0624 0.0507 0.0238
pvalue' (0.000) (0.041) (0.000)
(0.000) (0.000)
(0.000)
Panel B: Coeﬃcient of positive and negative values of each performance measure allowed to diﬀer°
All ﬁrms 6,174 EBEI ' RI ' EVA ' CFO
Adj. R` 0.1278 0.0732 0.0649 0.0280
pvalue' (0.000) (0.266) (0.000)
(0.000) (0.000)
(0.000)
ºUnderlying regressions are from Eq. (4): D
R
"b
"
#b
¹
X
R
/MVE
R¹
#b
`
X
R¹
/MVE
R¹
#e
R
, where
D
R
"marketadjusted returns; X"a given performance measure (CFO, EBEI, RI and EVA); and
MVE"the market value of equity three months after the beginning of the ﬁscal year. Performance
metrics are listed in order of Rsquares from highest (on the left) to lowest (on the right). Statistical
tests of diﬀerences in explanatory power across performance measures are presented centered in
parantheses below the adjusted Rsquares. See description of pvalue below.
'Twotailed pvalues in parentheses represent tests of the null hypothesis of no diﬀerence between
pairwise comparisons of adjusted Rsquares (Biddle et al., 1995). First row presents pvalue for
comparison between ﬁrst and second ranked measures, second and third ranked measures and third
and fourth ranked measures. On the the next row, comparisons are between ﬁrst and third ranked,
and second and fourth ranked measures. The last row compares ﬁrst and fourth ranked measures.
°Underlying regressions are from Eq. (4) modiﬁed to allow diﬀerent coeﬃcients on positive versus
negative values of the independent variables:
D
R
"b
"
#b
¹
X
Rºº'
/MVE
R¹
#b
`
X
R"°¨
/MVE
R¹
#b
`
X
R¹ºº'
/MVE
R¹
#b
"
X
R¹"°¨
/MVE
R¹
#e
R
(6)
where D
R
"marketadjusted returns; X"a given performance measure (CFO, EBEI, RI or EVA);
and MVE"market value of equity three months after the beginning of the ﬁscal year.
The underlying regressions in panel A constrain the coeﬃcients to be equal
across all ﬁrmyear observations. Hayn (1995), Burgstahler and Dichev (1997)
and Collins et al. (1997) provide evidence that loss ﬁrms have smaller earnings
response coeﬃcients than do proﬁtable ﬁrms. Because the valuerelevance of the
G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336 317
other performance measures (CFO, RI and EVA) could also vary with their sign
(O’Byrne, 1996), we repeat our tests for relative information content after
partitioning each performance measure into positive and negative values:
MktAdjRet
R
"b
"
#b
¹
X
Rºº'
/MVE
R¹
#b
`
X
R"°¨
/MVE
R¹
#b
`
X
R¹ºº'
/MVE
R¹
#b
"
X
R¹"°¨
/MVE
R¹
#e
R
. (6)
Panel B of Table 2 presents results for regression (6) for the complete sample
of 6,174 ﬁrmyear observations. Consistent with prior research, coeﬃcients
(available from the authors) are generally larger (in absolute value) and more
signiﬁcant for positive values of X
R
than for the negative values. Compared to
results reported above in Panel A, adjusted R`s increase for each performance
measure when allowing for separate coeﬃcients on positive and negative values.
This increase is largest for the EBEI regression with adjusted R` increasing from
9% to 12.8%. However, the ranking of performance measures remains identical
and statistical comparisons between regressions are nearly unchanged — earn
ings dominates each of the other three performance measures and all three
(EBEI, RI and EVA) dominate CFO. The only diﬀerence is that RI and EVA are
no longer statistically diﬀerent from each other.
Taken as a whole, the relative information content results show no evidence of
EVA(RI or CFO) dominating EBEI. Thus, we cannot support the Stern Stewart
claim that EVA has greater information content than earnings. In contrast, this
evidence points to earnings having higher relative information content than
EVA. In Section 5, we examine the sensitivity of these results to alternative
speciﬁcations. In Section 6, we discuss possible reasons why we fail to detect
stronger valuerelevance for EVA and RI.
4.2. Incremental information content tests
Table 3 presents results on the incremental information content of EVA
components from regression (7):
MktAdjRet
R
"b
"
#b
¹
CFO
R
#b
`
CFO
R¹
#b
`
Accrual
R
#b
"
Accrual
R¹
#b
`
ATInt
R
#b
"
ATInt
R¹
#b
`
CapChg
R
#b
`
CapChg
R¹
#b
"
AcctAdj
R
#b
¹"
AcctAdj
R¹
#e
R
. (7)
Predicted signs on each coeﬃcient are provided below the variable labels. We
expect a positive association between marketadjusted returns and the three
components CFO, Accrual and AcctAdj. We expect a negative association be
tween returns and the two components representing nonnegative capital costs,
ATInt and CapChg. Similar to the relative information content regressions in
Eq. (4), the lagged terms are predicted to have the opposite sign (footnote 9).
318 G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
T
a
b
l
e
3
T
e
s
t
s
o
f
i
n
c
r
e
m
e
n
t
a
l
i
n
f
o
r
m
a
t
i
o
n
c
o
n
t
e
n
t
o
f
E
V
A
c
o
m
p
o
n
e
n
t
s
:
C
F
O
,
o
p
e
r
a
t
i
n
g
a
c
c
r
u
a
l
s
,
a
f
t
e
r

t
a
x
i
n
t
e
r
e
s
t
,
c
a
p
i
t
a
l
c
h
a
r
g
e
,
a
c
c
o
u
n
t
i
n
g
a
d
j
u
s
t
m
e
n
t
s
(
H
'
)
º
O
b
s
.
C
o
n
s
t
a
n
t
C
F
O
R
C
F
O
R
¹
A
c
c
r
u
a
l
R
A
c
c
r
u
a
l
R
¹
A
T
I
n
t
R
A
T
I
n
t
R
¹
C
a
p
C
h
g
R
C
a
p
C
h
g
'
¹
A
c
c
A
d
j
'
A
c
c
A
d
j
'
¹
A
d
j
.
R
`
P
r
e
d
i
c
t
e
d
s
i
g
n
s
:
#
!
#
!
!
#
!
#
#
!
A
l
l
ﬁ
r
m
s
6
,
1
7
4
0
.
0
1
3
1
.
4
7
3
!
0
.
8
2
4
1
.
1
9
2
!
0
.
7
5
1
!
0
.
5
9
4
0
.
7
7
2
!
0
.
3
9
1
0
.
2
7
0
0
.
3
5
7
0
.
0
5
5
0
.
0
9
0
7
t

s
t
a
t
1
.
4
2
1
6
.
0
2
!
8
.
5
3
1
3
.
0
9
!
7
.
7
3
!
2
.
2
2
1
2
.
6
3
!
2
.
4
3
1
.
7
2
3
.
1
2
0
.
4
8
F

s
t
a
t
1
2
8
.
4
2
8
7
.
8
3
3
.
4
5
3
.
6
1
6
.
5
5
p

v
a
l
u
e
'
(
0
.
0
0
0
)
(
0
.
0
0
0
)
(
0
.
0
3
2
)
(
0
.
0
2
7
)
(
0
.
0
0
1
)
º
D
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
"
m
a
r
k
e
t

a
d
j
u
s
t
e
d
r
e
t
u
r
n
s
;
i
n
d
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
s
a
r
e
c
o
m
p
o
n
e
n
t
s
o
f
E
V
A
(
C
F
O
,
o
p
e
r
a
t
i
n
g
a
c
c
r
u
a
l
s
,
a
f
t
e
r

t
a
x
i
n
t
e
r
e
s
t
e
x
p
e
n
s
e
,
c
a
p
i
t
a
l
c
h
a
r
g
e
,
a
c
c
o
u
n
t
i
n
g
a
d
j
s
u
t
m
e
n
t
s
)
a
n
d
a
r
e
s
h
o
w
n
i
n
n
o
n

l
a
g
g
e
d
f
o
r
m
s
a
s
c
o
l
u
m
n
h
e
a
d
i
n
g
s
.
E
a
c
h
i
n
d
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
i
s
d
e
ﬂ
a
t
e
d
b
y
m
a
r
k
e
t
v
a
l
u
e
o
f
e
q
u
i
t
y
t
h
r
e
e
m
o
n
t
h
s
a
f
t
e
r
t
h
e
b
e
g
i
n
n
i
n
g
o
f
t
h
e
ﬁ
s
c
a
l
y
e
a
r
.
'
p

v
a
l
u
e
s
i
n
p
a
r
e
n
t
h
e
s
e
s
r
e
p
r
e
s
e
n
t
n
o
n

d
i
r
e
c
t
i
o
n
a
l
F

t
e
s
t
s
o
f
t
h
e
n
u
l
l
h
y
p
o
t
h
e
s
i
s
o
f
n
o
i
n
c
r
e
m
e
n
t
a
l
i
n
f
o
r
m
a
t
i
o
n
c
o
n
t
e
n
t
(
h
y
p
o
t
h
e
s
i
s
H
'
)
.
G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336 319
Fig. 3. Relative and incremental information content of EVA, residual income, earnings and
operating cash ﬂow.
In Panel A for the full sample, 9 out of 10 coeﬃcients are in the predicted
direction and signiﬁcant in onetail ttests at the 0.05 level or better. The
exception is the lagged term for AcctAdj, which is in the wrong direction. All of
the twotail Ftests are signiﬁcant at the 0.05 level or better. The relative sizes of
the Fstatistics suggest that CFO and Accrual make by far the largest incre
mental contributions to explaining marketadjusted returns, while ATInt,
CapChg and AcctAdj exhibit much smaller incremental contributions. When
combined with the relative information content ﬁndings above, these results
suggest that, while EVA components oﬀer some incremental information con
tent beyond earnings components, their contributions to the information con
tent of EVA are not suﬃcient for EVA to provide greater relative information
content than earnings.
Fig. 3 uses a Venn diagram to summarize our ﬁndings on relative and
incremental information content for the four information variables CFO, EBEI,
RI and EVA. The size of each circle represents relative information content and
the nonoverlapping areas represent incremental information content. EBEI
exhibits the largest relative information content among the measures. CFO, RI
and EVA protrude slightly from behind EBEI reﬂecting some limited incremen
tal information content beyond earnings. However, the overall minuscule in
crease in adjusted R` between the regression of returns on EBEI (9.04% in Panel
A of Table 2) and returns on EVA components (9.07% in Table 3) suggests that
the economic signiﬁcance of the incremental information content of the EVA
components is slight.
320 G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
5. Sensitivity analyses and extensions
In this section, we examine the sensitivity of the basic results reported above
to alternative speciﬁcations. We repeat selected information content tests by:
1) partitioning annual observations into ﬁve, nonoverlapping, twoyear test
periods (instead of one tenyear period); 2) evaluating subsets of ﬁrms that claim
to use EVA for internal business decisions; 3) changing the return interval from
oneyear to ﬁveyears; and 4) changing the return interval from oneyear
(contemporaneous) returns to twoyear (combined contemporaneous and one
year ahead) returns. Finally, we discuss a replication and extension of O’Byrne
(1996), where the dependent variable is the level of market value of the ﬁrm
(rather than returns). We conclude with an overall assessment of the results of
the sensitivity tests.
5.1. Partitioning the sample into subperiods
Results reported in Tables 2 and 3 pool observations over the ten years
1984—1993. In this section, we report relative and incremental information
content tests on annual data grouped into ﬁve, nonoverlapping, twoyear
periods. Because of survivorship bias in the Stern Stewart data, ﬁrmyear
observations increase from 1,015 in the 1984—85 period to 1,481 in 1992—93.
In pairwise comparisons of relative information content, adjusted R`s are
largest for EBEI in every twoyear period. However, in 1984—85 diﬀerences
between EBEI, EVA and RI are not statistically signiﬁcant at conventional
levels. Using a 5% cutoﬀ, in 1986—87, EBEI does not outperform RI (p"0.072)
but does outperform EVA (p"0.045). In 1988—89 and 1990—91, EBEI outper
forms each of the other performance measures at the 0.01 level or better. In
1992—93, EBEI does not outperform RI (p"0.083) but does outperform EVA
(p"0.006). Taken together, there is no evidence of EVA (RI or CFO) domina
ting EBEI. Thus we again cannot support the Stern Stewart claim that EVA has
greater information content than earnings. In contrast, the evidence points to
earnings having higher relative information content in many subperiods.
We also consider the 606 observations following the September 1993 Fortune
article that touted EVA as “The Real Key to Creating Wealth” (Tully, 1993).
The earnings regression again has the highest R` (11.2%), and the evidence is
suggestive of EBEI dominating EVA (p"0.049) and CFO (p"0.061) but not
RI. Thus market participants apparently did not begin using EVA for equity
valuation immediately following the appearance of the Fortune article.
In incremental information content tests, CFO and Accrual are signiﬁcant in
every twoyear period. Among the remaining EVA components (ATInt, CapChg
and AcctAdj), only 1 of 15 Fstatistics is signiﬁcant at the 5% level — AcctAdj in
the 1984—85 sample period. Results for the period after release of the 1993
Fortune article again show strong support for the incremental information
G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336 321
content of CFO and Accrual, but little evidence for the incremental signiﬁcance
of the remaining EVA components.
5.2. Adopters of ‘E»Alike’ performance measures
It is possible that ﬁrms adopt EVA at least in part because their past
experience indicates a relatively strong relation between EVA and stock returns.
Further, investors may become more attuned to the measure for ﬁrms that
adopt EVA. Thus, it is conceivable that the association between EVA and
returns is stronger for EVA adopters. To examine this possibility, we consider
separately four subsamples of ﬁrms that make some ‘use’ of EVAlike measures.
Firms in the ‘Any’ sample have disclosed that they use EVA (or some similar
concept) sometime during the period studied — even if that use appears to be
minimal. Firms in the ‘Performance’ sample provide more detail about their use
of an EVAlike measure for performance measurement and/or decision making.
Firms in the ‘Comp’ sample state that they use an EVAlike measure in senior
management incentive compensation plans and thus, presumably, also use it for
performance measurement and/or decision making. We include all available
data including years before the plan was implemented. The ‘Comp Year’ sample
restricts observations in the ‘Comp’ sample to only those years in which an
EVAbased compensation plan is in eﬀect. Thus, the ‘Comp Year’ sample is
a subset of the ‘Comp’ sample, which is a subset of the ‘Performance’ sample,
which, in turn, is a subset of the ‘Any’ sample.
Table 4 reports the results of relative information content tests for ﬁrms using
an EVAlike performance measure. EBEI exhibits the largest R`s for the ‘Any,’
‘Performance’ and ‘Comp’ groups, but EVA has the largest R` for the ‘Comp
Year’ sample. However, none of the performance measures diﬀer signiﬁcantly in
relative information content at the 5% level, and only 3 out of 18 comparisons at
the 10% level (EBEI'CFO for the ‘Any’ subsample (p"0.073); EBEI'RI
for the ‘Any’ and ‘Performance’ subsamples (p"0.094 and 0.059, respectively)).
While earnings is not as dominant in these smaller subsamples of EVA users,
neither do the ﬁndings show EVA dominating earnings in its association with
stock returns. The lower signiﬁcance levels may be attributable in part to the
smaller sample sizes used in these tests.
Table 5 reports tests of incremental information content for users of EVAlike
performance measures. In onetail ttests of individual slope coeﬃcients using
a 5% cutoﬀ (t"1.65), 13 of 16 are signiﬁcant for the CFO and Accrual
components while only 1 of 24 are individually signiﬁcant for the remaining
EVA components. With the exception of the ‘Comp Year’ group, none of the
twotail Ftests are signiﬁcant for components unique to EVA. In the small
‘Comp Year’ sample (n"35), both CapChg and AcctAdj have signiﬁcant
Fstatistics suggesting they make an incremental contribution to explaining
contemporaneous security returns in years where ﬁrms have EVAbased
322 G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
Table 4
Tests of relative information content (H
"
): Sample partitioned by relative ‘use’ of EVAº
Relative information content'
Rank order
of R° Obs. (1) (2) (3) (4)
‘Any’ 626 EBEI ' RI ' EVA ' CFO
Adj. R` 0.0799 0.0523 0.0484 0.0317
pvalue° (0.094) (0.867) (0.550)
(0.278) (0.486)
(0.073)
‘Performance’ 445 EBEI ' EVA ' CFO ' RI
Adj. R` 0.0461 0.0386 0.0262 0.0239
pvalue (0.765) (0.699) (0.938)
(0.481) (0.491)
(0.059)
‘Comp’ 344 EBEI ' CFO ' EVA ' RI
Adj. R` 0.0306 0.0292 0.0220 0.0181
pvalue (0.962) (0.834) (0.855)
(0.735) (0.737)
(0.412)
‘Comp Year’ 35 EVA ' RI ' EBEI ' CFO
Adj. R` 0.3072 0.2644 0.2366 0.1152
pvalue (0.352) (0.667) (0.393)
(0.481) (0.330)
(0.211)
ºFirms are categorized in their use of EVA as follows:
E ‘Any’ represents all ﬁrms that have mentioned that they use EVA for performance evaluation
and/or explicit incentive compensation — even if that use appears minimal.
E ‘Performance’ represents those ﬁrms that have mentioned that they use EVA for performance
measurement but do not disclose any use of EVA in their explicit incentive compensation plans.
E ‘Comp’ respresents those ﬁrms that use EVA in their explicit incentive compensation plans and,
presumably therefore, for performance measurement. Given the existence of a plan in any year, all
available ﬁrmyears are included.
E ‘Comp Year’ represents a subset of observations from ‘Comp’ only including years where ﬁrms
have an EVAbased compensation plan in place.
‘Comp Year’ is a subset of ‘Comp’; ‘Comp’ is a subset of ‘Performance’; and ‘Performance’ is a subset
of ‘Any’.
'Underlying regressions are from Eq. (4): D
R
"b
"
#b
¹
X
R
/MVE
R¹
#b
`
X
R¹
/MVE
R¹
#e
R
where
D
R
"marketadjusted returns; X"a given performance measure (EVA, RI, EBEI, CFO); and
MVE"the market value of equity three months after the beginning of the ﬁscal year. Performance
metrics are lsited in order of Rsquares from underlying regression Eq. (4), from highest (on the left)
to lowest (on the right). Statistical tests of diﬀerences in explantory power across performance
measures are presented centered in parentheses below the adjusted Rsquares. See description of
pvalues below.
°Twotailed pvalues in parentheses represent tests of the null hyposthesis of no diﬀerence between
pairwise comparisons of adjsuted Rsquares (Biddle et al., 1995). First row presents pvalues for
comparison between ﬁrst and second ranked measures, second and third ranked measures and third
and fourth ranked measures. On the next row, comparison are between ﬁrst and third ranked, and
second and fourth ranked measures. The last row compares ﬁrst and fourth ranked measures.
G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336 323
T
a
b
l
e
5
T
e
s
t
s
o
f
i
n
c
r
e
m
e
n
t
a
l
i
n
f
o
r
m
a
t
i
o
n
c
o
n
t
e
n
t
o
f
E
V
A
c
o
m
p
o
n
e
n
t
s
(
H
'
)
:
S
a
m
p
l
e
p
a
r
t
i
t
i
o
n
e
d
b
y
r
e
l
a
t
i
v
e
‘
u
s
e
’
o
f
E
V
A
º
'
O
b
s
.
C
o
n
s
t
a
n
t
C
F
O
R
C
F
O
R
¹
A
c
c
r
u
a
l
R
A
c
c
r
u
a
l
R
¹
A
T
I
n
t
R
A
T
I
n
t
R
¹
C
a
p
C
h
g
R
C
a
p
C
h
g
R
¹
A
c
c
t
A
d
j
R
A
c
c
t
A
d
j
R
¹
A
d
j
.
R
`
P
r
e
d
i
c
t
e
d
s
i
g
n
s
:
#
!
#
!
!
#
!
#
#
!
‘
A
n
y
’
6
2
6
!
0
.
0
1
2
1
.
1
3
8
!
0
.
6
2
4
0
.
8
4
3
!
0
.
8
9
6
!
1
.
1
0
5
1
.
4
3
4
!
0
.
1
2
3
!
0
.
2
1
5
0
.
3
2
4
!
0
.
1
2
6
0
.
1
0
5
4
t

