edu
¹This paper has beneﬁted from the comments of seminar participants at the Australian Graduate
School of Management, the University of Chicago, the University of Rochester’s Journal of
Accounting and Economics Conference and the University of Texas at Austin. We are grateful for the
comments of Andrew Alford, Brian Bushee, Ilia Dichev, John Hand, Trevor Harris, Bob Kaplan,
S.P. Kothari (the editor), James Myers and Scott Richardson. We are particularly grateful for the
detailed comments and suggestions of Bill Beaver (the referee and discussant) and Jim Ohlson (see
Ohlson, 1998). We thank I/B/E/S for the use of analyst forecast data. All views and errors are our
own.
Journal of Accounting and Economics 26 (1999) 1—34
An empirical assessment of the residual income
valuation model¹
Patricia M. Dechow, Amy P. Hutton, Richard G. Sloan*
School of Business Administration, University of Michigan, 701 Tappan Street, Ann Arbor,
MI 481091234, USA
Graduate School of Business Administration, Harvard University, Boston, MA 02163, USA
Received 1 October 1997; received in revised form 1 October 1998
Abstract
This paper provides an empirical assessment of the residual income valuation model
proposed in Ohlson (Ohlson, J.A., 1995. Earnings, book values and dividends in security
valuation. Contemporary Accounting Research 11, 661—687). We point out that existing
empirical research relying on Ohlson’s model is similar to past research relying explicitly
on the dividenddiscounting model. We establish that the key original empirical impli
cations of Ohlson’s model stem from the information dynamics that link current in
formation to future residual income. Our empirical results generally support Ohlson’s
information dynamics. However, we ﬁnd that our empirical implementation of Ohlson’s
model provides only minor improvements over existing attempts to implement the
dividenddiscounting model by capitalizing shortterm earnings’ forecasts in perpetu
ity. 1999 Published by Elsevier Science B.V. All rights reserved.
JEL classiﬁcation: M41; G14
Keywords: Capital markets; Valuation models
01654101/99/$ — see front matter 1999 Published by Elsevier Science B.V. All rights reserved.
PII: S 0 1 6 5  4 1 0 1 ( 9 8 ) 0 0 0 4 9  4
` See Palepu et al. (1996) for a discussion of the application of the model to equity valuation.
1. Introduction
A recent paper by Ohlson (1995) has stimulated interest in the residual income
formulation of the dividend discounting valuation model. This development has
potentially important implications for empirical researchers, as Ohlson’s model
speciﬁes the relation between equity values and accounting variables such as
earnings and book value. Existing empirical research has generally provided
enthusiastic support for the model, and the model is now proposed as an
alternative to the discounted cash ﬂow model in equity valuation.` Existing
empirical research argues that the model breaks new ground on two fronts.
First, the model predicts and explains stock prices better than the models based
on discounting shortterm forecasts of dividends and cash ﬂows (Bernard, 1995;
Penman and Sougiannis, 1996; Francis et al., 1997). Second, the model provides
a more complete valuation approach than popular alternatives (Frankel and
Lee, 1998).
In this paper, we evaluate the empirical implications of Ohlson’s model.
Central to our analysis is the incorporation of the residual income information
dynamics in Ohlson (1995). Past empirical applications of the residual income
valuation model ignore Ohlson’s information dynamics. In many cases, the
resulting valuation models are similar to past applications of the dividend
discounting model that capitalize current or forecasted earnings, but make no
appeal to book value or residual income (e.g., Whitbeck and Kisor, 1963;
Malkiel and Cragg, 1970; Kothari and Zimmerman, 1995).
Consistent with Ohlson’s information dynamics, we ﬁnd that residual income
follows a mean reverting process. In addition, we show that the rate of mean
reversion is systematically associated with ﬁrm characteristics suggested by
accounting and economic analysis. The rate of mean reversion is decreasing in
the quality of earnings, increasing in the dividend payout ratio and correlated
across ﬁrms in the same industry. We also ﬁnd that incorporating information in
analysts’ forecasts of earnings into the information dynamics increases forecast
accuracy. This result highlights the importance of information other than
current residual income in forecasting future residual income.
Our pricing tests indicate that stock prices partially reﬂect the mean reversion
in residual income. An important implication of this result is that book value
conveys additional information over earnings in explaining contemporaneous
stock prices. However, we also ﬁnd that book value provides very little addi
tional information about stock prices beyond that contained in analysts’ fore
casts of next year’s earnings. This result is somewhat surprising, because
analysts’ forecasts of next year’s earnings do not fully capture the longterm
mean reversion in residual income. Further tests help reconcile these seemingly
2 P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34
contradictory results by suggesting that observed stock prices seem to display
a lagged response to the longterm mean reversion in residual income.
We conclude that Ohlson’s formulation of the residual income valuation
model provides a parsimonious framework for incorporating information in
earnings, book value and earnings forecasts in empirical research. We illustrate
how many of the valuation relations implicit in past empirical research can be
considered as special cases of Ohlson’s model. However, we also ﬁnd that past
earnings and book value convey relatively little information about ﬁrm value
beyond that reﬂected in analysts’ forecasts of next year’s earnings. Thus, while
the model provides a unifying framework for earningsbased valuation research,
our eﬀorts at implementing the model provide only modest improvements in
explanatory power over past empirical research using analysts’ earnings fore
casts in conjunction with the traditional dividenddiscounting model. Neverthe
less, an important shortcoming of past research is that the relation between
earnings forecasts and future dividends has been speciﬁed in an ad hoc fashion.
By formalizing the information dynamics, Ohlson’s model provides a guiding
framework for future valuation research.
The remainder of the paper is organized as follows. Section 2 reviews
Ohlson’s formulation of the residual income valuation model and identiﬁes the
model’s empirical implications. Section 3 describes our research design and
variable measurement. Section 4 presents the empirical results and Section 5
concludes.
2. Model development
This section provides an empirically oriented review of the residual income
valuation model developed in Ohlson (1995). Our review emphasizes that the
model is a restated and restricted version of the standard dividenddiscounting
model. Empirical applications of the model that ignore Ohlson’s restrictions on
the timeseries properties of residual income are diﬃcult to distinguish from
empirical applications based on the standard dividend discounting model. We
illustrate this point with reference to existing empirical research employing the
residual income valuation model. We complete the section by outlining the key
issues in the empirical implementation of Ohlson’s valuation model.
2.1. Model review
The model is comprised of three basic assumptions. First, price is equal to the
present value of expected dividends:
P
R
"
`
O¯¹
E
R
[d
R>O
]
(1#r)O
, (1)
P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34 3
where P
R
is the price of the ﬁrm’s equity at time t, d
R
is net dividends paid at time
t, r is the (assumed constant) discount rate, E
R
[ ] is the expected value operator
conditioned on date t information.
Second, the clean surplus accounting relation:
b
R
"b
R¹
#x
R
!d
R
, (2)
where b
R
is the book value of equity at time t, and x
R
is earnings for the period
from t!1 to t.
This assumption allows future dividends to be expressed in terms of future
earnings and book values. Combining the clean surplus relation in Eq. (2) with
the dividend discounting model in Eq. (1) yields:
P
R
"
`
O¯¹
E
R
[b
R>O¹
#x
R>O
!b
R>O
]
(1#r)O
. (3)
Simple algebraic manipulation allows Eq. (3) to be rewritten as
P
R
"b
R
#
`
O¯¹
E
R
[x
R>O
!r.b
R>O¹
]
(1#r)O
!