s
t
a
t
.
!
0
.
4
5
5
.
3
0
!
2
.
6
1
4
.
0
4
!
3
.
4
1
!
1
.
3
5
1
.
6
9
!
0
.
5
8
!
1
.
0
5
0
.
9
5
!
0
.
0
4
3
F

s
t
a
t
.
1
4
.
0
9
1
0
.
4
0
1
.
4
5
2
.
0
1
0
.
4
6
p

v
a
l
u
e
°
(
0
.
0
0
0
)
(
0
.
0
0
0
)
(
0
.
2
3
5
)
(
0
.
1
3
6
)
(
0
.
6
3
4
)
‘
P
e
r
f
o
r

m
a
n
c
e
’
4
4
5
!
0
.
0
4
4
1
.
0
4
7
!
0
.
3
9
6
0
.
8
3
5
!
0
.
9
1
9
!
1
.
0
3
6
0
.
7
9
1
!
0
.
1
2
9
!
0
.
0
8
7
0
.
4
8
8
!
0
.
4
9
4
0
.
0
7
8
0
t

s
t
a
t
.
!
1
.
5
2
4
.
0
8
!
1
.
5
5
3
.
5
3
!
3
.
0
7
!
1
.
0
8
0
.
7
6
!
0
.
6
0
!
0
.
4
1
1
.
3
5
!
1
.
5
4
F

s
t
a
t
.
8
.
6
4
7
.
1
7
0
.
5
9
0
.
6
2
1
.
5
4
p

v
a
l
u
e
(
0
.
0
0
0
)
(
0
.
0
0
1
)
(
0
.
5
5
7
)
(
0
.
5
3
9
)
(
0
.
2
1
6
)
‘
C
o
m
p
’
3
4
4
!
0
.
0
5
5
0
.
8
6
8
!
0
.
3
6
5
0
.
7
0
9
!
0
.
9
5
8
!
1
.
0
2
3
0
.
3
7
6
0
.
3
4
4
!
0
.
3
6
8
0
.
4
2
5
!
0
.
5
6
2
0
.
0
6
2
5
t

s
t
a
t
.
!
1
.
7
0
3
.
0
5
!
1
.
2
3
2
.
6
0
!
2
.
6
1
!
1
.
0
0
0
.
3
7
0
.
6
2
!
0
.
7
1
1
.
0
9
!
1
.
5
1
F

s
t
a
t
.
4
.
7
2
4
.
7
0
0
.
6
1
0
.
2
5
1
.
3
4
p

v
a
l
u
e
(
0
.
0
0
9
)
(
0
.
0
1
0
)
(
0
.
5
4
4
)
(
0
.
7
7
9
)
(
0
.
2
6
5
)
324 G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
‘
C
o
m
p
y
e
a
r
’
3
5
!
0
.
0
1
5
0
.
3
0
8
!
1
.
7
6
9
0
.
7
5
9
!
2
.
1
7
8
!
1
.
6
8
7
!
1
.
2
4
0
3
.
1
3
0
!
0
.
8
4
1
!
2
.
1
2
2
!
4
.
1
6
7
0
.
3
3
4
4
t

s
t
a
t
.
!
1
.
7
0
3
.
0
5
!
1
.
2
3
2
.
6
0
!
2
.
6
1
!
1
.
0
0
0
.
3
7
0
.
6
2
!
0
.
7
1
1
.
0
9
!
1
.
5
1
F

s
t
a
t
.
0
.
8
7
2
.
0
2
1
.
2
6
4
.
0
4
3
.
6
6
p

v
a
l
u
e
(
0
.
4
3
0
)
(
0
.
1
5
4
)
(
0
.
3
0
3
)
(
0
.
0
3
1
)
(
0
.
4
1
)
º
F
i
r
m
s
a
r
e
c
a
t
e
g
o
r
i
z
e
d
i
n
t
h
e
i
r
u
s
e
o
f
E
V
A
a
s
f
o
l
l
o
w
s
:
‘
A
n
y
’
r
e
p
r
e
s
e
n
t
s
a
l
l
ﬁ
r
m
s
t
h
a
t
h
a
v
e
m
a
d
e
a
n
y
m
e
n
t
i
o
n
o
f
u
s
i
n
g
E
V
A
f
o
r
p
e
r
f
o
r
m
a
n
c
e
e
v
a
l
u
a
t
i
o
n
a
n
d
/
o
r
e
x
p
l
i
c
i
t
i
n
c
e
n
t
i
v
e
c
o
m
p
e
n
s
a
t
i
o
n
—
e
v
e
n
i
f
t
h
a
t
u
s
e
a
p
p
e
a
r
s
m
i
n
i
m
a
l
.
‘
P
e
r
f
o
r
m
a
n
c
e
’
r
e
p
r
e
s
e
n
t
s
t
h
o
s
e
ﬁ
r
m
s
t
h
a
t
h
a
v
e
m
e
n
t
i
o
n
e
d
t
h
a
t
t
h
e
y
u
s
e
E
V
A
f
o
r
p
e
r
f
o
r
m
a
n
c
e
m
e
a
s
u
r
e
m
e
n
t
b
u
t
d
o
n
o
t
d
i
s
c
l
o
s
e
a
n
y
u
s
e
o
f
E
V
A
i
n
t
h
e
i
r
e
x
p
l
i
c
i
t
i
n
c
e
n
t
i
v
e
c
o
m
p
e
n
s
a
t
i
o
n
p
l
a
n
s
.
‘
C
o
m
p
’
i
n
c
l
u
d
e
s
a
l
l
l
ﬁ
r
m

y
e
a
r
o
b
s
e
r
v
a
t
i
o
n
s
f
o
r
t
h
o
s
e
ﬁ
r
m
s
t
h
a
t
u
s
e
E
V
A
i
n
t
h
e
i
r
e
x
p
l
i
c
i
t
i
n
c
e
n
t
i
v
e
c
o
m
p
e
n
s
a
t
i
o
n
p
l
a
n
s
i
n
a
n
y
y
e
a
r
.
‘
C
o
m
p
Y
e
a
r
’
i
n
c
l
u
d
e
s
a
s
u
b
s
e
t
o
f
o
b
s
e
r
v
a
t
i
o
n
s
f
r
o
m
‘
C
o
m
p
’
f
o
r
o
n
l
y
t
h
o
s
e
y
e
a
r
s
w
h
e
r
e
t
h
e
E
V
A

b
a
s
e
d
c
o
m
p
e
n
s
a
t
i
o
n
p
l
a
n
i
s
i
n
e
ﬀ
e
c
t
.
‘
C
o
m
p
Y
e
a
r
’
i
s
a
s
u
b
s
e
t
o
f
‘
C
o
m
p
’
;
‘
C
o
m
p
’
i
s
a
s
a
s
u
b
s
e
t
o
f
‘
P
e
r
f
o
r
m
a
n
c
e
’
;
a
n
d
‘
P
e
r
f
o
r
m
a
n
c
e
’
i
s
a
s
u
b
s
e
t
o
f
‘
A
n
y
’
.
'
D
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
"
m
a
r
k
e
t

a
d
j
u
s
t
e
d
r
e
t
u
r
n
s
;
i
n
d
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
s
a
r
e
c
o
m
p
o
n
e
n
t
s
o
f
E
V
A
(
C
F
O
,
o
p
e
r
a
t
i
n
g
a
c
c
r
u
a
l
s
,
a
f
t
e
r

t
a
x
i
n
t
e
r
e
s
t
e
x
p
e
n
s
e
,
c
a
p
i
t
a
l
c
h
a
r
g
e
,
a
c
c
o
u
n
t
i
n
g
a
d
j
u
s
t
m
e
n
t
s
)
a
n
d
a
r
e
s
h
o
w
n
i
n
n
o
n

l
a
g
g
e
d
a
n
d
l
a
g
g
e
d
f
o
r
m
s
a
s
c
o
l
u
m
n
h
e
a
d
i
n
g
s
.
°
P

v
a
l
u
e
s
i
n
p
a
r
e
n
t
h
e
s
e
s
r
e
p
r
e
s
e
n
t
n
o
n

d
i
r
e
c
t
i
o
n
a
l
F

t
e
s
t
s
o
f
t
h
e
n
u
l
l
h
y
p
o
t
h
e
s
i
s
o
f
n
o
i
n
c
r
e
m
e
n
t
a
l
i
n
f
o
r
m
a
t
i
o
n
c
o
n
t
e
n
t
(
h
y
p
o
t
h
e
s
i
s
H
¹
)
.
G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336 325
¹` We also evaluated the relative and incremental information content using changes in (rather
than sums of) each performance measure over the ﬁveyear period. In relative information content
tests, once again EBEI outperformed EVA. In incremental information content tests, only AcctAdj
for the 1988—93 subperiod was signiﬁcant at the 5% level while none of the other components
unique to EVA were signiﬁcant in the predicted direction.
compensation plans in eﬀect. However, caution is warranted in drawing any
inferences from this result due to: the small size of the ‘Comp Year’ sample, the
surprising insigniﬁcant Fstatistics on CFO and Accrual, and the unexpected
signs on coeﬃcients on CapChg. Again, other than weakly suggestive results for
the ‘Comp Year’ sample, it does not appear users of EVA are adopting the
concept because of its stronger association with stock returns.
5.3. Fiveyear returns as the dependent variable
In this section we extend the return interval from one year to ﬁve years. Stern
Stewart reports its strongest results for EVA on ﬁveyear data (Stewart, 1991,
1994). In addition, because ﬁveyear data are less sensitive to the choice of
expectations models, these tests help address the possibility that the weaker
performance of EVAis due to a poorer expectations model. Regression (8) below
is used to evaluate relative information content comparisons. It includes non
lagged and lagged terms similar to the annual return regression (4) and are
analogous to the ‘level and changes’ speciﬁcation discussed in Section 2.3 and
footnote 9. Independent variables reﬂect ‘ﬁveyear sums’ in that each perfor
mance measure, X
'
, is summed over the most recent ﬁveyear period 1989—93
(for the nonlagged term) and summed over the prior ﬁve years, 1984—88 (for the
lagged term).
‘5year sums’: MktAdjRet
R
"b
"
#b
¹
X
R
/M»E
R`
#b
`
X
R`
/M»E
R`
#e
R
. (8)
Since all ten years of data are used to examine the association between ﬁveyear
returns and each performance measure, only one test period is reported in
Table 6. Results again show the earnings regression with the highest R` (31.2%)
followed by CFO (18.9%), EVA (14.5%) and RI (10.9%). The diﬀerences in
explanatory power between EBEI and each of the other three performance
measures are highly signiﬁcant.
In Table 7 we report incremental information content of EVA components
after extending the return interval from one year to ﬁve years. CFO and Accrual
are again highly signiﬁcant but the results on components unique to EVA
(CapChg and AcctAdj) are insigniﬁcant.¹`
326 G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
¹" We also examined the change in ‘market value added’ (deﬁned by Stern Stewart as ﬁrm market
value less invested capital) as a dependent variable with qualitatively similar ﬁndings.
Table 6
Tests of relative information content (H
"
): Returns measured over 5year periodsº
Relative information content
Rank order
of R` Obs. (1) (2) (3) (4)
‘5year sums’ 509 EBEI ' CFO ' EVA ' RI
Adj. R` 0.3118 0.1888 0.1446 0.1090
pvalue' (0.005) (0.264) (0.030)
(0.000) (0.051)
(0.000)
ºUnderlying regression is from Eq. (8):
‘5year sums’: D
R
"b
"
#b
¹
X
R
/MVE
R`
#b
`
X
R`
/MVE
R`
#e
R
,
where is deﬁned over the ﬁveyear intervals, 1989—93 (nonlag) and 1948—88 (lagged) terms,
respectively;
D
R
"marketadjusted returns measured over ﬁve years; X"a given performance measure (CFO,
EBEI, RI or EVA); and
MVE"market value of equity three months after the beginning of the ﬁscal year.
'pvalue in parentheses represent twotail tests of the null hypothesis of no diﬀerence between
pairwise comparisons of adjusted Rsquares (Biddle et al., 1995). First row present pvalues for
comparison between ﬁrst and second ranked measures, second and third ranked measures and third
and fourth ranked measures. On the next row, comparisons are between ﬁrst and third ranked, and
second and fourth ranked measures. The last row compares ﬁrst and fourth ranked measures.
5.4. Twoyear (contemporaneous and oneyear ahead) returns as
the dependent variable
To consider the possibility that equity market participants take longer to
learn about and impound EVA, we extend the return interval from the oneyear
contemporaneous period used above to a twoyear period that includes both the
contemporaneous and subsequent year. Consistent with results in Table 2, and
inconsistent with the conjecture that the market subsequently learns about the
importance of EVA, EBEI has signiﬁcantly higher association with twoyear
returns (adjusted R`"4.4%) than any of the other three information variables
(whose R`s range from 2% to 2.3%).
5.5. Market value of the ﬁrm as the dependent variable
Another claim made by Stern Stewart is EVA’s higher association with the
market value of the ﬁrm.¹" To test this claim, we replicate and extend a study
G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336 327
T
a
b
l
e
7
T
e
s
t
s
o
f
i
n
c
r
e
m
e
n
t
a
l
i
n
f
o
r
m
a
t
i
o
n
c
o
n
t
e
n
t
o
f
E
V
A
c
o
m
p
o
n
e
n
t
s
(
H
¹
)
:
R
e
t
u
r
n
s
m
e
a
s
u
r
e
d
o
v
e
r
ﬁ
v
e

y
e
a
r
p
e
r
i
o
d
s
º
'
O
b
s
.
C
o
n
s
t
a
n
t
C
F
O
R
C
F
O
R
¹
A
c
c
r
u
a
l
R
A
c
c
r
u
a
l
R
¹
A
T
I
n
t
R
A
T
I
n
t
R
¹
C
a
p
C
h
g
R
C
a
p
C
h
g
R
¹
A
c
c
t
A
d
j
R
A
c
c
t
A
d
j
R
¹
A
d
j
.
R
`
P
r
e
d
i
c
t
e
d
s
i
g
n
s
#
!
#
!
!
#
!
#
#
!
‘
5

y
e
a
r
s
u
m
s
’
5
0
9
!
0
.
3
7
3
2
.
1
2
8
!
0
.
7
3
1
1
.
6
5
9
!
0
.
0
7
2
!
0
.
5
0
9
0
.
0
8
9
!
0
.
0
8
8
0
.
2
7
5
0
.
5
4
9
0
.
4
8
7
0
.
3
2
4
1
t

s
t
a
t
.
!
2
.
5
4
7
.
9
3
!
2
.
5
6
5
.
9
9
!
0
.
2
5
!
0
.
1
1
0
.
1
7
!
0
.
4
2
1
.
2
1
1
.
3
9
1
.
6
4
F

s
t
a
t
.
3
3
.
1
6
1
8
.
0
1
0
.
0
1
0
.
8
1
2
.
7
5
p

v
a
l
u
e
°
(
0
.
0
0
0
)
(
0
.
0
0
0
)
(
0
.
9
8
6
)
(
0
.
4
4
7
)
(
0
.
0
6
5
)
º
D
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
"
m
a
r
k
e
t

a
d
j
u
s
t
e
d
r
e
t
u
r
n
s
;
i
n
d
e
p
e
n
d
e
n
t
v
a
r
i
b
l
e
s
(
E
V
A
c
o
m
p
o
n
e
n
t
s
)
a
r
e
s
h
o
w
n
a
c
o
l
u
m
n
h
e
a
d
i
n
g
s
.
'
U
n
d
e
r
l
y
i
n
g
r
e
g
r
e
s
s
i
o
n
a
s
f
o
l
l
o
w
s
‘
5

y
e
a
r
s
u
m
s
’
:
D
R
"
b
"
#
b
¹
X
R
/
M
V
E
R
`
#
b
`
X
R
`
/
M
V
E
R
`
#
b
`
X
R
/
M
V
E
R
`
#
b
"
X
R
`
/
M
V
E
R
`
#
2
#
b
"
X
R
/
M
V
E
R
`
#
b
¹
"
X
R
`
/
M
V
E
R
`
#
e
R
,
w
h
e
r
e
i
s
d
e
ﬁ
n
e
d
o
v
e
r
t
h
e
ﬁ
v
e

y
e
a
r
i
n
t
e
r
v
a
l
s
,
1
9
8
9
—
9
3
(
l
a
g
g
e
d
)
t
e
r
m
s
,
r
e
s
p
e
c
t
i
v
e
l
y
;
D
R
"
m
a
r
k
e
t

a
d
j
u
s
t
e
d
r
e
t
u
r
n
s
;
X
R
"
a
g
i
v
e
n
E
V
A
c
o
m
p
o
n
e
n
t
,
i
.
e
.
,
C
F
O
,
A
c
c
r
u
a
l
,
A
T
I
n
t
R
,
C
a
p
C
h
g
a
n
d
A
c
c
t
A
d
j
;
a
n
d
M
V
E
"
m
a
r
k
e
t
v
a
l
u
e
o
f
e
q
u
i
t
y
t
h
r
e
e
m
o
n
t
h
s
f
o
l
l
o
w
i
n
g
t
h
e
b
e
g
i
n
n
i
n
g
o
f
t
h
e
ﬁ
s
c
a
l
y
e
a
r
.
°
p