E
R
[b
R>`
]
(1#r)`
. (4)
The ﬁnal term in Eq. (4) is assumed to be zero, and ‘residual income’ or
‘abnormal earnings’ is deﬁned as
x
R
"x
R
!r.b
R¹
so that price can be expressed as the sum of book value and the present value of
future abnormal earnings:
P
R
"b
R
#
`
O¯¹
E
R
[x
R>O
]
(1#r)O
. (5)
Eq. (5) is the residual income version of the dividenddiscounting model. It is
important to note that Eq. (5) is just a restatement of the dividenddiscounting
model which in no way depends on the properties of accounting numbers other
than through the clean surplus relation. For example, given a stream of future
dividends, the value of b
R
and the values all the x
R>O
s could be picked as random
numbers. So long as the b
R>O
s are updated according to Eq. (2), the valuation
relation in Eq. (5) will yield the present value of the dividend stream. Another
way of illustrating the independence of Eq. (5) from accrual accounting concepts
is to redeﬁne b
R
as the ﬁrm’s cash balance at the end of period t and x
R
as the net
eﬀect of all nondividend cash ﬂows for period t. The resulting variables clearly
satisfy the clean surplus relation embodied in Eq. (2), and so the resulting
‘residual cash ﬂow valuation model’ is also a legitimate reformulation of the
dividend discounting formula. Thus, if accrual accounting is incrementally
useful over cash accounting in the valuation process, its usefulness must stem
from properties in addition to the clean surplus assumption.
4 P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34
` Similar terminal value assumptions are used by Francis et al. (1997), Lee et al. (1998) and
Penman and Sougiannis (1996).
From an empirical standpoint, Eq. (5) leaves the researcher in much the same
position as the dividenddiscounting model. The valuation relation cannot be
implemented without estimates of future book values. In order to estimate
future book values, the researcher must estimate future dividends. However,
once future dividends are estimated, the book value and earnings estimates
become redundant, and the researcher may just as well have used the dividend
discounting model in Eq. (1).
The above point is subtle, and overlooking it can lead empiricists to imple
ment the residual income valuation model by incorporating explicit estimates of
future dividends, without realizing that this makes the appeal to the residual
income formulation of the dividend discounting model somewhat redundant.
The point is illustrated by a recent application of the residual income valuation
model in Frankel and Lee (1998). They implement Eq. (5) by forecasting abnor
mal earnings for three periods and taking the last period in perpetuity as
follows:`
P
R
"b
R
#
f (1)
R
!r.b
R
(1#r)
#
f (2)
R
!r.b(1)
R
(1#r)`
#
f (3)
R
!r.b(2)
R
(1#r)`.r
,
where f (i)
R
is the period t consensus analyst forecast of earnings for period t#i,
b(i)
R
is b(i!1)
R
#f (i)
R
!d(i)
R
(the period t forecast of book value for period t#i),
and d(i)
R
is period t forecast of dividends for period t#i.
As a matter of algebra, this valuation expression reduces to
P
R
"
d(1)
R
(1#r)
#
d(2)
R
(1#r)`
#
f (3)
R
(1#r)`.r
.
Thus, the valuation model can be viewed as an application of the dividend
discounting formula in which explicit forecasts of dividends are provided for
the ﬁrst two periods and dividends are assumed to equal the forecast of period
t#3 earnings thereafter. The valuation model is readily interpretable in the
context of the original dividenddiscounting model, and the appeal to the
residual income formulation of the dividenddiscounting model is redundant. It
is also noteworthy that the book value of equity drops out of this particular
model.
The redundancy of the residual income valuation model applies more gener
ally to studies that generate explicit forecasts of earnings and book values (and
hence dividends) for several periods, and then use a terminal value assumption
to complete the valuation (e.g., Frankel and Lee, 1998; Francis et al., 1997).
Penman (1997) demonstrates how some of the more common terminal value
P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34 5
assumptions employed in the residual income valuation model are readily
interpretable in the context of the standard dividenddiscounting framework.
Thus, while the residual income formulation of the dividenddiscounting model
may have intuitive appeal because of its focus on accounting numbers, it
provides no new empirical implications in and of itself.
Both Ohlson (1995) and Lundholm (1995) emphasize that the original empiri
cal implications of Ohlson’s model depend critically on the third and ﬁnal
assumption regarding the abnormal earnings information dynamics. This as
sumption places restrictions on the standard dividenddiscounting model. From
a theoretical perspective, the ﬁrm is still being valued by discounting future
dividends. However, the third assumption speciﬁes the nature of the relation
between current information and the discounted value of future dividends.
Ohlson’s third assumption is that abnormal earnings satisfy the following
modiﬁed autoregressive process:
x
R>¹
"x
R
#v
R
#
¹R>¹
, (6a)
v
R>¹
"v
R
#
`R>¹
, (6b)
where v
R
is information about future abnormal earnings not in current abnormal
earnings,
GR
is the unpredictable, mean zero disturbance term, and and are
ﬁxed persistence parameters that are nonnegative and less than one.
Combining the residual income valuation model in Eq. (5) with the informa
tion dynamics in Eqs. (6a) and (6b) yields the following valuation function:
P
R
"b
R
#
¹
x
R
#
`
v
R
, (7)
where
¹
"/(1#r!) and
`
"(1#r)/[(1#r!)(1#r!)].
This valuation function does not require explicit forecasts of future dividends,
nor does it require additional assumptions about the computation of ‘terminal
value’. The information dynamics in Eqs. (6a) and (6b) along with the valuation
function in Eq. (7) embody the original empirical implications of Ohlson (1995).
2.2. Empirical implementation
Empirical implementation of the information dynamics in Eqs. (6a) and (6b)
and the valuation function in Eq. (7) requires three variables (b
R
, x
R
and v
R
) and
three parameters (, and r) to be provided as inputs. The ﬁrst two variables,
book value (b
R
) and earnings (x
R
), are readily available and easily measured. The
remaining variable, v
R
, and the three parameters are more diﬃcult to measure.
Turning ﬁrst to v
R
, it is well established that prices reﬂect information about
future earnings that is not contained in current earnings. Attempts to incorpor
ate this other information into valuation analyses date back at least as far as
Beaver et al. (1980). Eq. (6a) indicates that Ohlson deﬁnes his other information
variable, v
R
, as the diﬀerence between the conditional expectation of abnormal
6 P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34
" We are grateful to Jim Ohlson for suggesting this procedure for measuring v
R
(see Ohlson, 1998).
earnings for period t#1 based on all available information and the expectation
of abnormal earnings based only on current period abnormal earnings:
v
R
"E
R
[x
R>¹
]!x
R
.
Note that the conditional expectation of period t#1 abnormal earnings is
equal to the conditional expectation of period t#1 earnings less the product of
period t book value and the discount rate. We measure the period t conditional
expectation of period t#1 earnings using the consensus analyst forecast of
period t#1 earnings, denoted f
R
, so that
E
R
[x
R>¹
]"f
R
"f
R
!r.b
R
.
The other information, v
R
can then be measured as"
v
R
"f
R
!x
R
.
Finally, values for the three parameters , and r, must be established. We use
the average historical return on equities to measure r. We measure and using
their historical unconditional sample estimates. The estimation procedure is
described in more detail in Section 3. We refer to these estimates as and S,
respectively. We also develop a conditional forecast of using characteristics
suggested by accounting and economic analysis, which we refer to as . Details
of the estimation procedure are again provided in Section 3. The characteristics
that we use are described in more detail below.
The persistence of abnormal earnings is a function of the persistence of the
abnormal accounting rate of return and the growth rate in book value. Thus,
variables that forecast the persistence of accounting rates of return and the
growth rate in book value will determine . The extant accounting literature has
identiﬁed a number of factors aﬀecting the persistence of accounting rates of
return. First, Brooks and Buckmaster (1976) and Freeman et al. (1982) provide
evidence that extreme levels of earnings and extreme accounting rates of return
mean revert more quickly. Thus, we expect that will be smaller for ﬁrms with
extreme abnormal accounting rates of return. Second, it is well established that
nonrecurring special items, such as restructuring charges and asset write
downs, are less likely to persist (e.g., Fairﬁeld et al., 1996), so we expect that
will be lower for ﬁrms with extreme levels of special items. Third, Sloan (1996)
establishes that accounting rates of return are less likely to persist for ﬁrms with
extreme levels of operating accruals, so we expect that will be lower for ﬁrms
with extreme levels of operating accruals. Economic analysis points us to two
factors that are expected to relate to the persistence of abnormal earnings. First,
dividend policy serves as an indicator of expected future growth in the book
value of equity. Firms with growth opportunities tend to have lower payout
P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34 7
ratios. (e.g., Fazzari et al., 1988; Anthony and Ramesh, 1992). Thus, we expect
that ﬁrms with low payout policies will experience growth in the book value of
equity in the future, resulting in a higher . Second, we predict that a variety of
industryspeciﬁc factors should inﬂuence the persistence of abnormal earnings.
In particular, numerous studies suggest a link between industry structure and
ﬁrm proﬁtability (e.g., Scherer, 1980; Ahmed, 1994). We assume that the eﬀect of
industry speciﬁc factors should be relatively stable over time. We therefore
expect that the persistence of abnormal earnings should be increasing in the
historical persistence of abnormal earnings for ﬁrms in the same industry.
3. Research design
3.1. Model evaluation
We evaluate the empirical implications of Ohlson’s residual income valuation
model relative to several competing accountingbased valuation models. The
competing valuation models generally correspond to valuation models that
have been used in previous empirical research, and we show that they can all be
considered as special cases of Ohlson’s model. Our empirical analysis focuses on
the improvements provided by Ohlson’s model over these simpler and more
restrictive models. The additional restrictions range from ignoring the ‘other
information’ in analysts’ forecasts of earnings altogether, to setting the persist
ence parameters and to their polar extremes of 0 and 1. The competing
valuation models are summarized in Fig. 1.
The rows of Fig. 1 each summarize valuation models that make alternative
assumptions about the value of the abnormal earnings persistence parameter,
. The four rows consider values for of 0, 1, the unconditional estimate ()
and the conditional estimate (), respectively. The columns of Fig. 1 each
summarize valuation models that make alternative assumptions about the other
information variable, v
R
. The ﬁrst column ignores other information altogether,
and is therefore restricted to valuation models based on past abnormal earnings
alone. The remaining three columns summarize valuation models that incorpor
ate the other information variable into the valuation analysis. The columns
diﬀer with respect to the assumed value of the other information persistence
parameter, . The three columns consider values for of 0, 1 and the uncondi
tional estimate, S, respectively. Note that we superscript by the abnormal
earnings persistence parameter, . This is because we estimate S from a v
R
autoregression, and the measurement of v
R
depends on the value used for .
A priori, we are able to rule out several of the combinations of assumptions
about the parameters and . First, we rule out the use of with models
incorporating the other information variable, v
R
. We do this because several of
the conditioning variables relate to shortterm mean reversion in abnormal
8 P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34
` For example, if current abnormal earnings consist of a large negative special item, then we would
expect earnings to be temporarily low this period, resulting in a low conditional persistence
parameter, . However, we do not expect that a corresponding special item will be reported in next
period’s earnings. Thus, it makes little sense to apply the low conditional persistence of this period’s
abnormal earnings to the expectation of next period’s abnormal earnings.
earnings that is not necessarily expected to persist beyond the next period. Thus,
it makes little sense to apply the conditional persistence parameter for this
period’s abnormal earnings to the conditional expectation of next period’s
abnormal earnings.` Second, we rule out cases where one of the persistence
parameters is assumed to be 1 and the other persistence parameter is assumed to
be strictly positive. This combination of assumptions implies that abnormal
earnings are nonstationary. We ﬁnd this implication unappealing from an
economic standpoint, as it suggests that abnormal proﬁt opportunities will
never be competed away. This implication is also inconsistent with past empiri
cal evidence suggesting that accounting rates of return are mean reverting
(Freeman et al., 1982; Fairﬁeld et al., 1996).
Cells in Fig. 1 corresponding to one of the valuation models that we rule out
above are labeled ‘Not considered’. The remaining cells, list both the expectation
of next period’s abnormal earnings and the valuation function implied by the
corresponding valuation model. Below, we brieﬂy discuss each of the valuation
models and provide examples of prior research using the valuation models.
"0, ignore other information
This model assumes that expectations of future abnormal earnings are based
solely on information in current abnormal earnings and that abnormal earnings
are purely transitory. Consequently, expected future abnormal earnings are zero
and price is equal to book value. This restricted version of Ohlson’s model
corresponds to valuation models in which accounting earnings are assumed to
measure ‘value creation’ (e.g., Easton and Harris, 1991). Variants of this valu
ation model are implicit in many ‘levels’ studies in which market values are
regressed on book values (e.g., Barth, 1991).
"1, ignore other information
This model assumes that expectations of future abnormal earnings are based
solely on current abnormal earnings and that abnormal earnings persist indeﬁ
nitely. These assumptions imply that expected abnormal earnings equal current
abnormal earnings and price equals current earnings capitalized in perpetuity
plus any reinvested period t earnings. The intuition for including reinvested
period t earnings is that they will increase the book value base that is available
to generate earnings in the next period. This special case of Ohlson’s model
corresponds closely to the popular earnings capitalization valuation model in
P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34 9
10 P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34
F
i
g
.
1
.
S
u
m
m
a
r
y
o
f
t
h
e
i
m
p
l
i
c
a
t
i
o
n
s
o
f
t
h
e
a
c
c
o
u
n
t
i
n
g

b
a
s
e
d
v
a
l
u
a
t
i
o
n
m
o
d
e
l
s
e
x
a
m
i
n
e
d
i
n
t
h
e
e
m
p
i
r
i
c
a
l
t
e
s
t
s
.
E
a
c
h
c
e
l
l
c
o
n
t
a
i
n
s
t
h
e
f
o
r
e
c
a
s
t
o
f
n
e
x
t
p
e
r
i
o
d
’
s
a
b
n
o
r
m
a
l
e
a
r
n
i
n
g
s
(
E
R
[
x
? R
>
¹
]
)
a
n
d
t
h
e
c
o
n
t
e
m
p
o
r
a
n
e
o
u
s
f
o
r
e
c
a
s
t
o
f
s
t
o
c
k
p
r
i
c
e
(
P
R
)
f
o
r
t
h
e
r
e
s
p
e
c
t
i
v
e
m
o
d
e
l
.