v
a
l
u
e
s
i
n
p
a
r
e
n
t
h
e
s
e
s
r
e
p
r
e
s
e
n
t
n
o
n

d
i
r
e
c
t
i
o
n
a
l
F

t
e
s
t
s
o
f
t
h
e
n
u
l
l
h
y
p
o
t
h
e
s
i
s
o
f
n
o
i
n
c
r
e
m
e
n
t
a
l
i
n
f
o
r
m
a
t
i
o
n
c
o
n
t
e
n
t
(
h
y
p
o
t
h
e
s
i
s
H
¹
)
.
328 G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
`" O’Byrne scales EVA by k and capital and NOPAT only by Capital. We cannot replicate results
for free cash ﬂow because O’Byrne (1996) does not provide a precise deﬁnition.
`¹ O’Byrne argues that a nonzero intercept makes predicted M» a function of Capital and
therefore an EVA model in disguise. If so, it would follow from our ﬁndings that EBEI is a better
proxy for future EVA than is EVA.
authored by Stern Stewart vicepresident Stephen O’Byrne (1996). There are
three main diﬀerences between O’Byrne’s research and our tests reported above.
First, O’Byrne uses market value of the ﬁrm (debt plus equity) as the dependent
variable while we use marketadjusted returns. Second, he draws inferences by
comparing the magnitudes of R`s, while we draw inferences by relying on formal
statistical tests of relative information content. Third, and in our view most
importantly, O’Byrne makes a series of ‘adjustments’ only to the EVA regres
sions and uses R`s from these adjusted regressions to infer superiority of EVA
over competing information variables.
The initial relations tested in O’Byrne (before ‘adjustments’) are:`"
M»
R
/capital
R¹
"
"
#
¹
(EVA
R
/k)/Capital
R¹
#e
R
(9)
M»
R
/capital
R¹
"
"
#
¹
(NOPAT
R
)/Capital
R¹
#e
R
(10)
where
M»t/capital
R¹
the market value of debt plus equity deﬂated by beginning of
period capital.
EVA
R
economic value added for year t, i.e., NOPA¹
R
—k (capital
R¹
)
NOPAT
R
net operating proﬁts after tax for year t.
k Stern Stewart’s estimate of the ﬁrm’s weighted average cost of
capital.
Capital
R¹
Stern Stewart’s deﬁnition of assets (net of depreciation) inves
ted in goingconcern operating activities, or equivalently, con
tributed and retained debt and equity capital, at the beginning
of period t.
e
R
unexplained residual error.
O’Byrne compares R`s from the initial two models and reports 31% for the
EVA model (9) and 33% for the NOPAT model (10). Next, he makes a series of
adjustments to the EVA regressions by: (1) allowing separate coeﬃcients for
positive and negative values of EVA, (2) including the natural log of capital in
an attempt to capture diﬀerences in the way the market values ﬁrms of diﬀerent
sizes, and (3) including 57 industry dummy variables in order to capture
potential industry eﬀects. None of these adjustments are made for the NOPAT
regression, and O’Byrne further argues that a pure NOPAT model should be
forced through the origin (p. 120).`¹ After these adjustments, he reports a much
G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336 329
`` O’Byrne reports adjusted R` of 42% for the EVA model omitting industry intercept dummies.
`` Omitting scaling by the cost of capital (k) from regression Eq. (11) yields R`s of 53% for EBEI,
51% for EVA and 51% for NOPAT. Omitting industry dummies from Eq. (11) (while retaining
scaling by k) yields R`s of 47% for EBEI, 43% for EVA and 41% for NOPAT.
`" These results are not directly comparable with the returns results above since they employ
diﬀerent functional forms and additional variables due to O’Byrne (1996).
higher R` for the ﬁnal model containing EVA (56%) than for the ﬁnal model
containing NOPAT (which, because of the intercept restriction, falls to 17%).``
O’Byrne (1996) (p. 125) concludes:
“EVA, unlike NOPAT or other earnings measures like net income or
earnings per share, is systematically linked to market value. It should provide
a better predictor of market value than other measures of operating perfor
mance. And, as we have shown, it does provide a better predictor once we
understand and adjust for two critical relationships between EVA and market
value.”
Given the success of earnings in our returns tests discussed above, we add
EBEI to the consideration set and replicate O’Byrne’s ﬁnal model using 5,843
ﬁrmyear observations obtained from Stern Stewart as described earlier. In
Table 8, we treat EVA, NOPAT and EBEI as competing performance measures
and apply O’Byrne’s three adjustments to each variable (as described in regres
sion Eq. (11) in a note to the table). With this ‘level playing ﬁeld,’ EVA’s
superiority disappears. With all of O’Byrne’s adjustments (including industry
dummies), the EBEI regression has a signiﬁcantly higher association with ﬁrm
value (adjusted R`"53%) than the EVA regression (50%).`` After making the
same adjustments to the NOPAT regression, the R` of 49% is not signiﬁcantly
diﬀerent from the EVA regression. Thus, similar to results reported for our
returns tests above, results in Table 8 provide no evidence of the EVAregression
outperforming earnings in explaining deﬂated ﬁrm values.`"
5.6. Overall assessment of the sensitivity tests
Considering jointly the sensitivity analyses of relative information content
discussed in Sections 5.1, 5.2, 5.3, 5.4 and 5.5, we still ﬁnd no evidence to support
the Stern Stewart claim that EVA (or RI) outperform EBEI. In only one case
(the ‘Comp Year’ group in Table 4) does EVA and/or RI have a higher R` than
EBEI and this diﬀerence is not statistically signiﬁcant. In contrast, adjusted R` is
highest for EBEI in the remaining comparisons and EBEI signiﬁcantly out
performs EVA in several sensitivity tests at the 5% level.
330 G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
Table 8
Replication and extension of O’Byrne (1996): Tests of relative information content (H
"
) for EVA,
NOPAT and EBEI where the dependent variable is the market value of the ﬁrmº
Relative information content
Rank order of R` Obs. (1) (2) (3)
Sample size 5,843 EBEI ' EVA ' NOPAT
Adj. R` 0.5321 0.4965 0.4886
pvalue' (0.000) (0.413)
(0.000)
ºUnderlying regression is from O’Byrne (1996):
M»
R
/capital
R¹
"b
"
#b
¹
½
R NMQ
/capital
R¹
#b
`
½
R LCE
/capital
R¹
#b
`
ln(capital
R¹
)
#b
H
I
H
#e
R
, (11)
where M»"market value of debt and equity; ½"a given performance measure (EVA, NOPAT,
NI) deﬂated by the ﬁrm’s cost of capital (as estimated by Stern Stewart), where pos and neg refer to
positive and negative values of the performance measure, respectively; capital
R¹
"the ﬁrms’s
beginning of period contributed capital; and I is a dummy variable representing industry member
ship.
Performance metrics are listed in order of Rsquares fromhighest (on the left) to lowest (on the right).
Statistical tests of diﬀerences in explanatory power across performance measures are presented in
parentheses below the adjusted Rsquares. See description of pvalues below.
'Twotailed pvalues in parentheses represent tests of the null hypothesis of no diﬀerence between
pairwise comparisons of adjusted Rsquares (Biddle et al., 1995). First row present pvalues for
comparison between ﬁrst and second ranked measures and second and third ranked measures. On
the next row, comparisons are between ﬁrst and third ranked. Tests based on Vuong (1989) are
qualitatively identical.
In terms of incremental information content, the analyses in Section 5 provide
only limited evidence that components unique to EVA (i.e., CapChg and
AcctAdj) add to the information set that includes earnings (i.e., CFO and
Accruals), e.g., only two of the Fstatistics and none of the tstatistics on CapChg
and AcctAdj are signiﬁcant at the 5% level. Thus, from the sensitivity tests in
Section 5, we ﬁnd no evidence that EVA dominates earnings in its association
with stock returns or ﬁrm values.
6. Summary and potential limitations
Motivated by increased use in practice and increased interest in the media and
among academics, we examine the valuerelevance of EVA and residual income
compared to currentlymandated performance measures — earnings and cash
ﬂow from operations. There is little evidence to support the Stern Stewart claim
G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336 331
that EVA is superior to earnings in its association with stock returns or ﬁrm
values. In no case does EVA signiﬁcantly outperform EBEI in tests of relative
information content. On the contrary, in most cases the evidence suggests that
earnings outperforms EVA. Further, while the charge for capital and Stern
Stewart’s adjustments for accounting ‘distortions’ show some marginal evidence
of being incrementally important, this diﬀerence does not appear to be economi
cally signiﬁcant. Possible reasons why we do not detect stronger valuerelevance
for EVA include:
E Our research design uses current realizations, not future ﬂows, of each
performance measure. Equity valuation is ultimately the discounted present
value of future equity cash ﬂows (or dividends or RI or EVA). Even if EVA is
a good proxy for economic proﬁts, realized EVA may not outperform the
current realizations of other performance measures such as earnings in
proxying for future equity cash ﬂows. This is similar to the rationale used
to explain why EBEI generally outperforms CFO in relative information
content.
E Stern Stewart’s estimates of the charge for capital and accounting adjust
ments may contain measurement error relative to what the market is
using to value ﬁrms. Further, we use Stern Stewart’s publicly available
database which does not include many custom adjustments they use for their
clients.
E There exists little or no ‘surprise value’ in components unique to EVA
including the capital charge and Stern Stewart’s accounting adjustments. For
example, if the cost of capital and the amount of capital are slow to change (or
the changes are predictable months or years in advance), the market should
long ago have impounded these data. However, over ﬁveyear return inter
vals, the opportunity for surprise should be larger, and results reported in
Section 5.3 do not lend support for the superiority of EVA over this longer
return interval.
E Data needed to compute EVA are not easily estimated and the market does
not have these data during our test period. Recall that we assume that the
market has access to suﬃcient data within three months of a ﬁrm’s ﬁscal year
end such that EVA (and its components) can be reliably estimated by that
time. This potential issue is mitigated in tests that use alternative dependent
variables (i.e., ﬁveyear return intervals in Section 5.3, twoyear return inter
vals that include both contemporaneous and oneyear ahead returns in
Section 5.4, and ﬁrm values in Section 5.5). Again the evidence does not
support the superiority of EVA.
E In attempting to approximate economic proﬁts, adjustments made by Stern
Stewart may remove accruals that market participants use to infer ﬁrms’
future prospects. These could be discretionary accruals that managers use to
‘signal’ future prospects or nondiscretionary accruals that are byproducts of
332 G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
`` Collins and DeAngelo (1990), Subramanyam (1996), and Hunt et al. (1997) report evidence
consistent with discretionary accruals increasing the informativeness of earnings. Wu (1997) presents
an agency model in which ﬁrms will optimally choose accounting adjustments for internal perfor
mance metrics that serve to reduce their correlation with stock returns.
`" Wallace (1996, 1998) reports that some adopters of EVA feel they must still base their external
performance on earnings because this is the measure on which ﬁnancial analysts continue to
focus.
the accounting process.`` Thus, in constructing EVA, it is possible that
Stern Stewart obtains a measure that is closer in level to economic proﬁts
(than say EBEI), but at the same time reduces its association with stock
returns.
E In violation of our maintained hypothesis of semistrong market eﬃciency,
the market may have failed to recognize the reporting beneﬁts of EVA
through the period we study, consistent with the notion of ‘earnings
myopia’.`" As more data become available, future studies will be able to
assess whether market participants have come to appreciate EVA. It also is
possible to imagine a new equilibrium in which ﬁrms would disclose EVA
rather than earnings. However, this would subject EVA to many of the same
legal and regulatory inﬂuences, and as a consequence, the resulting metric
might closely resemble earnings (or what earnings might become). For this
reason, and despite its alleged advantages for internal decision making, we do
not anticipate that EVA will displace earnings for ﬁnancial reporting pur
poses.
Until further research can be conducted, our conclusion is that, although for
some ﬁrms EVA may be an eﬀective tool for internal decision making, perfor
mance measurement and incentive compensation, it does not dominate earnings
in its association with stock market returns for the sample ﬁrms and period
studied. To the contrary, and in contrast to claims by Stern Stewart, our
evidence suggests that earnings generally dominates EVA in valuerelevance to
market participants.
An avenue for future research suggested by the ﬁndings of this study is to
examine more closely which components of EVA and earnings contribute to, or
subtract from, information content. For example, ex ante, capitalizing R&D and
marketing costs should only add to EVA’s information content given both are
generally expensed in the determination of earnings. In contrast, by estimating
taxes paid in cash (rather than tax expense), EVA may lose information content
by removing valuerelevant deferred tax accruals. Research at this level of detail
requires data that are currently unavailable from Stern Stewart on individual
adjustments used in the calculation of EVA.
G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336 333
Acknowledgements
Helpful comments were provided by Bill Alberts, Mary Barth, William
Beaver, Dave Burgstahler, George Foster, Bob Holthausen (the referee), Alister
Hunt, Jane Kennedy, Dawn Matsumoto, Brian McCulloch, Jim Patell, Mark
Peecher, Terry Shevlin, D. Shores, Naomi Soderstrom, Jerry Zimmerman, and
workshop participants at the Chinese University of Hong Kong, City University
of Hong Kong, European Accounting Association Meetings, Hong Kong Uni
versity of Science & Technology, National Chengchi University, Northwestern
University, New York University, Stanford University, and the University of
Washington. Funding was provided by the Accounting Development Fund at
the University of Washington, the Herbert O. Whitten Professorship and
a Hong Kong Research Grants Council Direct Allocation Grant.
References
American Institute of Certiﬁed Public Accountants, Special committee on ﬁnancial reporting, 1994.
Improving business reporting — a customer focus, comprehensive report of the special committee
on ﬁnancial reporting, American Institute of Certiﬁed Public Accountants.
Anthony, R., 1973. Accounting for the cost of equity. Harvard Business Review 51, 88—102.
Anthony, R., 1982a. Equity interest — its time has come. Journal of Accountancy 154, 76—93.
Anthony, R., 1982b. Recognizing the cost of interest on equity. Harvard Business Review 60,
91—96.
Biddle, G., Seow, G., 1991. The estimation and determinants of association between returns and
earnings: Evidence from crossindustry comparisons. Journal of Accounting, Auditing and Fi
nance 6, 183—232.
Biddle, G., Seow, G., Siegel, A., 1995. Relative versus incremental information content. Contempor
ary Accounting Research 12, 1—23.
Biddle, G., Siegel, A., 1996. Testing relative information using nested likelihood models. Unpub
lished paper, University of Washington, Seattle, WA.
Bowen, R., Burgstahler, D., Daley, L., 1986. Evidence on the relationships between earnings and
various measures of cash ﬂow. The Accounting Review 61, 713—725.
Bowen, R., Burgstahler, D., Daley, L., 1987. The incremental information content of accrual versus
cash ﬂows. The Accounting Review 62, 723—747.
Bowen, R., Johnson, M., Shevlin, T., Shores, D., 1989. Informational eﬃciency and the information
content of earnings during the stock market crash of 1987. Journal of Accounting and Economics
11, 225—254.
Burgstahler, D., Dichev, I., 1997. Earnings, adaptation and equity value. The Accounting Review 72,
187—215.
Canning, J.B., 1929. The Economics of Accountancy. Ronald Press, New York.
Collins, D.W., DeAngelo, L., 1990. Accounting information and corporate governance: market and
analyst reactions to earnings of ﬁrms engaged in proxy contests. Journal of Accounting and
Economics 13, 213—247.
Collins, D., Pincus, M., Xie, H., 1997. Equity valuation and negative earnings: The role of book value
of equity, Unpublished paper, University of Iowa, Iowa City, IA.
Davidson, R., MacKinnon, J., 1981. Estimation and Inference in Econometrics. Oxford University
Press, New York.
334 G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
Dechow, P., Lys, T., Sabino, J., 1996. Using longwindow return studies in addressing recognition
issues: An evaluation of alternative research approaches. Unpublished paper, University of
Pennsylvania, Philadelphia, PA.
Easton, P., Harris, T., 1991. Earnings as an explanatory variable for returns. Journal of Accounting
Research 29, 19—36.
Edey, H.C., 1957. Business valuation, goodwill and the superproﬁt method, Accountancy, Jan
uary/February.
Edwards, E.O., Bell, P.W., 1961. The Theory and Measurement of Business Income. University of
California Press, Berkeley.
Feltham, G.A., Ohlson, J.A., 1995. Valuation and clean surplus accounting for operating and
ﬁnancing activities. Contemporary Accounting Research 11, 689—731.
Fisher, A., 1995. Creating stockholder wealth. Fortune 132, 105—117.
Hamilton, R., 1777. An Introduction to Merchandize. Edinburgh.
Hayn, C., 1995. The information content of losses. Journal of Accounting and Economics 20,
125—153.
Hotelling, H., 1940. The selection of variates for use in prediction with some comments
on the general problem of nuisance parameters. The Annals of Mathematical Statistics,
271—283.
Hunt, A., Moyer, S. E., Shevlin, T., 1997. Earnings volatility, earnings management, and equity
value, Unpublished paper, University of Washington, Seattle, WA.
Kay, J.A., 1976. Accountants, too, could be happy in a golden age: The accountant’s rate of proﬁt
and the internal rate of return. Oxford Economic Papers 28, 447—460.
Kennedy, P., 1985. A Guide to Econometrics. The MIT Press, Cambridge, MA.
Lieber, R., 1996. Who are the real wealth creators? Fortune, 134, 107—116.
Marshall, A., 1890. Principles of Economics. The Macmillan Press Ltd, London, New York.
O’Byrne, S., 1996. EVA and market value. Journal of Applied Corporate Finance 9, 116—125.
Ohlson, J., 1995. Earnings, book value, and dividends in security valuation. Contemporary
Accounting Research 11, 661—687.
Peasnell, K.V., 1981. On capital budgeting and income measurement. ABACUS 17, 52—67.
Peasnell, K.V., 1982. Some formal connections between economic values and yields and accounting
numbers. Journal of Business Finance and Accounting 9, 361—381.
Preinreich, G.A.D., 1936. The law of goodwill. The Accounting Review 12, 317—329.
Preinreich, G.A.D., 1937. Goodwill in accountancy. The Journal of Accountancy, July, 28—50.
Preinreich, G.A.D., 1938. Annual survey of economic theory: the theory of depreciation, Econo
metrica, January, 219—231.
Solomons, D., 1965. Division Performance: Measurement and Control. Financial Executives
Research Foundation, New York.
Stern Stewart EVA Roundtable, 1994. Journal of Applied Corporate Finance 7, 46—70.
Stewart III, G.B., 1991. The Quest For Value. Harper Business, New York.
Stewart III, G.B., 1994. EVA: fact or fantasy? Journal of Applied Corporate Finance 7, 71—84.
Subramanyam, K.R., 1996. The pricing of discretionary accruals. Journal of Accounting and
Economics 22, 250—281.
Teitelbaum, R., 1997. America’s greatest wealth creators. Fortune 136, 265—276.
Tully, S., 1993. The real key to creating wealth. Fortune 128, 38—50.
Tully, S., 1994. America’s best wealth creators. Fortune 130, 143—162.
Vuong, Q., 1989. Likelihood ratio tests for model selection and nonnested hypotheses. Econo
metrica 57, 307—333.
Walbert, L., 1994. The Stern Stewart performance 1000: Using EVA to build market value. Journal
of Applied Corporate Finance 6, 109—120.
Wallace, J., 1996. Adopting Residual IncomeBased Compensation Plans: Evidence of Eﬀects on
Management Actions. UMI Dissertation Services, Ann Arbor.
G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336 335
Wallace, J., 1997. Adopting residual incomebased compensation plans: Do you get what you pay
for? Journal of Accounting and Economics 24, (this issue).