A
b
n
o
r
m
a
l
e
a
r
n
i
n
g
s
f
o
r
y
e
a
r
t
i
s
d
e
ﬁ
n
e
d
a
s
x
R
"
x
R
!
r
.
b
R
,
w
h
e
r
e
x
R
d
e
n
o
t
e
s
e
a
r
n
i
n
g
s
b
e
f
o
r
e
e
x
t
r
a
o
r
d
i
n
a
r
y
i
t
e
m
s
a
n
d
d
i
s
c
o
n
t
i
n
u
e
d
o
p
e
r
a
t
i
o
n
s
f
o
r
y
e
a
r
t
;
r
d
e
n
o
t
e
s
t
h
e
d
i
s
c
o
u
n
t
r
a
t
e
(
a
s
s
u
m
e
d
t
o
b
e
1
2
%
)
;
b
R
d
e
n
o
t
e
s
b
o
o
k
v
a
l
u
e
o
f
e
q
u
i
t
y
a
t
t
h
e
e
n
d
o
f
y
e
a
r
t
;
i
s
t
h
e
ﬁ
r
s
t
o
r
d
e
r
a
u
t
o
r
e
g
r
e
s
s
i
o
n
c
o
e
ﬃ
c
i
e
n
t
f
o
r
a
b
n
o
r
m
a
l
e
a
r
n
i
n
g
s
a
n
d
i
s
e
s
t
i
m
a
t
e
d
u
s
i
n
g
a
l
l
h
i
s
t
o
r
i
c
a
l
l
y
a
v
a
i
l
a
b
l
e
d
a
t
a
f
r
o
m
1
9
5
0
t
h
r
o
u
g
h
t
h
e
f
o
r
e
c
a
s
t
y
e
a
r
i
n
a
p
o
o
l
e
d
t
i
m
e

s
e
r
i
e
s
c
r
o
s
s

s
e
c
t
i
o
n
a
l
r
e
g
r
e
s
s
i
o
n
;
i
s
t
h
e
p
r
e
d
i
c
t
e
d
v
a
l
u
e
o
f
f
r
o
m
t
h
e
r
e
g
r
e
s
s
i
o
n
m
o
d
e
l
s
p
e
c
i
ﬁ
e
d
i
n
T
a
b
l
e
2
a
n
d
e
s
t
i
m
a
t
e
d
u
s
i
n
g
a
l
l
h
i
s
t
o
r
i
c
a
l
l
y
a
v
a
i
l
a
b
l
e
d
a
t
a
f
r
o
m
1
9
5
0
t
h
r
o
u
g
h
t
h
e
f
o
r
e
c
a
s
t
y
e
a
r
;
i
s
t
h
e
ﬁ
r
s
t
o
r
d
e
r
a
u
t
o
r
e
g
r
e
s
s
i
o
n
c
o
e
ﬃ
c
i
e
n
t
f
o
r
t
h
e
o
t
h
e
r
i
n
f
o
r
m
a
t
i
o
n
v
a
r
i
a
b
l
e
,
v
R
,
a
n
d
i
s
e
s
t
i
m
a
t
e
d
u
s
i
n
g
a
l
l
h
i
s
t
o
r
i
c
a
l
l
y
a
v
a
i
l
a
b
l
e
d
a
t
a
f
r
o
m
1
9
5
0
t
h
r
o
u
g
h
t
h
e
f
o
r
e
c
a
s
t
y
e
a
r
i
n
a
p
o
o
l
e
d
t
i
m
e

s
e
r
i
e
s
c
r
o
s
s

s
e
c
t
i
o
n
a
l
r
e
g
r
e
s
s
i
o
n
.
v
R
i
s
d
e
ﬁ
n
e
d
a
s
v
R
"
f
R
!
x
R
,
w
h
e
r
e
t
h
e
p
e
r
i
o
d
t
c
o
n
s
e
n
s
u
s
a
n
a
l
y
s
t
f
o
r
e
c
a
s
t
o
f
a
b
n
o
r
m
a
l
e
a
r
n
i
n
g
s
f
o
r
t
h
e
n
e
x
t
p
e
r
i
o
d
i
s
d
e
ﬁ
n
e
d
a
s
f
R
"
f
R
!
r
.
b
R
f
R
d
e
n
o
t
e
s
t
h
e
I
/
B
/
E
/
S
c
o
n
s
e
n
s
u
s
f
o
r
e
c
a
s
t
o
f
e
a
r
n
i
n
g
s
f
o
r
y
e
a
r
t
#
1
m
e
a
s
u
r
e
d
i
n
t
h
e
ﬁ
r
s
t
m
o
n
t
h
f
o
l
l
o
w
i
n
g
t
h
e
a
n
n
o
u
n
c
e
m
e
n
t
o
f
e
a
r
n
i
n
g
s
f
o
r
y
e
a
r
t
.
P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34 11
which earnings are assumed to follow a random walk and the future dividend
payout ratio is assumed to be 100% (e.g., Kothari, 1992; Kothari and Zimmer
man, 1995). The model is considered in more detail in Ohlson (1991). An
important feature of the model is that book value does not enter the valuation
function.
", ignore other information
This model assumes that expectations of future abnormal earnings are based
solely on current abnormal earnings, and that abnormal earnings mean revert at
their unconditional historical rate. Expected abnormal earnings equal current
abnormal earnings multiplied by the persistence parameter, . Price is a linear
function of book value and current abnormal earnings. The relative weight on
book value (abnormal earnings) is decreasing (increasing) in the persistence
parameter, ". Thus, this model combines elements of the two preceding
models.
", ignore other information
This model is identical to the model discussed directly above, except that the
unconditional estimate of the persistence parameter () is replaced by the
conditional estimate of the persistence parameter ().
"0, "0
This valuation model incorporates the other information in the conditional
forecast of next period’s abnormal earnings, but assumes that both abnormal
earnings and the other information variable are purely transitory. Expected
abnormal earnings are equal to the consensus analyst forecast of abnormal
earnings. Note that expected abnormal earnings equal the consensus analyst
forecast of abnormal earnings by construction for all of the models incorporat
ing the other information variable. Price is equal to book value plus the
discounted value of the forecast of next period’s abnormal earnings. Abnormal
earnings have no implications for ﬁrm value beyond next period, because
forecasted abnormal earnings are assumed to be purely transitory. Consequently,
current book value receives a heavy weighting in the valuation function. This
model corresponds to Penman and Sougiannis’ (1996) application of the residual
income valuation model with a one period horizon and no terminal value.
"1, "0
Unlike the prior model, this model assumes that forecasted abnormal earn
ings persist indeﬁnitely, so price is equal to the forecast of next period’s earnings
capitalized in perpetuity. Variants of this model have long been popular in
empirical applications of the dividenddiscounting model. For example, Whit
beck and Kisor (1963) and Vander Weider and Carleton (1988) model the ratio
of price to the consensus analyst forecast of next period earnings as a function of
12 P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34
the dividend payout ratio and expected growth in earnings. More recently, this
model has formed the basis for the computation of terminal values in empirical
applications of the residual income valuation model using ﬁnite horizon data.
Frankel and Lee (1998), Lee et al. (1998), Penman and Sougiannis (1996) and
Francis et al. (1997) all compute terminal value by assuming that abnormal
earnings in the terminal year either remain constant in perpetuity, or grow at
some nominal rate (e.g., 4%). Note, however, that this model does not allow for
mean reversion in abnormal earnings, and so does not place any weight on
current book value in the valuation function.
", "0
This model allows for gradual mean reversion in next period’s expected
abnormal earnings by assuming that equals its historical unconditional value.
Price is a linear function of book value and the forecast of next period’s
abnormal earnings. The relative weight on book value (forecasted abnormal
earnings) is decreasing (increasing) in the persistence parameter, ". Thus,
this model combines elements of the two preceding models. While this model is
appealing in that it combines analysts’ forecast data with information in book
value, it has received little attention in the empirical literature. Bernard (1995)
captures the spirit of this model by regressing price on book value and short
term forecasts of abnormal earnings.
"0, "1
This valuation model is identical to the model obtained by assuming that
("1, "0), which is discussed above. The intuition for this result is that
"0 implies that v measures the complete expectation of next period’s abnor
mal earnings. Assuming that "1 then has the eﬀect of allowing the expectation
of next period’s abnormal earnings to persist indeﬁnitely. More generally, note
that the valuation function is always symmetric in and (see Ohlson, 1998).
"0, "S
The symmetry of the valuation function implies that this valuation model is
identical to the model obtained by assuming that (", "0), which is
discussed above. One apparent diﬀerence between the two models that is evident
from Fig. 1 is the substitution of S for in the valuation function. The
diﬀerence is reconciled by noting that when "0, v captures the entire expecta
tion of next period’s abnormal earnings, so that S reﬂects the persistence of next
period’s abnormal earnings.
", "S
The ﬁnal valuation model sets both and equal to their historical uncondi
tional values. This model represents our best attempt to implement the residual
income valuation model proposed by Ohlson (1995). Allowing both abnormal
P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34 13
" An alternative procedure would be to use a deﬁnition of earnings that incorporates extraordin
ary items and to then incorporate the lower persistence of the comprehensive earnings number in the
persistence parameter, . This is the procedure that we adopt for special items.
earnings and the other information variable to each have their own persistence
parameters produces a valuation function in which price is a linear combination
of book value, current abnormal earnings and the other information variable.
This valuation model implies that book value, current abnormal earnings and
the other information embedded in the forecast of next period’s abnormal
earnings all contain incremental information about price.
3.2. Data and variable measurement
The empirical analysis uses three data sources. Historical accounting data are
obtained from the COMPUSTAT ﬁles. Our primary empirical analysis uses
annual ﬁnancial statement data from 1976 to 1995. Stock return data are
obtained from the CRSP daily ﬁles. All of our empirical tests employ with
dividend stock returns and buyandhold returns. Analyst forecast data is
obtained from the I/B/E/S ﬁles. Combining the three databases gives us a total
of 50,133 observations. The empirical analysis is conducted using pershare
data. All of our tests use earnings measured before extraordinary items. Strictly
speaking, excluding extraordinary items from earnings violates the clean surplus
assumption underlying the theoretical development of the residual income
valuation model. However, from a practical perspective, extraordinary items are
nonrecurring, and so their inclusion is unlikely to enhance the prediction of
abnormal earnings."
Our analysis requires a measure of the discount rate, r. Note that the discount
rate enters all of the models in a similar fashion, and our objective is not to
evaluate alternative methods for estimating discount rates. Moreover, attempts
to document predictable variation in expected returns that are consistent with
the predictions of asset pricing models have met with limited success. Thus, we
use a discount rate of 12%, which approximates the longrun average realized
return on US equities. The relative rankings of the models in the empirical tests
are robust to discount rates ranging from 9% to 15%.
The unconditional value of used in the Ohlson valuation model is estimated
separately for each ﬁscal year. An abnormal earnings autoregression is esti
mated using all available observations from previous years, going back as far as
1950. All variables are scaled by market value of equity to control for hetero
scedasticity, and the 1% most extreme observations are winsorised so that they
do not have an undue inﬂuence on the regressions. The resulting estimate of the
autoregressive parameter, , is used to implement the unconditional version of
Ohlson’s model.
14 P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34
The conditional value of the autoregressive parameter, , is estimated in
a similar manner. We ﬁrst construct the ﬁve variables that are hypothesized to
be associated with crosssectional variation in the persistence of abnormal
earnings. The ﬁrst variable (q1) measures the magnitude of abnormal earnings,
and is computed as the absolute value of the ratio of abnormal earnings to
lagged book value. The second variable (q2) measures the magnitude of special
items, and is computed as the absolute value of the ratio of special items to
lagged book value. The third variable (q3) measures the magnitude of operating
accruals, and is measured as the absolute value of the ratio of operating accruals
to lagged total assets. Operating accruals are computed in the standard way
(e.g., Sloan, 1996). The fourth variable (div) measures the dividend payout
policy and is computed as the ratio of dividends to earnings over the most recent
ﬁscal year. If the dividend payout ratio is negative due to negative earnings, we
use the ratio from the most recent previous year in which the ﬁrm reported
positive earnings. If the ratio is greater than one, we set it to one, because
a payout ratio greater than one cannot be sustained indeﬁnitely. The ﬁfth
variable (ind) measures the historical persistence of abnormal earnings for ﬁrms
in the same industry. We use twodigit SIC codes to measure industry member
ship. A ﬁner partitioning results in an unsatisfactorily low number of observa
tions for some industries. A pooled industryspeciﬁc abnormal earnings
autoregression is used to estimate the historical persistence parameter for each
SIC grouping. The regressions use all available observations from 1950 through
the previous year. Next, is estimated via an abnormal earnings autoregres
sion in which each of the ﬁve determinants of are included as interactive
eﬀects:
x
R
"
"
#
¹
x
R¹
#
`
(x
R¹
q1
R¹
)#
`
(x
R¹
q2
R¹
)#
"
(x
R¹
q3
R¹
)
#
`
(x
R¹
div
R¹
)#
"
(x
R¹
ind
R¹
)#
R
.
A separate regression is estimated for each ﬁscal year in the sample, with each
regression using all available observations in the sample from previous years,
going back as far as 1950. The estimate for each ﬁrmyear is then computed
using the parameter estimates from this regression and the ﬁrmyears actual
values of q1, q2, q3, div and ind:
A"
¹
#
`
q1
R
#
`
q2
R
#
"
q3
R
#
`
div
R
#
"
ind
R
.
If one of the variables required to compute is missing, then is set equal
to .
Finally, we estimate S, through an ‘other information’ autoregression, using
the same procedure that we used to estimate . One complication that arises in
the estimation of S is that the measurement of the other information variable, v,
depends on an assumed value of . Recall from Fig. 1 that we only require
P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34 15
` We also consider the model with ("0, "S). In this case, S measures the persistence of
abnormal earnings, which is given by .
a measure of in the situation where ".` Hence, we need only estimate the
other information autoregression with v measured using . We measure
v
R
using the estimate of obtained from all data available through the end of
period t!1.
4. Empirical results
4.1. Timeseries behavior of abnormal earnings
We begin our empirical analysis by evaluating how well the evolution of
abnormal earnings is described by Ohlson’s information dynamics. We test
ﬁve aspects of the timeseries behavior of abnormal earnings. First, we
examine whether the autoregressive coeﬃcient, , diﬀers reliably from the polar
extremes of 0 and 1. Second, we examine whether the ﬁrstorder autoregressive
process is suﬃcient for abnormal earnings by adding additional lags of abnor
mal earnings. Third, we relax the constraints that the autoregressive process
places on the earnings and book value components of abnormal earnings.
Fourth, we estimate by allowing the autoregressive coeﬃcient on abnormal
earnings to vary as a function of our conditioning variables. Finally, we examine
whether the autoregressive coeﬃcient, , diﬀers reliably from the polar extremes
of 0 and 1.
The ﬁrst three tests are presented in Table 1 and employ pooled timeseries
and crosssectional regression analysis. Panel A reports the results from a ﬁrst
order abnormal earnings autoregression. The autoregressive coeﬃcient,
¹
, is
0.62 with a tstatistic of 138.31. Thus, the hypotheses that
¹
"1 and
¹
"0
respectively are both strongly rejected. The plots in Fig. 2 illustrate the superior
forecasting ability of a timeseries model that incorporates the gradual mean
reversion in abnormal earnings. The plots compare the predictive ability of
timeseries models setting equal to 0, 1 and , respectively. The ﬁgure is
constructed by ﬁrst ranking all sample observations on deﬂated abnormal
earnings and equally assigning the ranked observations to deciles. The mean
values of abnormal earnings for the highest and lowest deciles are then plotted
over the next four years, along with the values implied by each of the models. It
can be readily seen that the model using " tracks subsequent abnormal
earnings the most closely. Note also that while the model using "0 does
a poor job of predicting shortterm abnormal earnings, it does a relatively good
job of tracking longterm abnormal earnings, because mean reversion in abnor
mal earnings is almost complete after four years.