Wallace, J., 1998. EVA ﬁnancial systems: Management perspectives. Advances in Management
Accounting 6, 1—15.
White, H., 1980. A heteroskedasticityconsistent covariance matrix estimator and a direct test for
heteroskedasticity. Econometrica 48, 817—838.
Wu, M., 1997. Who makes accounting ‘adjustments’ and why? An agency perspective. Unpublished
paper, New York University, New York, NY.
Zarowin, S., 1995. The future of ﬁnance. Journal of Accountancy 180 (August) 47—49.
336 G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
302
G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
Marshall, 1890). In the twentieth century, this concept has been operationalized under various labels including residual income. Residual income has been recommended as an internal measure of businessunit performance (Solomons, 1965) and as an external performance measure for ﬁnancial reporting (Anthony, 1973, 1982a,b). General Motors applied this concept in the 1920s and General Electric coined the term ‘residual income’ in the 1950s and used it to assess the performance of its decentralized divisions (Stern Stewart EVA Roundtable, 1994). More recently, Stern Stewart & Company has advocated that a trademarked variant of residual income, economic value added (EVA ), be used instead of earnings or cash from operations as a measure of both internal and external performance. They argue: “Abandon earnings per share” (Stewart, 1991) (p. 2). “Earnings, earnings per share, and earnings growth are misleading measures of corporate performance” (Stewart, 1991), (p. 66). “The best practical periodic performance measure is economic value added (EVA)” (Stewart, 1991 (p. 66). “Forget EPS, ROE and ROI. EVA is what drives stock prices” (Stern Stewart advertisement in Harvard Business Review, November—December, 1995, p. 20). Stewart (1994) cites inhouse research indicating that “EVA stands well out from the crowd as the single best measure of wealth creation on a contemporaneous basis” and “EVA is almost 50% better than its closest accountingbased competitor in explaining changes in shareholder wealth” (p. 75). This study provides independent empirical evidence on the information content of EVA, residual income, and two mandated performance measures, earnings and cash ﬂow from operations. Our inquiry is motivated by: the claims cited above, interest in EVA in the business press, increasing use of EVA by ﬁrms, increasing interest in EVA among academics, and potential interest in EVA among accounting policy makers. Citations of EVA in the business press have grown exponentially, rising from 1 in 1989 to 294 in 1996 (Lexis/Nexis ‘allnews’ library). Fortune has touted EVA as “The Real Key to Creating Wealth” (30 September 1993), “A New Way to Find Bargains” (9 December 1996), and has begun augmenting its wellknown ‘500’ ranking with an annual
Residual income is generally deﬁned as aftertax operating proﬁts less a charge for invested capital. Operating proﬁts are proﬁts before deducting the aftertax cost of interest expense. The ﬁrm’s weighted average cost of debt and equity capital is deducted in the capital charge. Other labels include: abnormal earnings (Feltham and Ohlson, 1995); excess earnings (Canning, 1929, Preinreich, 1936, 1937, 1938); excess income (Kay, 1976; Peasnell, 1981, 1982); excess realizable proﬁt (Edwards and Bell, 1961); and superproﬁts (Edey, 1957). Stern Stewart & Company is a New Yorkbased consulting ﬁrm that markets the ‘EVA Financial Management System’ for internal and external performance measurement and incentive compensation. Performance measures marketed by competing ﬁrms include cashﬂow return on investment (CFROI) by Boston Consulting Group’s HOLT Value Associates, discounted cashﬂow analysis (DCA) by Alcar, discounted economic proﬁts (EP) by Marakon Associates, and economic value management (EVM) by KPMG Peat Marwick.
G.C. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336
303
‘Performance 1000’ based on data from Stern Stewart (Tully, 1993, 1994; Fisher, 1995; Lieber, 1996; Teitelbaum, 1997). Companies that have adopted EVA for performance measurement and/or incentive compensation include AT&T, Coca Cola, Eli Lilly, Georgia Paciﬁc, Polaroid, Quaker Oats, Sprint, Teledyne and Tenneco. The ‘EVA Financial Management System’ is alleged to encourage managers to act more like owners by helping managers make improved operating, ﬁnancing and investment decisions. Evidence provided in Wallace (1997) suggests that managers compensated on the basis of EVA (instead of earnings) take actions consistent with EVAbased incentives. Recently, academics have shown interest in models of equity valuation that express ﬁrm value in terms of book value and the expected stream of residual income or ‘abnormal earnings’ (Ohlson, 1995; Feltham and Ohlson, 1995). Our study provides empirical evidence on whether current period realizations of residual income (RI) and EVA are more closely associated with stock returns than are traditional accounting measures such as earnings and cash from operations (CFO). Finally, data on the information content of EVA and RI provide potentially useful input to the normative policy debate on what performance measure(s) should be reported in ﬁnancial statements. Financial reporting has been criticized for lowquality and lack of relevance in today’s informationrich environment. The AICPA Special Committee on Financial Reporting (1994), the Jenkins Committee, makes suggestions for improving ﬁnancial reporting that are consistent with ﬁrms using EVA for internal decision making and external reporting. A prediction from an April 1995 AICPA workshop on the future of ﬁnancial management is that EVA will replace EPS in ¹he ¼all Street Journal’s regular stock and earnings reports (Zarowin, 1995) (p. 48). Widespread interest in revisiting the quality of ﬁnancial reporting suggests that alternatives to currently mandated performance measures should be evaluated for
CFO Basil Anderson of Scott Paper states: ‘‘We used to have diﬀerent ﬁnancial measures for diﬀerent purposes — discounted cash ﬂow for capital decisions, another measure for rewarding performance and the like. 2 Now EVA is one measure that integrates all that. 2 it oﬀers an excellent link to the creation of shareholder value’’ (Walbert, 1994) pp. 111—112. Jim Meenan, CFO of AT&T’s communications services group expresses a similar view: ‘‘Every decision is now based on EVA. The motivation of our business units is no longer just to make a proﬁt. The drive is to earn the cost of capital. 2 when you drive your business units toward EVA, you’re really driving the correlation with market value’’ (Walbert, 1994) (p. 112). Eugene Vesell, managing director of Oppenheimer Capital states: ‘‘The ﬁrst thing we look at when we pick companies is, are they motivated by EVA? We prefer it to measures like EPS or return on equity.’’ (Tully, 1994) (p. 143). We emphasize that our results are only an input to the policy making process. Each of the measures we consider may have value in other decision contexts, e.g., cash from operations may provide valuable information to lenders and suppliers about liquidity. Questions regarding cost and best source(s) of data are beyond the scope of this research.
(In Section 5.. The ﬁrst (of two) empirical questions we address is Q1: Do EVA and/or RI dominate currently mandated performance measures. tests of question 1 indicate that earnings (R"12. in explaining contemporaneous annual stock returns? This relative information content question examines which variables (EVA. Biddle et al. earnings and operating cash ﬂow. The remainder of the paper is organized as follows. This ﬁnding is supported across a number of alternative speciﬁcations. while each component is signiﬁcantly associated with marketadjusted returns.8%) is signiﬁcantly more highly associated with marketadjusted annual returns than are RI (R"7. we decompose EVA into components (e.8%). RI. cash from operations. This study provides evidence that we hope will be useful to policy makers who may be interested in EVA or RI as replacements (or complements) to earnings and CFO as key measures of ﬁrm performance. Considering the relative and incremental information content results together. components unique to EVA (capital charge and accounting adjustments) are typically not signiﬁcant. capital charge. For the full sample. / Journal of Accounting and Economics 24 (1997) 301–336 valuerelevance. tests across alternative speciﬁcations indicate that. we examine whether EVA and/or RI complement currently mandated performance measures. and describes statistical tests for relative and incremental information content.g. earnings and CFO: Q2: Do components unique to EVA and/or RI help explain contemporaneous stock returns beyond that explained by CFO and earnings? This is equivalent to asking: Does the market appear to value a given EVA component beyond the information contained in the other components? To address this incremental information content question. CFO or earnings) have a greater association with contemporaneous stock returns and provides a direct test of one of Stern Stewart’s claims about the superiority of EVA.) Using a sample of 6.5%) and that all three of these measures dominate CFO (R"2. Section 2 provides a description of EVA and its components. and accounting ‘adjustments’) and evaluate the contribution of each component toward explaining contemporaneous stock returns. Section 3 . Further. while cash ﬂow and accrual components are consistently signiﬁcant. operating accruals.C. the EVA components do not appear to be economically signiﬁcant. Second.174 ﬁrmyears representing both adopters and nonadopters of EVA over the period 1984—1993.3%) or EVA (R"6.304 G. neither EVA nor RI appears to dominate earnings in its association with stock market returns. presents hypotheses.5 we examine separately another Stern Stewart claim that EVA outperforms earnings in explaining ﬁrm values.
Next. where k " Stern Stewart’s estimate of the ﬁrm’s weighted average cost of capital.C. and noncurrent portion of deferred taxes. amortization. NOPAT separates operating activities from ﬁnancing activities by adding back the aftertax eﬀect of debt ﬁnancing charges (interest expense) included in EBEI. depreciation. Section 5 reports various extensions and sensitivity analyses. earnings before extraordinary items (EBEI). hypotheses and statistical tests 2..1. 2. variable deﬁnitions.G. and descriptive statistics. current liabilities (other than notes payable and current portion of longterm debt). .g. where ATInt"the aftertax equivalent of book interest expense. Residual income diﬀers from EBEI in that it measures operating performance (NOPAT) net of a charge for the cost of all debt and equity capital employed: RI"NOPAT!(k*Capital). earnings. We begin by partitioning earnings into operating cash ﬂows and accruals: EBEI"CFO#Accrual. Biddle et al. Accrual " total accruals related to operating (as opposed to investing or ﬁnancing) activities. We close with a summary and a discussion of potential factors contributing to the failure of EVA and/or RI to dominate earnings. residual income (RI) and economic value added (EVA). we deﬁne net operating proﬁts after tax (NOPAT) as EBEI plus the aftertax cost of interest expense NOPAT"EBEI#ATInt. where CFO " net cash provided by operating activities. e. residual income and E»A This section describes linkages between operating cash ﬂows (CFO). Section 4 provides empirical results on the relative and incremental information content of EVA and its components. noncash current assets. Linkages between operating cash ﬂow. / Journal of Accounting and Economics 24 (1997) 301–336 305 reports sample selection criteria. Components of EVA.
is consistent with the ﬁrm creating wealth for the residual claimants. Stern Stewart do not disclose complete details about their accounting adjustments. 742—743). / Journal of Accounting and Economics 24 (1997) 301–336 Capital " Stern Stewart’s deﬁnition of assets (net of depreciation) invested in goingconcern operating activities. 1991) (pp.. In any given year.C. adding net capitalized intangibles (including R&D). Positive RI reﬂects proﬁts in excess of that required by debt and equity capital suppliers and. 28—30) argues that research and development costs should be capitalized (if material) and amortized. Negative RI is consistent with decreasing shareholder wealth. adding allowances for bad debts. NOPAT is adjusted by adding back the period’s R&D expense and deducting amortization of the R&D asset. adding unrecorded goodwill.306 G. the net eﬀect is an increase (decrease) in NOPAT if R&D expense is greater (less) than R&D amortization. or equivalently. Capital is higher by the amount of the net capitalized R&D asset. subtracting marketable securities and construction in progress (because neither contributes to current operating activities). AcctAdj " Stern Stewart adjustments to accounting measures of capital. 1991) (Chapter 2): EVA"NOPAT#AcctAdj !k*[Capital#AcctAdj ]. This requires adjustments to both NOPAT (via AcctAdj ) and to Capital (via AcctAdj ). inventory obsolescence. EVA is Stern Stewart’s proprietary version of RI. adding cumulative goodwill amortization. at the beginning of the period. . etc. AcctAdj reﬂect the cumulative eﬀect on Capital of the capitalization and amortization of current and past R&D expenditures. Stern Stewart attempts to improve on RI by adjusting NOPAT and Capital for what they view to be ‘distortions’ in the accounting model of performance measurement (Stewart. Other adjustments to NOPAT include: adding the change in bad debt allowances. Other adjustments to Capital include: capitalization and amortization of certain marketing costs. and subtracting an estimate of taxes owed for the period (Stewart. AcctAdj and AcctAdj are not examined individually in subsequent empirical tests because Stern Stewart does not disclose them separately. Biddle et al. warranties. e. 1991) (pp. net of taxes (Stewart. Stewart (1991) (pp.. At any point in time. 112—117). the shareholders. As an example of a common accounting adjustment. contributed and retained debt and equity capital. adding other operating income. adding goodwill amortization.g. adding the LIFO reserve. asset lives and amortization patterns. adding the present value of noncapitalized long term leases. adding the change in the LIFO reserve. where AcctAdj " Stern Stewart adjustments to accounting measures of operating proﬁts. thus. and adding (subtracting) cumulative unusual losses (gains).
. forwardlooking and can form estimates of performance measures.. EVA can be decomposed into its component parts: EVA"CFO#Accrual#ATInt!CapChg#AcctAdj. 1 summarizes these relations by showing how EVA components combine into other performance measures. Following Biddle et al. i.e. where CapChg"k*Capital AcctAdj"AcctAdj !(k*AcctAdj ). EBEI and RI. 1995.G. or valuerelevance.. i.2.e. we draw a distinction between relative and incremental information content. we use stock market returns to compare the information content.g. 2. / Journal of Accounting and Economics 24 (1997) 301–336 307 Fig. Hypotheses By assuming that equity markets are (semistrong) eﬃcient. 1987). Biddle et al. CFO.. incremental information content comparisons assess whether one measure provides valuerelevant data beyond that provided by another measure and apply when assessing the information content of a supplemental disclosure or the information of a component measure (e. EBEI. . Fig. RI and EVA.. Bowen et al. of CFO. 1. Components of economic value added (EVA). We use this decomposition to examine the incremental information content of EVA components. Relying on the above deﬁnitions.C. when only one measure can be chosen. Relative information content comparisons are appropriate when one desires a ranking of performance measures by information content or when making mutually exclusive choices among performance measures. In contrast.
CFO. Statistical tests A standard approach for assessing information content is to examine the statistical signiﬁcance of the slope coeﬃcient. RI and EVA. 1 by testing the null hypotheses that individual components of EVA do not provide incremental information content beyond other components that also comprise CFO and EBEI: H : Component X does not provide information content beyond that pro' vided by the remaining components X —X where X —X are components of EVA (i. RI or EVA). in the following ordinary leastsquares regression (that omits ﬁrm subscripts): D "b #b FE /MVE #e R 6R R\ R (1) where. ATInt. Because little is known about suitable proxies for market expectations for performance measures other than earnings. CapChg and AcctAdj).3. CFO. EBEI. . RI and EVA have equal relative information content: H : The information content of measure X is equal to that of X 0 where X and X represent pairwise combinations from the set of performance measures: CFO.308 G. 2. b . X (e.C.e. We examine the incremental value relevance of EVA components summarized in Fig. FE /MVE is the unexpected realization (or forecast 6R R\ error) for a given accounting measure. Biddle et al.g. EBEI. D is the dependent variable. we take a neutral position and conduct twotail tests of the null hypotheses that CFO. Rejection of H is viewed as evidence of incremental information ' content. a measure of (abnormal or unexpected) R returns for time period t. Rejection of H is viewed as evidence of 0 a signiﬁcant diﬀerence in relative information content. MVE . we use an approach from Biddle and The MVE deﬂator is measured 3 months after the prior year end to be consistent with the start of the returns period measured by the dependent variable. / Journal of Accounting and Economics 24 (1997) 301–336 Despite claims by Stern Stewart and others that EVA and RI are more valuerelevant to market participants than EBEI and CFO.. Accrual. EBEI.. and e is R\ R a random disturbance term. scaled by the beginningofperiod market value of the ﬁrm’s equity.
(3) relates abnormal returns and (scaled) lagged measures of accounting for performance. ARIMA.C. This is one way of addressing the potential concern that (say) EVA is less well predicted by past observations of EVA than (say) EBEI is predicted by past values of EBEI. Results based on these speciﬁcations are qualitatively similar to those reported and are available from the authors. / Journal of Accounting and Economics 24 (1997) 301–336 309 Seow (1991) and Biddle et al. is predicted by lagged values of each of the other variables — CFO. R R R\ R\ R\ R (4) This ‘onelag’ version is equivalent to the ‘levels and changes’ speciﬁcation proposed by Easton and Harris (1991). RI and EVA. say EVA. (3) encompasses a range of alternative speciﬁcations for market expectations. Thus each information variable. constant stock price multiple.G. but it is motivated diﬀerently. E(b )"b . . (2) into Eq. and combined ‘levels and changes’ speciﬁcations. (3) is ﬂexible in terms of allowing any number of lagged observations to be included as explanatory variables. Substituting Eq. Although Eq. and E(b)"!b G G\ G i'1. including randomwalk. In Eq.. Biddle et al. in the presence of possible structural change across time. (3). where E(b )"b !b . It is then assumed that market expectaR R R tions are formed according to a discrete linear stochastic process (in autoregressive form): E(X )" # X # X # X #2 R R\ R\ R\ (2) where the is a constant and ’s are autoregressive parameters. (1) yields: D "b #b (X !( # X # X # X #2))/MVE R R R\ R\ R\ R\ #e "b #b X /MVE #b X /MVE #b X /MVE R R R\ R\ R\ R\ R\ #b X /MVE #2#e . R\ R\ R (3) Eq. 1995 that estimates market expectations ‘jointly’ with slope coeﬃcients. Eq. the proxy for market expectations is estimated jointly with the slope coeﬃcient (b) using the same data and optimization criterion (minimum G mean squared errors). It also is in a more convenient form that allows the slope or ‘response’ coeﬃcient (b ) to be We also consider a speciﬁcation that allows each information variable to be predicted by lagged observations of all of the information variables. EBEI. we limit Eq. This is accomplished by ﬁrst expressing the forecast error as the diﬀerence between the realized value of a performance measure and the market’s expectation: FE "X !E(X ). (3) to one lag: D "b #b X /MVE #b X /MVE #e .