16 P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34
Table 1
Autoregressive properties of abnormal earnings
Panel A: Pooled analysis with one lag
x
GR>¹
"
"
#
¹
x
GR
#
GR>¹
"
¹
R`
!0.02 0.62 0.34
(!29.04) (138.31)
Panel B: Pooled analysis with four lags
x
GR>¹
"
"
#
¹
x
GR
#
`
x
GR¹
#
`
x
GR`
#
"
x
G R`
#
GR>¹
"
¹
`
`
"
R`
!0.01 0.59 0.07 0.01 0.01 0.35
(!12.36) (68.31) (7.50) (0.86) (1.59)
Panel C: ºnconstrained estimation with one lag
x
GR>¹
"
"
#
¹
x
GR
#
`
b
GR¹
#
GR>¹
"
¹
`
R`
0.02 0.47 !0.09 0.40
(17.16) (80.12) (!77.64)
Notes: Sample consists of 50,133 annual observations from 1976 to 1995. All variables are scaled by
the market value of equity at the end of year t. Figures in parentheses are tstatistics.
Abnormal earnings for year t is deﬁned as:
x
R
"x
R
!r.b
R¹
where x
R
denotes earnings before extraordinary items for year t, r denotes the discount rate (assumed
to be 12%), and b
R
denotes book value of equity at the end of year t.
` BarYosef et al. (1996) investigate the appropriateness of the single lag information dynamic in
a more general framework and ﬁnd that a second lag achieves modest statistical signiﬁcance.
Panel B of Table 1 reports results including additional lags of abnormal
earnings to examine whether the ﬁrstorder autoregressive process is suﬃcient.
Inclusion of three additional lags of abnormal earnings has a trivial impact,
increasing the explanatory power from 0.34 to 0.35. Only the second lag is
statistically signiﬁcant (t"7.50), but the coeﬃcient magnitude is only 0.07
versus 0.59 on the ﬁrst lag. Thus, the ﬁrst order autoregressive process appears
to provide a reasonable empirical approximation.` Finally, Panel C reports the
P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34 17
Fig. 2. Comparison of the actual timeseries properties of abnormal earnings with the properties
predicted by a ﬁrstorder autoregressive process with alternative values for the autoregressive
coeﬃcient, . The graph is formed by taking observations in the extreme deciles of abnormal
earnings performance in year 0 and plotting the mean level of abnormal earnings performance for each
decile over the following four years. The sample consists of 50,133 observations from 1976 to 1995.
results of regressions of abnormal earnings on lagged abnormal earnings and the
book value component of lagged abnormal earnings. If the ﬁrstorder autoreg
ressive process is appropriate, then the additional book value term should not
load in the regression. However, we see that book value loads with a signiﬁ
cantly negative coeﬃcient and that the inclusion of book value leads to a decline
in the coeﬃcient on abnormal earnings. Feltham and Ohlson (1995) suggest that
the negative loading on book value can be interpreted as ‘aggressive’ ac
counting. However, unreported tests reveal that this unconstrained speciﬁcation
is not signiﬁcantly helpful in forecasting future abnormal earnings and so it is
not pursued further.
18 P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34
Table 2
Determinants of the persistence of abnormal earnings
x
R
"
"
#
¹
x
R¹
#
`
(x
R¹
q1
R¹
)#
`
(x
R¹
q2
R¹
)#
"
(x
R¹
q3
R¹
)#
`
(x
R¹
div
R¹
)
#
"
(x
R¹
ind
R¹
)#
R
"
¹
`
`
"
`
"
R`
Predicted
sign
? ? ! ! ! ! #
!0.02 0.61 !0.37 !1.21 !0.17 !0.11 0.61 0.40
(!30.97) (13.22) (!28.68) (!35.59) (!3.77) (!7.80) (8.10)
Notes: Sample consists of 50,133 observations from 1976 to 1995. Abnormal earnings are scaled by
market value of equity at the end of year t. Figures in parentheses are tstatistics.
Abnormal earnings for year t is deﬁned as
x?
R
"x
R
!r.b
R¹
where x
R
denotes earnings before extraordinary items and discontinued operations for year t,
r denotes the discount rate (assumed to be 12%), and b
R
denotes book value of equity at the end of
year t;
q1
R
is deﬁned as the absolute value of abnormal earnings for year t divided by book value of equity at
the beginning of year t;
q2
R
is deﬁned as the absolute value of special accounting items divided by book value of equity at the
beginning of year t;
q3
R
is deﬁned as the absolute value of accounting accruals divided by total assets at the beginning of
year t;
div
R
is deﬁned as dividends paid during year t divided by earnings before extraordinary items and
discontinued operations for year t;
ind
R
is deﬁned as the ﬁrst order autoregressive coeﬃcient from an abnormal earnings autoregression
for all ﬁrms in the same two digit SIC code as the observation. The autoregression is conducted
using all available ﬁrms on the COMPUSTAT annual tapes in the same two digit SIC code from
1950 through year t.
Table 2 analyses variation in the autoregressive coeﬃcient,
¹
. Recall that
this coeﬃcient measures the persistence of abnormal earnings and is hy
pothesized to have ﬁve determinants. Persistence is hypothesized to be lower
when earnings contain more transitory accounting items, measured by the
empirical constructs q1, q2 and q3. Persistence is also hypothesized to be
decreasing in the dividend yield (div) and increasing in the historical level of
industrywide abnormal earnings persistence (ind). Table 2 reports results from
allowing each of the hypothesized determinants of persistence to enter as
interactive variables in the abnormal earnings autoregression. Inclusion of the
ﬁve interactive eﬀects increases the explanatory power of the regression from
0.34 to 0.40. All of the interactive eﬀects enter with their hypothesized signs
and are statistically signiﬁcant. These results conﬁrm that the persistence of
P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34 19
Table 3
Autoregressive properties of v
R
, the other information embedded in analysts’ forecasts of next
period’s abnormal earnings
Pooled analysis with one lag
v
R>¹
"
"
#
¹
v
R
#
`R>¹
"
¹
R`
0.01 0.32 0.08
(38.79) (57.94)
Notes: Sample consists of 50,133 annual observations from 1976 to 1995. All variables are scaled by
the market value of equity at the end of year t. Figures in parentheses are tstatistics.
The other information variable is deﬁned as
v
R
"f
R
!x
R
where the period t consensus analyst forecast of abnormal earnings for the next period is deﬁned as
f
R
"f
R
!r.b
R
and abnormal earnings for period t is deﬁned as
x
R
"x
R
!r.b
R¹
f
R
denotes the I/B/E/S consensus forecast of earnings for year t#1 measured in the ﬁrst month
following the announcement of earnings for year t;
is the ﬁrst order autoregression coeﬃcient for abnormal earnings and is estimated using all
historically available data from 1950 through year t in a pooled timeseries crosssectional regres
sion;
x
R
denotes earnings before extraordinary items for year t;
r denotes the discount rate (assumed to be 12%);
b
R
denotes book value of equity at the end of year t.
abnormal earnings varies in a systematic and predictable manner. Conse
quently, the conditional estimates of that we use to implement Ohlson’s
valuation model should oﬀer additional predictive ability with respect to future
abnormal earnings.
Finally, Table 3 examines the autoregressive properties of the other informa
tion variable, v. The estimate of the ﬁrstorder autoregressive coeﬃcient on the
other information,
¹
, is 0.32 with a tstatistic of 57.94. Thus, the other
information mean reverts at about twice the rate of abnormal earnings. How
ever,
¹
, is also signiﬁcantly diﬀerent from the polar extremes of 0 and 1 that are
implicitly assumed in many of the valuation models used in past research. Thus,
we expect that incorporating more precise estimates of this coeﬃcient should
improve our ability to forecast future abnormal earnings and hence predict
contemporaneous stock prices.
20 P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34
Table 4
Relative forecasting ability of alternative modes for predicting next year’s abnormal earnings
Panel A: Predictions for models ignoring ‘other information’, computed as
E
R
[x
R>¹
]"x
R
Mean forecast
error
Mean absolute
forecast error
Mean square
forecast error
"0 !0.029 0.087 0.033
"1 0.006 0.081 0.032
" !0.008 0.077 0.030
" !0.006 0.076 0.028
Panel B: Prediction for models incorporating ‘other information’, computed as
E
R
[x
R>¹
]"f
R
!0.032 0.052 0.015
Notes: Sample consists of 50,133 observation from 1976 to 1995. Forecast errors are scaled by the
market value of equity at the end of year t.
The forecast error for year t is computed by subtracting the forecast of abnormal earning for year
t#1 from the realized abnormal earnings for year t#1.
Abnormal earnings for year t is deﬁned as
x
R
"x
R
!r.b
R
and the period t consensus analysts’ forecast of abnormal earnings for period t#1 is deﬁned as
f
R
"f
R
!r.b
R
where
x
R
denotes earnings before extraordinary items and discontinued operations for year t;
r denotes the discount rate (assumed to be 12%);
b
R
denotes book value of equity at the end of year t;
f
R
denotes the I/B/E/S consensus forecast of earnings for year t#1 measured in the ﬁrst month
following the announcement of earnings for year t;
is the ﬁrst order autoregression coeﬃcient for abnormal earnings and is estimated using all
historically available data from 1950 through the forecast year in a pooled timeseries crosssectional
regression;
is the predicted value of from the regression model speciﬁed in Table 2 and estimated using all
historically available data from 1950 through the forecast year.
4.2. Prediction of next period abnormal earnings
Statistics on the bias and accuracy of the predictions of next period abnormal
earnings generated by each of the valuation models are reported in Table 4. The
mean forecast error measures forecast bias, while the mean absolute forecast
error and the mean square forecast error measure forecast accuracy. All forecast
errors are deﬂated by market value and the extreme 1%of the forecast errors are
P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34 21
winsorised. Panel A reports forecast errors for each of the models that ignore the
other information variable, v, while panel B reports forecast errors for the
models that incorporate v. Recall that the forecast of next period abnormal
earnings is equal to the consensus analyst estimate of abnormal earnings for all
of the models that incorporate v. Hence, we only report one set of forecast errors
for these models. We measure the analysts’ earnings estimates using the I/B/E/S
mean consensus earnings estimates provided in the month immediately follow
ing the announcement of the annual earnings data used in the timeseries
models. This ensures that all of the forecasting variables are measured at similar
points in time.
The mean forecast error is close to zero for the models using "1, "
and ", and is slightly negative for the model using "0 (!0.029). This
latter result indicates that, on average, ﬁrms fell slightly short of achieving
a return on equity equal to the assumed cost of capital of 12%. The mean
forecast error is also negative using the consensus analyst forecast, reﬂecting
overoptimism in analysts’ forecasts. The measures of forecast accuracy indicate
that the predictive abilities of the models ignoring the other information in
analysts’ forecasts are all very close. The model using has only slightly
smaller forecast errors than the model using , indicating that our eﬀorts to
conditionally estimate the persistence parameter add relatively little to the
forecasting ability of the model. The model using "1 is slightly less accurate
than the two versions using estimates of , and the model using "0 is the
least accurate of all. The results in panel B indicate that analysts’ forecasts of
abnormal earnings are much more accurate than the forecasts generated by the
historical timeseries models. This result highlights the important role of the
other information embedded in analysts’ forecasts in predicting future abnormal
earnings.
4.3. Explanation of contemporaneous stock prices
The relative ability of the competing valuation models to explain contempor
aneous stock prices is evaluated in Table 5. Panel A of Table 5 reports results
for the four models ignoring the other information. All of these models generate
large positive mean forecast errors, indicating that they undervalue equities
relative to the stock market. The undervaluation is smallest for the model using
"0 (forecast error"0.291) and greatest for the model using "1 (forecast
error"0.378). The measures of forecast accuracy are similar for the models
using "0, "
S
and ", respectively. However, the model using "1
is considerably less accurate than the other three models. To understand this
result, recall from Fig. 2 that the model using "1 model generates poor
forecasts of longrun abnormal earnings. Since expectations of longrun abnor
mal earnings are included in the computation of stock price, this model therefore
generates relatively poor forecasts of stock price. The mediocre showing of the
22 P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34
Table 5
Relative forecasting ability of alternative modles for explaining contemporaneous stock prices
Panel A: Price estimates for models ignoring ‘other information’, computed as
P
R
"b
R
#
1#r!
x
R
Mean forecast
error
Mean absolute
forecast error
Mean square
forecast error
"0 0.291 0.461 0.284
"1 0.378 0.519 0.363
" 0.320 0.461 0.284
" 0.326 0.465 0.291
Panel B: Price estimates for models incorporating ‘other information’, computed as
P
R
"b
R
#
1#r!
x
R
#
1#r
(1#r!)(1#r!)
v
R
("0, "0) 0.285 0.445 0.266
("1, "0) and ("0, "1) 0.227 0.402 0.232
(", "0) and ("0, "S) 0.278 0.427 0.248
(", "S) 0.259 0.419 0.241
Notes: Sample consists of 50,133 observations from 1976 to 1995. Forecast errors are scaled by stock
price at the end of year t
The forecast error for year t is computed by subtracting the forecast stock price for year t from the
observed market stock price at the end of the month following the announcement of earnings for
year t.
Abnormal earnings for year t is deﬁned as
x
R
"x
R
!r.b
R
where x
R
denotes earnings before extraordinary items and discontinued operations for year t,
r denotes the discount rate (assumed to be 12%), and b
R
denotes book value of equity at the end of
year t;
is the ﬁrst order autoregression coeﬃcient for abnormal earnings and is estimated using all
historically available data from 1950 through the forecast year in a pooled timeseries crosssectional
regression;
is the predicted value of from the regression model speciﬁed in Table 2 and estimated using all
historically available data from 1950 through the forecast year;
S is the ﬁrst order autoregression coeﬃcient for the other information variable, v
R
, and is estimated
using all historically available data from1950 through the forecast year in a pooled timeseries cross
sectional regression.
v
R
is deﬁned as
v
R
"f
R
!x
R
where the period t consensus analyst forecast of abnormal earnings for the next period is deﬁned as
f
R
"f
R
!r.b
R
f
R
denotes the I/B/E/S consensus forecast of earnings for year t#1 measured in the ﬁrst month
following the announcement of earnings for year t.
P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34 23
model using " is somewhat surprising. Table 4 illustrates that this model
generates the most accurate forecasts of next period’s abnormal earnings among
the four models ignoring the other information. Thus, the poor showing of this
model in the pricing tests raises the possibility that stock prices do not reﬂect
rational expectations of future abnormal earnings. We explore this issue in more
detail later in the paper.