(4a)) and deriving the onelag speciﬁcation (in Eq. it is ﬁnite sample exact. Under usual regularity conditions (uncorrelated homoskedastic errors). (4) where b "a #a and b "!a . (4c) R R\ R\ R\ R Eq. a nonlinear null hypothesis is obtained that involves quadratic forms of regression coeﬃcients. The Biddle—Seow—Siegel test derives from Hotelling (1940). Lys and Sabino (1996) provide evidence that Vuong’s test outperforms the Jtest. (4). we employ a statistical test from Biddle et al. . The coeﬃcient(s) on the nonlag term(s) can be interpreted directly as ‘response’ coeﬃcient(s). it is well suited to evaluate the signiﬁcance of relative information content comparisons in accounting contexts.1. / Journal of Accounting and Economics 24 (1997) 301–336 observed directly (rather than being derived from separate coeﬃcients on levels and changes). (1995). The relation between the two speciﬁcations can be illustrated by starting with the levelschanges speciﬁcation (in Eq. in Eq. Davidson and MacKinnon’s nonnested ‘Jtest’ and Vuong’s likelihood ratio test are ‘pairwise tests for model selection’ designed to assess which of two competing models is closer (in terms of KullbackLiebler distance) to the ‘truth’. Biddle and Siegel (1996) provide evidence that the Biddle—Seow—Siegel test performs as well as or better than Vuong’s test in calibration and power. (4c) the response coeﬃcient is (a #a ). As discussed in Biddle et al. The Jtest also can yield ambiguous results. we replicated our relative information content tests using Vuong’s test with qualitatively similar results as discussed brieﬂy in Section 4 below. 2. As conﬁrming evidence.3.C. Both are valid only asymptotically and may have poor ﬁnite sample properties. (4c)): D "a #a X /MVE #a (X !X )/MVE #e (4a) R R R\ R R\ R\ R "a #a X /MVE #a X /MVE !a X /MVE #e (4b) R R\ R R\ R\ R\ R "a #(a #a )X /MVE !a X /MVE #e . It is tested using a Wald test (Kennedy. (1995) that allows a test of the null hypothesis of no diﬀerence in the ability of two competing sets of independent variables to explain variation in the dependent variable. b (b ) is predicted to be positive (negative). which is problematic in applications assessing relative information content. Since a and a are both expected to be positive. we make six pairwise comparisons of regressions among the accounting performance measures CFO. Using this test. as speciﬁed in Eq..g. e.310 G. The test is constructed as a comparison of Rs. RI and EVA. this method for assessing relative information content compares favorably with alternative tests provided in Davidson and MacKinnon (1981) and Vuong (1989). EBEI. Tests for relative information content To assess relative information content. Dechow. (4c) corresponds to Eq. 1985) of estimated coeﬃcients and their heteroskedasticityadjusted variance—covariance matrix. and can be used in conjunction with White’s (1980) correction for heteroskedastic errors. By using a lackofﬁt measure deﬁned as the average of the sum of squared residuals and the sum of squared prediction errors. Biddle et al. As a result. generalizes to any number of predictor variables.
for the onelag speciﬁcation in Eq.271 observations) due to either missing Compustat or CRSP (Center for Research in Securities Prices) data or to provide a lagged observation for each variable.g. Speciﬁcally.3.1. Sample selection. The initial sample of 1000 ﬁrms (8. For their corporate clients. White’s (1980) correction is employed in both the relative and incremental information content tests. 6 H : b "b "0. Sample selection Data used in this study were purchased directly from Stern Stewart & Co. 7 where b .57% of the 6. b .83% to 1. incremental information content is assessed by examining the statistical signiﬁcance of regression slope coeﬃcients. data greater (less) than 4 standard deviations from the median of the ﬁrmyear observations are assigned a value equal to the median plus (minus) 4 standard deviations. Tests for incremental information content Following standard methodology (e. and cost of capital for ﬁrms with ﬁscal years ending June 1983 to May 1994 (see variable deﬁnitions below). (1987)).G. The total number of observations reset for any variable range from 51 to 97. capital.524 ﬁrmyear observations) is reduced by 219 ﬁrms (2. (4) generalized to two accounting performance measures X and ½. incremental information content is assessed using ttests on individual coeﬃcients and Ftests of the joint null hypotheses: H : b "b "0. or 0.174 sample ﬁrm years. variable deﬁnitions and descriptive statistics 3. both the dependent and independent variables are winsorized to $4 standard deviations from the median.. Stern Stewart make ‘a handful’ of standard adjustments. The resulting sample has 6. Stern Stewart make additional custom adjustments (not available to the public). In other words.2. . (5) R\ R\ R To control for the potential eﬀects of heteroskedastic errors. For their publicly available database used in this study. Biddle et al. (5) below: D "b #b X /MVE #b X /MVE #b ½ /MVE R R R\ R\ R\ R R\ #b ½ /MVE #e . 3. These data include up to eleven annual observations for economic value added (EVA).C. Bowen et al. Next. b and b are from Eq. We also delete 79 extreme outlier observations deﬁned as more than 8 standard deviations from the median.174 ﬁrmyear observations for 773 ﬁrms. / Journal of Accounting and Economics 24 (1997) 301–336 311 2.
Bowen et al. Biddle et al. EBEI. 3. Rs in twoyear subperiods from 1988 onward are slightly higher than for the two. operating activities — net cash ﬂow.. / Journal of Accounting and Economics 24 (1997) 301–336 These data are compiled by Stern Stewart & Company from Business ¼eek’s listing of the 1. data item D110.312 G. . For years after 1987 Compustat data item D308. is used. MktAdjRet Market adjusted return computed from CRSP data as a ﬁrm’s 12month compounded stock return less the 12month compounded valueweighted marketwide return.3. market adjusted returns. Consistent with the possibility that pre1988 measures of CFO are noisy.000 largest ﬁrms in market capitalization.2. cash ﬂow from operations is estimated similar to Bowen et al. Dependent variable Our dependent variable. depending upon the year of the observation. 1995.000 ﬁrms. are deﬁned below: CFO Cash ﬂow from operations obtained from the statement of cash ﬂows or the statement of changes in ﬁnancial position. funds from operations — total. Biddle et al. is commonly used in information content studies to measure unexpected returns (e. The listing is published annually.C. Stern Stewart introduced its ﬁrst 1000 ranking for the calendar year ended 1988.. Independent variables and descriptive data — relative information content tests The four measures of accounting performance in the relative information content tests.g. 1987) as funds from operations (D110) plus the change in current liabilities (D5) less the change in debt in current liabilities (D34) less the change in current assets (D4) plus the change in cash and cash equivalents (D1). 3. twoyear subperiods before 1988. is used if the ﬁrm used the cash deﬁnition of funds for the statement of changes in ﬁnancial position. RI and EVA. Stern Stewart modiﬁes this list by ﬁrst removing utilities and ﬁnancial institutions.. If the ﬁrm used the working capital deﬁnition of funds in any year prior to 1988. 1989). A 12month nonoverlapping period ending three months following the ﬁrm’s ﬁscal yearend is chosen to allow time for information contained in the ﬁrm’s annual report to be impounded in stock market prices. CFO. and then adding ﬁrms from prior Business ¼eek 1000 listings to bring the sample back to 1. (1986. For years prior to 1988.
Despite a survivorship bias in the data. EBEI has the highest correlation with market adjusted returns. Economic value added obtained from the Stern Stewart 1000 database. Low values of EVA and RI are also consistent with a potential upward bias in Stern Stewart’s cost of capital estimates.4 below for deﬁnitions of ATInt and CapChg. operating accruals. The ﬁrm’s tax rate is assumed to be zero if net operating losses are present. Residual income equals earnings plus aftertax interest expense less a charge on all capital (RI"EBEI#ATInt!CapChg). aftertax interest expense. capital charge and accounting adjustments: Accrual Operating accruals deﬁned as earnings less cash ﬂow from operations (Accruals"EBEI!CFO). 3. winsorized variables R\ pooled across time are provided in Panel A of Table 1. Descriptive data on these deﬂated. In order to reduce heteroscedasticity in the data. Near zero EVA and RI is consistent with a competitive economy where even the typical large ﬁrm has diﬃculty earning more than its cost of capital. Biddle et al. Independent variables and descriptive data — incremental information content tests The independent variables in the incremental information content tests are the ﬁve components of EVA described in Section 2. median RI is negative in every year and median EVA is negative in 7 out of 10 years. EBEI has the lowest standard deviation among the four performance measures consistent with the smoothing eﬀects of accruals. ATInt . we deﬂate all independent variables by the market value of equity three months after the beginning of the ﬁscal year (MVE ).4. / Journal of Accounting and Economics 24 (1997) 301–336 313 EBEI RI EVA Earnings deﬁned as Compustat data item D18. 1: CFO (deﬁned above). net income before extraordinary items. Correlations between the independent variables are all positive and signiﬁcant except that EVA and RI are negatively correlated with CFO. ATInt is nonnegative. Otherwise the maximum statutory corporate tax rate is used for the given year. 2.C.1 and summarized in Fig. Accruals can be positive or negative but are more likely to be negative (reﬂecting noncash expenses such as depreciation and amortization). After tax interest expense computed as 1 minus the ﬁrm’s tax rate multiplied by interest expense (D15). CFO has the largest ﬁrmyear mean and median followed by EBEI. EVA and RI.G. Undeﬂated median values of each performance measure are plotted across time in Fig. See Section 3. Correlations among these measures are provided in Panel A of Table 1.
Pearson correlation coeﬃcients '0.0407 are signiﬁcant at (0.00 The sample has 6.900 !0.125 0.00 0.00 Panel B: Descriptive statistics on the dependent and independent variable in incremental information content tests Dependent variable MktAdjRet Descriptive statistics Mean Median Std.00 !0.01 '0.133 !0.00 !0.782 0.011 0.134 !0. Biddle et al.00 0.10 '0.026 !0. Dev.751 0.034 0. All independent variables are deﬂated by the market value of equity three months after the beginning of the ﬁscal year.155 0.652 0.362 0.049 !0.007 0. Correlations MktAdjRet EBEI R EVA R RI R CFO R 1. Correlations MktAdjRet CFO R Accrual R ATInt R CapChg R AccAdj R 1.082 !0.138 0. Dev.362 0.00 0.111 0.247 0.592 0.065 0.021 !0.580 0.137 0.307 0.004 0.007 !0.00 0.900 !0.133 EBEI R Independent variables EVA R RI R CFO R 1.039 1.00 0.016 0.210 1.011 0.00 0.017 0.056 !0. / Journal of Accounting and Economics 24 (1997) 301–336 Table 1 Descriptive statistics for pooled data Panel A: Descriptive statistics on the dependent and independent variable in relative information content tests Dependent variable MktAdjRet Descriptive statistics Mean Median Std.314 G.018 !0.055 CFO R Independent variables Accrual R ATInt R CapChg R AccAdj R 1.0204 are signiﬁcant at (0.122 1.129 !0.153 0.118 0.059 0.086 !0.142 0.118 0.011 1.001 .149 0.C.138 1.059 0.363 0.055 0.142 0.501 !0.127 0.057 !0.0319 are signiﬁcant at (0. All variables are winsorized $4 standard deviations from the median.174 ﬁrmyear observations.007 0.057 1.469 0.054 0.
See Stewart (1991). CapChg AcctAdj Capital charge deﬁned as the ﬁrm’s weighted average cost of debt and equity capital times its beginning of year capital. 2. According to Stewart (1994). Stern Stewart estimate the cost of capital by weighting the cost of equity (applying the capital asset pricing model) and the aftertax cost of debt. .C.G. 741—745. Median values of performance measures. Our deﬁnition of RI incorporates Stern Stewart adjustments to capital. Accounting adjustments reﬂect Stern Stewart’s net annual adjustments to earnings and capital. 1984—93. and are deﬁned as economic value added less residual income (AcctAdj"EVA!RI). AcctAdj can be positive or negative. Biddle et al. Both of these items are obtained from Stern Stewart. Capital is a proxy for all cash invested in the business since a company’s inception. Data were not available from Stern Stewart to calculate capital before accounting adjustments. especially pp. CapChg is positive since both the cost of capital and capital are positive. / Journal of Accounting and Economics 24 (1997) 301–336 315 Fig.
the Vuong test provides greater statistical signiﬁcance. Nearly identical inferences are obtained using the Vuong (1989) test. These results suggest that.00001) for each performance measure based on the full sample of 6. . in terms of relative information content. pvalues from twotailed statistical tests of relative information content are shown centered in parentheses for each of the six possible pairwise comparisons. EBEI. The correlation between CFO and AcctAdj is insigniﬁcant. with EBEI having a signiﬁcantly larger adjusted R (9%) than each of the other three performance measures. 1996).174 ﬁrmyears. consistent with ﬁrms with higher operating cash ﬂows also having higher debt and equity costs. identical inferences are obtained at conventional signiﬁcance levels for all pairwise comparisons. Empirical results 4.C. / Journal of Accounting and Economics 24 (1997) 301–336 Descriptive data on these deﬂated. Biddle et al. earnings signiﬁcantly outperforms RI. 4. we predict a positive (negative) slope coeﬃcient on contemporaneous (lagged) observations of each performance measure. CFO. Adjusted Rs from these regressions are provided in Table 2.4%). RI signiﬁcantly outperforms EVA (although the gap here is smaller). Both mean and median Accrual and AcctAdj are negative. For example. and all three outperform CFO. consistent with its asymptotic nature and tendency to reject the null observed in simulation tests (Biddle and Siegel. The negative correlation between CFO and Accrual is again consistent with the accrual process smoothing earnings relative to the underlying operating cash ﬂows. Results in Panel A of Table 2 are based on Eq. (4) and each of the six pairwise diﬀerences in R are signiﬁcant at conventional levels. In results available from the authors. winsorized EVA components are provided in panel B of Table 1. RI and EVA. and both have a signiﬁcantly larger adjusted Rs than CFO (2. CFO has by far the largest correlation with marketadjusted returns. In general. On average.2%) than does the EVA regression (5. The negative coeﬃcient on the lagged term follows from the prediction that changes in the performance measures also are positively associated with stock returns (see Section 2. coeﬃcient b (b ) is positive (negative) and signiﬁcant (at (0. Relative information content tests Relative information content is assessed by comparing adjusted Rs from four separate regressions. one for each performance measure.316 G. Correlations between CFO.1%). consistent with some smoothing of the underlying operating cash ﬂows.1. The RI regression has a signiﬁcantly larger adjusted R (6. especially footnote 9). ATInt and CapChg are positive and signiﬁcant.3. for the relative comparisons in Panel A of Table 2 and Table 4. The highest R is shown on the left (which in both panels is from the EBEI regression) and the lowest is shown on the far right (which in both panels is from the CFO regression).
(4) modiﬁed to allow diﬀerent coeﬃcients on positive versus negative values of the independent variables: D "b #b X /MVE #b X /MVE #b X /MVE R R R\ R R\ R\ R\ #b X /MVE #e (6) R\ R\ R where D "marketadjusted returns. EBEI.0238 The underlying regressions in panel A constrain the coeﬃcients to be equal across all ﬁrmyear observations. First row presents pvalue for comparison between ﬁrst and second ranked measures. X"a given performance measure (CFO. residual income. 1995). Because the valuerelevance of the . Twotailed pvalues in parentheses represent tests of the null hypothesis of no diﬀerence between pairwise comparisons of adjusted Rsquares (Biddle et al. See description of pvalue below. RI 0. Performance metrics are listed in order of Rsquares from highest (on the left) to lowest (on the right). R pvalue 6.000) EVA 0. X"a given performance measure (CFO.0507 ' (0. RI and EVA).000) (0. second and third ranked measures and third and fourth ranked measures.000) (0.0904 ' (0.0624 ' (0. Hayn (1995). Biddle et al.0732 ' (0.174 EBEI 0. RI or EVA). where R R R\ R\ R\ R D "marketadjusted returns.C.000) CFO 0. / Journal of Accounting and Economics 24 (1997) 301–336 317 Table 2 Tests of the relative information content of EVA . R and MVE"market value of equity three months after the beginning of the ﬁscal year.000) CFO 0.000) Underlying regressions are from Eq. Burgstahler and Dichev (1997) and Collins et al.000) Panel B: Coeﬃcient of positive and negative values of each performance measure allowed to diﬀer All ﬁrms Adj. Underlying regressions are from Eq.0649 ' (0. comparisons are between ﬁrst and third ranked. On the the next row. EBEI.041) (0.266) (0.000) EVA 0. (4): D "b #b X /MVE #b X /MVE #e .. The last row compares ﬁrst and fourth ranked measures.000) (0. Statistical tests of diﬀerences in explanatory power across performance measures are presented centered in parantheses below the adjusted Rsquares. earnings and operating cash ﬂow (H ) 0 Rank order of R Observations (1) Relative information content (2) (3) (4) Panel A: Coeﬃcient of positive and negative values of each performance measure constrained to be equal All ﬁrms Adj.G.000) (0.0280 RI 0. and R MVE"the market value of equity three months after the beginning of the ﬁscal year.1278 ' (0. and second and fourth ranked measures. (1997) provide evidence that loss ﬁrms have smaller earnings response coeﬃcients than do proﬁtable ﬁrms.174 EBEI 0. R pvalue 6.
adjusted Rs increase for each performance measure when allowing for separate coeﬃcients on positive and negative values. we discuss possible reasons why we fail to detect stronger valuerelevance for EVA and RI. the ranking of performance measures remains identical and statistical comparisons between regressions are nearly unchanged — earnings dominates each of the other three performance measures and all three (EBEI.318 G. .C. We expect a negative association between returns and the two components representing nonnegative capital costs.2. (4). this evidence points to earnings having higher relative information content than EVA. ATInt and CapChg. We expect a positive association between marketadjusted returns and the three components CFO. we examine the sensitivity of these results to alternative speciﬁcations. coeﬃcients (available from the authors) are generally larger (in absolute value) and more signiﬁcant for positive values of X than for the negative values. 1996). Accrual and AcctAdj. In Section 5. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336 other performance measures (CFO. we cannot support the Stern Stewart claim that EVA has greater information content than earnings. the lagged terms are predicted to have the opposite sign (footnote 9). the relative information content results show no evidence of EVA (RI or CFO) dominating EBEI. In Section 6. Incremental information content tests Table 3 presents results on the incremental information content of EVA components from regression (7): MktAdjRet "b #b CFO #b CFO #b Accrual R R R\ R #b Accrual #b ATInt #b ATInt R\ R R\ #b CapChg #b CapChg #b AcctAdj R R\ R #b AcctAdj #e . Similar to the relative information content regressions in Eq. (7) R\ R Predicted signs on each coeﬃcient are provided below the variable labels.8%.174 ﬁrmyear observations. The only diﬀerence is that RI and EVA are no longer statistically diﬀerent from each other. However. (6) R\ R\ R\ R\ R Panel B of Table 2 presents results for regression (6) for the complete sample of 6. RI and EVA) could also vary with their sign (O’Byrne. Compared to R results reported above in Panel A. This increase is largest for the EBEI regression with adjusted R increasing from 9% to 12. RI and EVA) dominate CFO. Taken as a whole. Thus. we repeat our tests for relative information content after partitioning each performance measure into positive and negative values: MktAdjRet "b #b X /MVE #b X /MVE R R R\ R R\ #b X /MVE #b X /MVE #e . In contrast. 4. Consistent with prior research.
000) 0.63 3.055 3. ' .270 !2. R Obs. independent variables are components of EVA (CFO.72 3.192 !0.824 16.000) 1.221 2.013 1.09 !7.174 0.42 (0.02 !8. pvalues in parentheses represent nondirectional Ftests of the null hypothesis of no incremental information content (hypothesis H ).43 1. accounting adjsutments) and are shown in nonlagged forms as column headings. ! Constant CFO R CFO R\ Accrual Accrual R R\ Predicted signs: !0.594 0.83 (0. operating accruals.45 (0.C.55 (0.001) 0. aftertax interest expense.0907 G. Biddle et al.751 13.73 87.Table 3 Tests of incremental information content of EVA components: CFO.42 1.391 0.032) !0. capital charge. accounting adjustments (H ) ' ATInt R # ! # # R\ ! ATInt CapChg R CapChg \ AccAdj AccAdj \ ! Adj.027) # ! # All ﬁrms tstat Fstat pvalue 6. operating accruals.12 0.48 6. / Journal of Accounting and Economics 24 (1997) 301–336 319 Dependent variable"marketadjusted returns.357 0.772 !2. Each independent variable is deﬂated by market value of equity three months after the beginning of the ﬁscal year.61 (0. capital charge.473 !0. aftertax interest.53 128.
9 out of 10 coeﬃcients are in the predicted direction and signiﬁcant in onetail ttests at the 0. / Journal of Accounting and Economics 24 (1997) 301–336 In Panel A for the full sample. All of the twotail Ftests are signiﬁcant at the 0. residual income. The size of each circle represents relative information content and the nonoverlapping areas represent incremental information content.04% in Panel A of Table 2) and returns on EVA components (9. .05 level or better. EBEI.320 G. earnings and operating cash ﬂow. RI and EVA protrude slightly from behind EBEI reﬂecting some limited incremental information content beyond earnings.C. the overall minuscule increase in adjusted R between the regression of returns on EBEI (9.07% in Table 3) suggests that the economic signiﬁcance of the incremental information content of the EVA components is slight. 3 uses a Venn diagram to summarize our ﬁndings on relative and incremental information content for the four information variables CFO. The exception is the lagged term for AcctAdj. these results suggest that. Fig. their contributions to the information content of EVA are not suﬃcient for EVA to provide greater relative information content than earnings. Fig. while ATInt. Biddle et al. The relative sizes of the Fstatistics suggest that CFO and Accrual make by far the largest incremental contributions to explaining marketadjusted returns. while EVA components oﬀer some incremental information content beyond earnings components. Relative and incremental information content of EVA . RI and EVA. However. When combined with the relative information content ﬁndings above. 3. CFO.05 level or better. CapChg and AcctAdj exhibit much smaller incremental contributions. which is in the wrong direction. EBEI exhibits the largest relative information content among the measures.
In 1992—93. However. Finally.481 in 1992—93. EVA and RI are not statistically signiﬁcant at conventional levels. nonoverlapping. we examine the sensitivity of the basic results reported above to alternative speciﬁcations.006). In pairwise comparisons of relative information content. We conclude with an overall assessment of the results of the sensitivity tests. twoyear test periods (instead of one tenyear period). Thus we again cannot support the Stern Stewart claim that EVA has greater information content than earnings.049) and CFO (p"0. We also consider the 606 observations following the September 1993 Fortune article that touted EVA as “The Real Key to Creating Wealth” (Tully. In 1988—89 and 1990—91. where the dependent variable is the level of market value of the ﬁrm (rather than returns). ﬁrmyear observations increase from 1. we discuss a replication and extension of O’Byrne (1996). and the evidence is suggestive of EBEI dominating EVA (p"0. In incremental information content tests. EBEI does not outperform RI (p"0. Among the remaining EVA components (ATInt. EBEI outperforms each of the other performance measures at the 0. EBEI does not outperform RI (p"0.015 in the 1984—85 period to 1.072) but does outperform EVA (p"0. we report relative and incremental information content tests on annual data grouped into ﬁve. and 4) changing the return interval from oneyear (contemporaneous) returns to twoyear (combined contemporaneous and oneyear ahead) returns. Thus market participants apparently did not begin using EVA for equity valuation immediately following the appearance of the Fortune article. 3) changing the return interval from oneyear to ﬁveyears. Biddle et al. twoyear periods.045). adjusted Rs are largest for EBEI in every twoyear period. Sensitivity analyses and extensions In this section. in 1984—85 diﬀerences between EBEI.2%). In contrast.1. in 1986—87. 5. Because of survivorship bias in the Stern Stewart data. nonoverlapping. CapChg and AcctAdj). We repeat selected information content tests by: 1) partitioning annual observations into ﬁve.C.G. only 1 of 15 Fstatistics is signiﬁcant at the 5% level — AcctAdj in the 1984—85 sample period. the evidence points to earnings having higher relative information content in many subperiods. / Journal of Accounting and Economics 24 (1997) 301–336 321 5. Results for the period after release of the 1993 Fortune article again show strong support for the incremental information .083) but does outperform EVA (p"0. 2) evaluating subsets of ﬁrms that claim to use EVA for internal business decisions. The earnings regression again has the highest R (11. 1993). Partitioning the sample into subperiods Results reported in Tables 2 and 3 pool observations over the ten years 1984—1993. Taken together. there is no evidence of EVA (RI or CFO) dominating EBEI. CFO and Accrual are signiﬁcant in every twoyear period.061) but not RI. In this section. Using a 5% cutoﬀ.01 level or better.
but little evidence for the incremental signiﬁcance of the remaining EVA components. investors may become more attuned to the measure for ﬁrms that adopt EVA. Biddle et al. Table 5 reports tests of incremental information content for users of EVAlike performance measures. respectively)). which. The ‘Comp Year’ sample restricts observations in the ‘Comp’ sample to only those years in which an EVAbased compensation plan is in eﬀect. 13 of 16 are signiﬁcant for the CFO and Accrual components while only 1 of 24 are individually signiﬁcant for the remaining EVA components. While earnings is not as dominant in these smaller subsamples of EVA users. Thus. neither do the ﬁndings show EVA dominating earnings in its association with stock returns. To examine this possibility.65). Firms in the ‘Any’ sample have disclosed that they use EVA (or some similar concept) sometime during the period studied — even if that use appears to be minimal.’ ‘Performance’ and ‘Comp’ groups. Further. EBEI'RI for the ‘Any’ and ‘Performance’ subsamples (p"0. both CapChg and AcctAdj have signiﬁcant Fstatistics suggesting they make an incremental contribution to explaining contemporaneous security returns in years where ﬁrms have EVAbased . Adopters of ‘E»Alike’ performance measures It is possible that ﬁrms adopt EVA at least in part because their past experience indicates a relatively strong relation between EVA and stock returns. none of the performance measures diﬀer signiﬁcantly in relative information content at the 5% level.C. Firms in the ‘Performance’ sample provide more detail about their use of an EVAlike measure for performance measurement and/or decision making. With the exception of the ‘Comp Year’ group. Firms in the ‘Comp’ sample state that they use an EVAlike measure in senior management incentive compensation plans and thus. the ‘Comp Year’ sample is a subset of the ‘Comp’ sample. We include all available data including years before the plan was implemented. In the small ‘Comp Year’ sample (n"35). Table 4 reports the results of relative information content tests for ﬁrms using an EVAlike performance measure. The lower signiﬁcance levels may be attributable in part to the smaller sample sizes used in these tests. presumably.322 G. it is conceivable that the association between EVA and returns is stronger for EVA adopters. also use it for performance measurement and/or decision making.094 and 0. In onetail ttests of individual slope coeﬃcients using a 5% cutoﬀ (t"1. / Journal of Accounting and Economics 24 (1997) 301–336 content of CFO and Accrual. However. 5. but EVA has the largest R for the ‘Comp Year’ sample. which is a subset of the ‘Performance’ sample.073). Thus. none of the twotail Ftests are signiﬁcant for components unique to EVA.2. EBEI exhibits the largest Rs for the ‘Any. and only 3 out of 18 comparisons at the 10% level (EBEI'CFO for the ‘Any’ subsample (p"0. in turn. is a subset of the ‘Any’ sample. we consider separately four subsamples of ﬁrms that make some ‘use’ of EVAlike measures.059.
0292 ' (0. (4): D "b #b X /MVE #b X /MVE #e where R R R\ R\ R\ R D "marketadjusted returns.481) (0. (4). First row presents pvalues for comparison between ﬁrst and second ranked measures. all available ﬁrmyears are included.412) ‘Comp Year’ Adj.699) (0.855) EVA 0.0799 ' (0. .938) (2) RI 0.278) (0.0523 ' (0.962) (0.059) ‘Comp’ Adj. 1995). 626 (1) EBEI 0.0262 ' (0.0181 CFO 0.0239 RI 0.330) EBEI 0. E ‘Comp Year’ represents a subset of observations from ‘Comp’ only including years where ﬁrms have an EVAbased compensation plan in place.667) (0. R pvalue Obs. The last row compares ﬁrst and fourth ranked measures.867) (0.0484 ' (0. / Journal of Accounting and Economics 24 (1997) 301–336 Table 4 Tests of relative information content (H ): Sample partitioned by relative ‘use’ of EVA 0 Relative information content Rank order of R ‘Any’ Adj.3072 ' (0. ‘Comp Year’ is a subset of ‘Comp’. from highest (on the left) to lowest (on the right).481) (0. ‘Comp’ is a subset of ‘Performance’.735) (0. Twotailed pvalues in parentheses represent tests of the null hyposthesis of no diﬀerence between pairwise comparisons of adjsuted Rsquares (Biddle et al.486) (3) EVA 0.073) ‘Performance’ Adj.0317 RI 0. presumably therefore. and ‘Performance’ is a subset of ‘Any’. Performance metrics are lsited in order of Rsquares from underlying regression Eq.0461 ' (0. for performance measurement.765) (0.352) (0. R pvalue 35 EVA 0. See description of pvalues below. and second and fourth ranked measures. RI. comparison are between ﬁrst and third ranked. Biddle et al.491) CFO 0. Given the existence of a plan in any year. and R MVE"the market value of equity three months after the beginning of the ﬁscal year.C. R pvalue 344 EBEI 0.737) EVA 0.G.211) RI 0.550) (4) 323 CFO 0.0220 ' (0. X"a given performance measure (EVA..2366 ' (0.0386 ' (0.1152 Firms are categorized in their use of EVA as follows: E ‘Any’ represents all ﬁrms that have mentioned that they use EVA for performance evaluation and/or explicit incentive compensation — even if that use appears minimal.094) (0. On the next row. R pvalue 445 EBEI 0. Statistical tests of diﬀerences in explantory power across performance measures are presented centered in parentheses below the adjusted Rsquares. Underlying regressions are from Eq.0306 ' (0. E ‘Performance’ represents those ﬁrms that have mentioned that they use EVA for performance measurement but do not disclose any use of EVA in their explicit incentive compensation plans.834) (0. CFO). EBEI. E ‘Comp’ respresents those ﬁrms that use EVA in their explicit incentive compensation plans and.393) CFO 0. second and third ranked measures and third and fourth ranked measures.2644 ' (0.
012 !0.126 !0.779) 0.52 4.562 !1.624 5.25 (0. pvalue !1.215 !0.434 !1.105 1.488 1.396 0.58 !1.60 !0.45 (0.539) 0.40 (0.368 0.344 !0.036 0.05 2.C. Fstat.055 !1.265) 0.425 1.123 !0.010) .62 !0.30 !2.61 4.324 0.023 0.46 (0. Biddle et al.958 !2.235) !1.216) 0.53 !3.1054 Adj.129 !0.23 4.376 !1.000) !0.17 (0.41 10.000) 3.08 0.494 !1.76 0.087 !0.45 1.047 !0.557) !1.919 !0.365 3.61 (0.843 !0.634) !0.69 1. / Journal of Accounting and Economics 24 (1997) 301–336 ‘Performance’ tstat. Fstat.70 0.64 (0.791 !1.55 8.01 (0.868 !0.59 (0.54 (0.0625 0.00 0.34 (0.70 (0.138 !0. pvalue !0.61 14.71 0.62 (0.043 0.0780 G.37 0. pvalue 344 !0.09 (0. Fstat.35 !0.09 !0.41 0.60 !0. R Obs.136) 445 !0.54 1.544) ATInt CapChg CapChg AcctAdj AcctAdj R R\ R R\ # 0.896 4. ! Constant CFO R CFO R\ Accrual Accrual R R\ # ! # Predicted signs: ‘Any’ 626 tstat.044 1.324 Table 5 Tests of incremental information content of EVA components (H ): Sample partitioned by relative ‘use’ of EVA ' ATInt R # ! # R\ ! !1.04 !3.07 7.000) 0.35 1.95 ! !0.05 !1.009) 0.835 0.72 (0.51 1.08 !1.001) ‘Comp’ tstat.709 2.
/ Journal of Accounting and Economics 24 (1997) 301–336 325 Firms are categorized in their use of EVA as follows: ‘Any’ represents all ﬁrms that have made any mention of using EVA for performance evaluation and/or explicit incentive compensation—even if that use appears minimal.841 !2. pvalue !1.35 !2.430) 2.05 !1.031) !1.61 2. capital charge. Fstat.41) 0.122 !4.23 0.37 1.62 1.015 0.71 4. ‘Comp Year’ includes a subset of observations from ‘Comp’ for only those years where the EVAbased compensation plan is in eﬀect. Dependent variable"marketadjusted returns.70 3.51 3.09 !1.00 0.308 !1. and ‘Performance’ is a subset of ‘Any’.759 3.240 !0. Biddle et al.154) !1.167 !0.3344 ‘Comp year’ tstat.66 (0.02 (0. ‘Comp’ is as a subset of ‘Performance’.687 !1.C. ‘Comp’ includes alll ﬁrmyear observations for those ﬁrms that use EVA in their explicit incentive compensation plans in any year.87 (0.60 G.26 (0. operating accruals.178 !2. ‘Performance’ represents those ﬁrms that have mentioned that they use EVA for performance measurement but do not disclose any use of EVA in their explicit incentive compensation plans.769 0. accounting adjustments) and are shown in nonlagged and lagged forms as column headings.303) !0. ‘Comp Year’ is a subset of ‘Comp’.04 (0. . independent variables are components of EVA (CFO.130 0. Pvalues in parentheses represent nondirectional Ftests of the null hypothesis of no incremental information content (hypothesis H ). aftertax interest expense.
and the unexpected signs on coeﬃcients on CapChg. In addition. once again EBEI outperformed EVA.9%). X . only one test period is reported in Table 6. . because ﬁveyear data are less sensitive to the choice of expectations models. 5. CFO and Accrual are again highly signiﬁcant but the results on components unique to EVA (CapChg and AcctAdj) are insigniﬁcant.3. (8) R R\ R\ R\ R Since all ten years of data are used to examine the association between ﬁveyear returns and each performance measure. In incremental information content tests. the surprising insigniﬁcant Fstatistics on CFO and Accrual. is summed over the most recent ﬁveyear period 1989—93 (for the nonlagged term) and summed over the prior ﬁve years. Results again show the earnings regression with the highest R (31. only AcctAdj for the 1988—93 subperiod was signiﬁcant at the 5% level while none of the other components unique to EVA were signiﬁcant in the predicted direction. ‘5year sums’: MktAdjRet R "b #b X /M»E #b X /M»E #e .5%) and RI (10. these tests help address the possibility that the weaker performance of EVA is due to a poorer expectations model.C.3 and footnote 9.9%).2%) followed by CFO (18. Again. Independent variables reﬂect ‘ﬁveyear sums’ in that each performance measure. it does not appear users of EVA are adopting the concept because of its stronger association with stock returns. We also evaluated the relative and incremental information content using changes in (rather than sums of) each performance measure over the ﬁveyear period. 1994). In relative information content tests. 1984—88 (for the lagged term). other than weakly suggestive results for the ‘Comp Year’ sample. 1991.326 G. Stern Stewart reports its strongest results for EVA on ﬁveyear data (Stewart. Regression (8) below is used to evaluate relative information content comparisons. In Table 7 we report incremental information content of EVA components after extending the return interval from one year to ﬁve years. caution is warranted in drawing any inferences from this result due to: the small size of the ‘Comp Year’ sample. It includes nonlagged and lagged terms similar to the annual return regression (4) and are analogous to the ‘level and changes’ speciﬁcation discussed in Section 2. EVA (14. Fiveyear returns as the dependent variable In this section we extend the return interval from one year to ﬁve years. Biddle et al. / Journal of Accounting and Economics 24 (1997) 301–336 compensation plans in eﬀect. However. The diﬀerences in explanatory power between EBEI and each of the other three performance measures are highly signiﬁcant.
C. EBEI has signiﬁcantly higher association with twoyear returns (adjusted R"4.4%) than any of the other three information variables (whose Rs range from 2% to 2.1446 ' (0.5. RI or EVA). and MVE"market value of equity three months after the beginning of the ﬁscal year. R R R\ R\ R\ R where is deﬁned over the ﬁveyear intervals.3118 ' (0. / Journal of Accounting and Economics 24 (1997) 301–336 Table 6 Tests of relative information content (H ): Returns measured over 5year periods 0 Rank order of R ‘5year sums’ Adj. Biddle et al. second and third ranked measures and third and fourth ranked measures.G.1090 ‘5year sums’: D "b #b X /MVE #b X /MVE #e . 509 (1) EBEI 0.4. First row present pvalues for comparison between ﬁrst and second ranked measures. 1989—93 (nonlag) and 1948—88 (lagged) terms. (8): Relative information content (2) CFO 0.000) (0. we extend the return interval from the oneyear contemporaneous period used above to a twoyear period that includes both the contemporaneous and subsequent year. pvalue in parentheses represent twotail tests of the null hypothesis of no diﬀerence between pairwise comparisons of adjusted Rsquares (Biddle et al. To test this claim. Market value of the ﬁrm as the dependent variable Another claim made by Stern Stewart is EVA’s higher association with the market value of the ﬁrm.1888 ' (0.005) (0. Consistent with results in Table 2. X"a given performance measure (CFO.051) (3) EVA 0. R pvalue Obs. R EBEI.264) (0.000) Underlying regression is from Eq.030) (4) 327 RI 0. 5. 5. and second and fourth ranked measures. 1995). .3%). and inconsistent with the conjecture that the market subsequently learns about the importance of EVA. On the next row. D "marketadjusted returns measured over ﬁve years. Twoyear (contemporaneous and oneyear ahead) returns as the dependent variable To consider the possibility that equity market participants take longer to learn about and impound EVA.. The last row compares ﬁrst and fourth ranked measures. respectively. comparisons are between ﬁrst and third ranked. we replicate and extend a study We also examined the change in ‘market value added’ (deﬁned by Stern Stewart as ﬁrm market value less invested capital) as a dependent variable with qualitatively similar ﬁndings.
56 33. 1989—93 (lagged) terms.000) Dependent variable"marketadjusted returns.01 (0.072 !0.3241 ‘5year sums’ tstat.447) !0.373 2.e. R R R\ R\ R\ R R\ R\ R\ R R\ R\ R\ R where is deﬁned over the ﬁveyear intervals.93 !2.128 !0.659 0.39 # ! # 509 !0. / Journal of Accounting and Economics 24 (1997) 301–336 D "b #b X /MVE #b X /MVE #b X /MVE #b X /MVE #2#b X /MVE #b X /MVE #e .328 Table 7 Tests of incremental information content of EVA components (H ): Returns measured over ﬁveyear periods ATInt R # ! # # ! R\ ! ATInt CapChg CapChg AcctAdj AcctAdj R R\ R R\ Adj. and MVE"market value of equity three months following the beginning of the ﬁscal year.01 (0.75 (0.275 0. R Obs.21 0.11 0. R R i. Accrual. Biddle et al. X "a given EVA component. CFO.731 1.64 2.. ! Constant CFO R CFO R\ Accrual Accrual R R\ Predicted signs !0. ATInt .549 1.509 0.99 !0.986) !0.25 18.54 7. D "marketadjusted returns.000) 5. Fstat.17 0. respectively.089 !0. CapChg and AcctAdj.C.42 1.088 0.065) 0.16 (0. Underlying regression as follows ‘5year sums’: G.487 1. . pvalue !2. independent varibles (EVA components) are shown a column headings. R pvalues in parentheses represent nondirectional Ftests of the null hypothesis of no incremental information content (hypothesis H ).81 (0.
Second. EVA economic value added for year t. O’Byrne makes a series of ‘adjustments’ only to the EVA regressions and uses Rs from these adjusted regressions to infer superiority of EVA over competing information variables. . and (3) including 57 industry dummy variables in order to capture potential industry eﬀects. We cannot replicate results for free cash ﬂow because O’Byrne (1996) does not provide a precise deﬁnition.C. and in our view most importantly. 120). Third. e unexplained residual error. he makes a series of adjustments to the EVA regressions by: (1) allowing separate coeﬃcients for positive and negative values of EVA. None of these adjustments are made for the NOPAT regression. R O’Byrne compares Rs from the initial two models and reports 31% for the EVA model (9) and 33% for the NOPAT model (10). or equivalently.. i. at the beginning of period t. he reports a much # (EVA /k)/Capital #e R R\ R # (NOPAT )/Capital #e R R\ R (9) (10) O’Byrne scales EVA by k and capital and NOPAT only by Capital. NOPAT R k Stern Stewart’s estimate of the ﬁrm’s weighted average cost of capital. / Journal of Accounting and Economics 24 (1997) 301–336 329 authored by Stern Stewart vicepresident Stephen O’Byrne (1996). After these adjustments. (2) including the natural log of capital in an attempt to capture diﬀerences in the way the market values ﬁrms of diﬀerent sizes. Capital Stern Stewart’s deﬁnition of assets (net of depreciation) invesR\ ted in goingconcern operating activities. he draws inferences by comparing the magnitudes of Rs. First. it would follow from our ﬁndings that EBEI is a better proxy for future EVA than is EVA. The initial relations tested in O’Byrne (before ‘adjustments’) are: M» /capital " R R\ M» /capital " R R\ where M»t/capital the market value of debt plus equity deﬂated by beginning of R\ period capital. If so.e.G. Next. O’Byrne argues that a nonzero intercept makes predicted M» a function of Capital and therefore an EVA model in disguise. There are three main diﬀerences between O’Byrne’s research and our tests reported above. and O’Byrne further argues that a pure NOPAT model should be forced through the origin (p. contributed and retained debt and equity capital. NOPA¹ —k (capital ) R R R\ net operating proﬁts after tax for year t. while we draw inferences by relying on formal statistical tests of relative information content. Biddle et al. O’Byrne uses market value of the ﬁrm (debt plus equity) as the dependent variable while we use marketadjusted returns.
the EBEI regression has a signiﬁcantly higher association with ﬁrm value (adjusted R"53%) than the EVA regression (50%). as we have shown. (11) in a note to the table). And.2. Omitting industry dummies from Eq. O’Byrne (1996) (p. / Journal of Accounting and Economics 24 (1997) 301–336 higher R for the ﬁnal model containing EVA (56%) than for the ﬁnal model containing NOPAT (which.3.5. it does provide a better predictor once we understand and adjust for two critical relationships between EVA and market value. With this ‘level playing ﬁeld.843 ﬁrmyear observations obtained from Stern Stewart as described earlier. NOPAT and EBEI as competing performance measures and apply O’Byrne’s three adjustments to each variable (as described in regression Eq. unlike NOPAT or other earnings measures like net income or earnings per share.4 and 5. we add EBEI to the consideration set and replicate O’Byrne’s ﬁnal model using 5. falls to 17%). is systematically linked to market value.6. After making the same adjustments to the NOPAT regression. similar to results reported for our returns tests above. O’Byrne reports adjusted R of 42% for the EVA model omitting industry intercept dummies. Omitting scaling by the cost of capital (k) from regression Eq. 125) concludes: “EVA.1. . the R of 49% is not signiﬁcantly diﬀerent from the EVA regression. Thus. Overall assessment of the sensitivity tests Considering jointly the sensitivity analyses of relative information content discussed in Sections 5. 5.’ EVA’s superiority disappears. because of the intercept restriction. we still ﬁnd no evidence to support the Stern Stewart claim that EVA (or RI) outperform EBEI. These results are not directly comparable with the returns results above since they employ diﬀerent functional forms and additional variables due to O’Byrne (1996). Biddle et al. In Table 8.330 G. 51% for EVA and 51% for NOPAT. (11) (while retaining scaling by k) yields Rs of 47% for EBEI.C. (11) yields Rs of 53% for EBEI. 5. we treat EVA. 43% for EVA and 41% for NOPAT. It should provide a better predictor of market value than other measures of operating performance. In contrast. results in Table 8 provide no evidence of the EVA regression outperforming earnings in explaining deﬂated ﬁrm values. With all of O’Byrne’s adjustments (including industry dummies). adjusted R is highest for EBEI in the remaining comparisons and EBEI signiﬁcantly outperforms EVA in several sensitivity tests at the 5% level. In only one case (the ‘Comp Year’ group in Table 4) does EVA and/or RI have a higher R than EBEI and this diﬀerence is not statistically signiﬁcant. 5. 5.” Given the success of earnings in our returns tests discussed above.
e. / Journal of Accounting and Economics 24 (1997) 301–336 331 Table 8 Replication and extension of O’Byrne (1996): Tests of relative information content (H ) for EVA .4886 In terms of incremental information content. NOPAT. 6. Tests based on Vuong (1989) are qualitatively identical.413) (3) NOPAT 0. the analyses in Section 5 provide only limited evidence that components unique to EVA (i. e. First row present pvalues for comparison between ﬁrst and second ranked measures and second and third ranked measures.g. Thus. we examine the valuerelevance of EVA and residual income compared to currentlymandated performance measures — earnings and cash ﬂow from operations. comparisons are between ﬁrst and third ranked.. respectively. There is little evidence to support the Stern Stewart claim . CapChg and AcctAdj) add to the information set that includes earnings (i. Performance metrics are listed in order of Rsquares from highest (on the left) to lowest (on the right). we ﬁnd no evidence that EVA dominates earnings in its association with stock returns or ﬁrm values. CFO and Accruals). only two of the Fstatistics and none of the tstatistics on CapChg and AcctAdj are signiﬁcant at the 5% level. Summary and potential limitations Motivated by increased use in practice and increased interest in the media and among academics.. and I is a dummy variable representing industry membership. Biddle et al. (2) EVA 0. ½"a given performance measure (EVA.. On the next row. Twotailed pvalues in parentheses represent tests of the null hypothesis of no diﬀerence between pairwise comparisons of adjusted Rsquares (Biddle et al.e. See description of pvalues below.4965 ' (0.G.5321 ' (0. 1995). (11) H H R where M»"market value of debt and equity.000) Underlying regression is from O’Byrne (1996): M» /capital "b #b ½ /capital #b ½ /capital #b ln(capital ) R R\ R NMQ R\ R LCE R\ R\ #b I #e .C. Statistical tests of diﬀerences in explanatory power across performance measures are presented in parentheses below the adjusted Rsquares. NI) deﬂated by the ﬁrm’s cost of capital (as estimated by Stern Stewart).000) (0. capital "the ﬁrms’s R\ beginning of period contributed capital.843 (1) EBEI 0.. 5. where pos and neg refer to positive and negative values of the performance measure. 0 NOPAT and EBEI where the dependent variable is the market value of the ﬁrm Relative information content Rank order of R Sample size Adj. R pvalue Obs. from the sensitivity tests in Section 5.
Again the evidence does not support the superiority of EVA. Biddle et al. These could be discretionary accruals that managers use to ‘signal’ future prospects or nondiscretionary accruals that are byproducts of . E Data needed to compute EVA are not easily estimated and the market does not have these data during our test period. not future ﬂows. if the cost of capital and the amount of capital are slow to change (or the changes are predictable months or years in advance). Further. This potential issue is mitigated in tests that use alternative dependent variables (i.4. Possible reasons why we do not detect stronger valuerelevance for EVA include: E Our research design uses current realizations. Even if EVA is a good proxy for economic proﬁts. while the charge for capital and Stern Stewart’s adjustments for accounting ‘distortions’ show some marginal evidence of being incrementally important. Recall that we assume that the market has access to suﬃcient data within three months of a ﬁrm’s ﬁscal year end such that EVA (and its components) can be reliably estimated by that time. the opportunity for surprise should be larger. ﬁveyear return intervals in Section 5. On the contrary.3. realized EVA may not outperform the current realizations of other performance measures such as earnings in proxying for future equity cash ﬂows. E Stern Stewart’s estimates of the charge for capital and accounting adjustments may contain measurement error relative to what the market is using to value ﬁrms. and results reported in Section 5. In no case does EVA signiﬁcantly outperform EBEI in tests of relative information content. E In attempting to approximate economic proﬁts.332 G. of each performance measure. we use Stern Stewart’s publicly available database which does not include many custom adjustments they use for their clients.5).C. the market should long ago have impounded these data. and ﬁrm values in Section 5.3 do not lend support for the superiority of EVA over this longer return interval. Further. this diﬀerence does not appear to be economically signiﬁcant.e.. in most cases the evidence suggests that earnings outperforms EVA. / Journal of Accounting and Economics 24 (1997) 301–336 that EVA is superior to earnings in its association with stock returns or ﬁrm values. adjustments made by Stern Stewart may remove accruals that market participants use to infer ﬁrms’ future prospects. This is similar to the rationale used to explain why EBEI generally outperforms CFO in relative information content. over ﬁveyear return intervals. Equity valuation is ultimately the discounted present value of future equity cash ﬂows (or dividends or RI or EVA). twoyear return intervals that include both contemporaneous and oneyear ahead returns in Section 5. However. For example. E There exists little or no ‘surprise value’ in components unique to EVA including the capital charge and Stern Stewart’s accounting adjustments.
our evidence suggests that earnings generally dominates EVA in valuerelevance to market participants. For example.G. . Subramanyam (1996). and despite its alleged advantages for internal decision making. although for some ﬁrms EVA may be an eﬀective tool for internal decision making. Wu (1997) presents an agency model in which ﬁrms will optimally choose accounting adjustments for internal performance metrics that serve to reduce their correlation with stock returns. An avenue for future research suggested by the ﬁndings of this study is to examine more closely which components of EVA and earnings contribute to. and in contrast to claims by Stern Stewart. consistent with the notion of ‘earnings myopia’. Until further research can be conducted. For this reason. / Journal of Accounting and Economics 24 (1997) 301–336 333 the accounting process. Biddle et al. (1997) report evidence consistent with discretionary accruals increasing the informativeness of earnings. EVA may lose information content by removing valuerelevant deferred tax accruals. capitalizing R&D and marketing costs should only add to EVA’s information content given both are generally expensed in the determination of earnings. it is possible that Stern Stewart obtains a measure that is closer in level to economic proﬁts (than say EBEI). or subtract from. and Hunt et al. Research at this level of detail requires data that are currently unavailable from Stern Stewart on individual adjustments used in the calculation of EVA. It also is possible to imagine a new equilibrium in which ﬁrms would disclose EVA rather than earnings. this would subject EVA to many of the same legal and regulatory inﬂuences. in constructing EVA. Thus. 1998) reports that some adopters of EVA feel they must still base their external performance on earnings because this is the measure on which ﬁnancial analysts continue to focus. ex ante. However. and as a consequence. it does not dominate earnings in its association with stock market returns for the sample ﬁrms and period studied. but at the same time reduces its association with stock returns. by estimating taxes paid in cash (rather than tax expense). In contrast. our conclusion is that. To the contrary. information content. the market may have failed to recognize the reporting beneﬁts of EVA through the period we study. Wallace (1996.C. we do not anticipate that EVA will displace earnings for ﬁnancial reporting purposes. E In violation of our maintained hypothesis of semistrong market eﬃciency. performance measurement and incentive compensation. As more data become available. future studies will be able to assess whether market participants have come to appreciate EVA. the resulting metric might closely resemble earnings (or what earnings might become). Collins and DeAngelo (1990).
Contemporary Accounting Research 12.W... Collins. New York University. H.. R. European Accounting Association Meetings. MacKinnon. The Accounting Review 62. Iowa City. Jim Patell. WA. 1929. comprehensive report of the special committee on ﬁnancial reporting. The Accounting Review 72. 76—93. J. Testing relative information using nested likelihood models. 1981. University of Iowa.. Daley. Siegel. 1997. Northwestern University. Anthony. L. Improving business reporting — a customer focus. and workshop participants at the Chinese University of Hong Kong. Journal of Accounting. 1991. American Institute of Certiﬁed Public Accountants. Shevlin. 187—215.334 G. Biddle. National Chengchi University. 1990. D. D. New York. Burgstahler. the Herbert O. G. 1989. Relative versus incremental information content. A. Hong Kong University of Science & Technology. William Beaver.. Daley. D. Whitten Professorship and a Hong Kong Research Grants Council Direct Allocation Grant. Unpublished paper. 1996. G. University of Washington.. M. 1982b. Equity valuation and negative earnings: The role of book value of equity.C. 213—247. 183—232. Harvard Business Review 60. A. R. Mark Peecher. 91—96. Journal of Accounting and Economics 11.. R. 1—23. M. 1987. G. Recognizing the cost of interest on equity. Mary Barth.. G. 88—102. Shores. Special committee on ﬁnancial reporting. L.... Anthony. Unpublished paper. The Economics of Accountancy. D.. G. D. Canning. 1986. New York. Pincus. D.B. 713—725. Jane Kennedy. Anthony.. R. Collins. / Journal of Accounting and Economics 24 (1997) 301–336 Acknowledgements Helpful comments were provided by Bill Alberts. Seow. I. Earnings. Stanford University. City University of Hong Kong. Biddle. R... 1982a.. Auditing and Finance 6... Biddle et al. The incremental information content of accrual versus cash ﬂows. Bowen. Naomi Soderstrom. Siegel. Jerry Zimmerman. Equity interest — its time has come. 723—747.. J. IA.. Terry Shevlin. The Accounting Review 61.. 225—254. Davidson. Seow. R. Bowen. 1995.. Bowen. Estimation and Inference in Econometrics. 1994.. Burgstahler. adaptation and equity value. Burgstahler. Dave Burgstahler. . Biddle. Evidence on the relationships between earnings and various measures of cash ﬂow. Journal of Accountancy 154. Dawn Matsumoto.. George Foster.. The estimation and determinants of association between returns and earnings: Evidence from crossindustry comparisons. Funding was provided by the Accounting Development Fund at the University of Washington. L.. D. Bob Holthausen (the referee). 1997. Accounting information and corporate governance: market and analyst reactions to earnings of ﬁrms engaged in proxy contests. Ronald Press. Oxford University Press. Brian McCulloch. Alister Hunt. T. DeAngelo. Dichev. Seattle. 1973. Informational eﬃciency and the information content of earnings during the stock market crash of 1987. Xie. Accounting for the cost of equity. Harvard Business Review 51. Johnson. Journal of Accounting and Economics 13. R.. and the University of Washington. Shores... References American Institute of Certiﬁed Public Accountants..
Hotelling. University of Pennsylvania.G. P. Peasnell.R. Edey.. Walbert. The Journal of Accountancy. The information content of losses. goodwill and the superproﬁt method.. 1938. 1991. Seattle. 109—120. New York.. University of California Press. J. Bell. The Stern Stewart performance 1000: Using EVA to build market value. Who are the real wealth creators? Fortune. Teitelbaum.. O’Byrne. Accountants. S. 1989.A. C. Earnings volatility. A. Vuong. G. 116—125.. 107—116. 661—687..A.. Sabino.. 38—50. 1961. 1994.. Business valuation. E. Tully. 1994. J. Valuation and clean surplus accounting for operating and ﬁnancing activities.. R. 1997. Wallace. A.. K.W. 1976. 1996. 1965. Biddle et al. 1995. Easton. New York. The MIT Press. A. and equity value. S. H. 361—381.V..O. Stern Stewart EVA Roundtable. P. Annual survey of economic theory: the theory of depreciation.. 1985. 143—162. Hamilton.. Adopting Residual IncomeBased Compensation Plans: Evidence of Eﬀects on Management Actions. 1940.B. D. 105—117. 1996. ABACUS 17. The selection of variates for use in prediction with some comments on the general problem of nuisance parameters. Unpublished paper. An Introduction to Merchandize. 46—70. 1995. The Quest For Value. 307—333.D. 265—276. Using longwindow return studies in addressing recognition issues: An evaluation of alternative research approaches. G. 1937. Tully. 1996. The Macmillan Press Ltd. Econometrica. P. 271—283. UMI Dissertation Services. Oxford Economic Papers 28. America’s best wealth creators. The pricing of discretionary accruals. The Theory and Measurement of Business Income. Subramanyam. Edinburgh. Stewart III. 689—731. Edwards.. The law of goodwill. 1994. The Annals of Mathematical Statistics. G. G.A. Journal of Accounting and Economics 22...V. Hayn.. 1995.. Ohlson. Kay. Journal of Accounting and Economics 20.. earnings management. Lieber. Feltham. Likelihood ratio tests for model selection and nonnested hypotheses.. EVA and market value. Journal of Applied Corporate Finance 6. MA. J. New York.. . Earnings. America’s greatest wealth creators.C.C. J.. Berkeley. Preinreich.. T. Fortune 128.. 1996. Journal of Business Finance and Accounting 9. Principles of Economics.B. Harper Business. 1995. 1982. Ohlson.A. July. Harris. 219—231. The Accounting Review 12. Division Performance: Measurement and Control. S. 1777. Peasnell. Contemporary Accounting Research 11. 52—67.D. Journal of Applied Corporate Finance 9. Lys. 1996. / Journal of Accounting and Economics 24 (1997) 301–336 335 Dechow. 28—50. K. Some formal connections between economic values and yields and accounting numbers. T. 1997. PA. University of Washington. too. 71—84. H. Contemporary Accounting Research 11.. Fortune 136... 1890. Financial Executives Research Foundation. Earnings as an explanatory variable for returns. EVA: fact or fantasy? Journal of Applied Corporate Finance 7.. Q.. R. January/February.. Marshall. Creating stockholder wealth. January.A. 19—36. Unpublished paper. 1936. and dividends in security valuation.D. T. 1981. 447—460. Philadelphia. 125—153. J. On capital budgeting and income measurement. Cambridge... R. Journal of Applied Corporate Finance 7... 317—329. P. The real key to creating wealth.. K. Goodwill in accountancy. Hunt. Solomons. Fortune 130. Accountancy. 1994. S. Preinreich. Preinreich. 134. G. G. Kennedy. London. Stewart III.A. L. A Guide to Econometrics. book value. Fisher.. 250—281. 1993. Shevlin. E. WA.. Ann Arbor. 1991. Moyer. 1957. could be happy in a golden age: The accountant’s rate of proﬁt and the internal rate of return. Econometrica 57. Journal of Accounting Research 29. Fortune 132.
/ Journal of Accounting and Economics 24 (1997) 301–336 Wallace. . J. M.. 817—838. Wallace. New York University.. (this issue). NY. Biddle et al. A heteroskedasticityconsistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica 48. Journal of Accountancy 180 (August) 47—49. J. Zarowin. Unpublished paper. 1995. The future of ﬁnance. Adopting residual incomebased compensation plans: Do you get what you pay for? Journal of Accounting and Economics 24.336 G. S. 1998... 1—15. 1997.C. EVA ﬁnancial systems: Management perspectives. 1997. 1980.. Who makes accounting ‘adjustments’ and why? An agency perspective. White. New York. Wu. H. Advances in Management Accounting 6.
This action might not be possible to undo. Are you sure you want to continue?