Panel B of Table 5 reports results for the models incorporating the informa
tion in analysts’ forecasts. The mean forecast errors indicate that these models
also undervalue relative to the market. However, the undervaluations are not as
large as they were for the models ignoring other information. The undervalu
ations are surprising, because the results in Table 3 indicate that the analysts’
forecasts of future abnormal earnings are overoptimistic. All of the models
incorporating the other information have lower forecast errors than the models
using historical data. These results are consistent with the superior predictive
accuracy of analysts’ forecasts with respect to future abnormal earnings. Of the
models incorporating other information, the model using ("1, "0) pro
vides the most accurate forecasts of stock prices. Recall from Fig. 1 that this
model simply capitalizes the forecast of next period’s earnings in perpetuity and
ignores information in book value. This result is surprising, because book value
contains additional information about longrun abnormal earnings that should
be rationally reﬂected in stock prices. Thus, the strong showing of the model
using ("1, "0) in the pricing tests again raises the possibility that stock
prices do not reﬂect rational expectations of future abnormal earnings.
In Table 6, we investigate the ability of the information variables used in the
valuation models to explain stock prices without imposing the restrictions
implied by the valuation models. Panel A of Table 6 reports results of annual
crosssectional regressions of stock price on historical book value and earnings.
These two explanatory variables are the information variables used in the
valuation models that ignore other information. Both book value and earnings
load positively and signiﬁcantly in the regressions. The fact that book value
loads in addition to earnings indicates that book value contains value relevant
information beyond that already in earnings. We can obtain further insights
from the regressions by comparing the estimated coeﬃcients to values implied
by Ohlson’s model in conjunction with representative parameter values. The
formulae for the predicted valuation coeﬃcients on book value and earnings are
taken from Ohlson (1995), (p. 670). Using r"12% (longrun historical average)
and "0.62 (historical average from (Table 1) gives:
¹
"1!r./(1#r!)"0.85; and
`
"(#.r)/(1#r!)"1.39.
The corresponding mean values (standard errors) on the empirical regression
coeﬃcients are
¹
"0.40 (0.074) and
`
"3.88 (0.262). Thus, stock prices appear
24 P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34
Table 6
Unconstrained regressions of stock price on the variables used in the valuation models
Panel A: Regressions of price on book value and earnings
P
R
"#
¹
b
R
#
`
x
R
#
R
Coeﬀ. Mean Std. err. Min. Q1 Med. Q3 Max
9.72 0.408 7.65 8.07 9.57 10.92 13.63
¹
0.40 0.074 !0.18 0.05 0.51 0.68 0.81
`
3.88 0.262 2.43 3.07 3.68 4.74 6.27
R` 0.40 0.015 0.40 0.51 0.53 0.59 0.67
1Panel B: Regressions of price on book value, earnings and the consensus analyst forecast of next year’s
earnings
P
R
"#
¹
b
R
#
`
x
R
#
`
f
R
#
R
Coeﬀ. Mean Std. Err. Min. Q1 Med. Q3 Max
4.25 0.353 1.64 3.00 4.53 5.09 7.05
¹
0.24 0.035 !0.06 0.09 0.26 0.39 0.42
`
0.05 0.150 !0.82 !0.53 0.03 0.56 1.34
`
5.79 0.256 3.97 4.85 5.89 6.64 8.07
R` 0.69 0.019 0.56 0.61 0.68 0.74 0.86
Notes: Statistics reported are based on the estimates from 20 annual crosssectional regressions from
1976 to 1995. Sample consists of 50,133 observations from 1976 to 1995. All variables are measured
on a pershare basis.
P
R
denotes the stock price measured at the end of the month following the announcement of earnings
for year t.
x
R
denotes earnings before extraordinary items and discontinued operations for year t.
b
R
denotes book value of equity at the end of year t.
f
R
denotes the I/B/E/S consensus forecast of earnings for year t#1 measured in the ﬁrst month
following the announcement of earnings for year t.
to place too low a weight on book value and too high a weight on earnings. The
value of required to justify the empirical regression coeﬃcients is approxim
ately "0.85. One interpretation of these results is that they are driven by
a misspeciﬁcation in Ohlson’s valuation model. An alternative interpretation is
that stock prices do not reﬂect rational expectations, because investors overesti
mate the persistence of abnormal earnings.
The regressions reported in panel B of Table 6 employ the information
variables used in the valuation models incorporating other information. In
addition to book value and earnings, these regressions also include the consen
sus analyst forecast of next period’s earnings. The explanatory power of these
P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34 25
regressions are considerably higher than in panel A, indicating that the analysts’
forecasts contain incremental information about ﬁrm value. Book value loads
positively and signiﬁcantly, though the coeﬃcient is much smaller than in the
regressions excluding the analyst forecast variable. This result indicates that
book value contains some value relevant information beyond that in analysts’
forecasts of next year’s earnings. Earnings loads with a small and statistically
insigniﬁcant coeﬃcient, suggesting that analysts’ forecast of next year’s earnings
subsume value relevant information in current earnings. Finally, the analysts’
forecast of next year’s earnings loads with a positive and statistically signiﬁcant
coeﬃcient.
We can again obtain further insights from the regressions by comparing the
estimated coeﬃcients to values implied by Ohlson’s model in conjunction with
representative parameter values. The formulae for the predicted valuation
coeﬃcients on book value and earnings are from Ohlson (1998) (p. 14). Using
r"12% (longrun historical average) and "0.62 (historical average from
Table 1) and "0.32 (historical average from Table 3) gives:
¹
"[(1#r)(1!)(1!)]/[(1#r!)(1#r!)]"0.72,
`
"[!(1#r)..]/[(1#r!)(1#r!)]"!0.55, and
`
"(1#r)/[(1#r!)(1#r!)]"2.80.
The corresponding mean values (standard errors) on the empirical regression
coeﬃcients are
¹
"0.24 (0.035),
`
"0.05 (0.150) and
`
"5.79 (0.256). Thus,
stock prices place too low a weight on book value and too high a weight on the
analysts’ forecast of next year’s earnings. For example, a (,) combination of
approximately (0.95,0.00) would be required to approximate the empirical
regression coeﬃcients. One interpretation of these results is that they are driven
by a misspeciﬁcation in Ohlson’s valuation model. An alternative interpretation
is that stock prices do not reﬂect rational expectations, because investors tend to
overestimate the persistence of shorttermearnings forecasts. We investigate this
possibility in the next section.
4.4. Prediction of future stock returns
Thus far, our pricing tests have focused on the ability of the competing
valuation models to predict contemporaneous stock prices. In this section, we
consider whether the values implied by the competing models are able to predict
future stock returns. These additional tests are motivated by the apparent
inconsistencies between the abnormal earnings prediction results in Table 4 and
the valuation results in Tables 5 and 6. In particular, the results in Table 4
26 P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34
Table 7
Predictive ability of ratios of implied model values to observed market values with respect to stock
returns over the following year
Panel A: Implied values ignoring ‘other information’, computed as
P
R
"b
R
#
1#r!
x
R
Portfolio "0 "1 " "
1 (Lowest) 0.143 0.159 0.140 0.136
2 0.171 0.143 0.174 0.159
3 0.153 0.161 0.152 0.165
4 0.169 0.158 0.162 0.159
5 0.181 0.160 0.170 0.173
6 0.170 0.166 0.181 0.175
7 0.191 0.182 0.180 0.187
8 0.196 0.202 0.197 0.194
9 0.206 0.222 0.203 0.212
10 (Highest) 0.215 0.235 0.234 0.235
Hedge 0.072 0.076 0.094 0.099
(tstatistic) (1.94) (2.24) (2.39) (2.44)
indicate that the model using " provides more accurate forecasts of future
abnormal earnings than the models using " and "0. However, the
results in Table 5 indicate that the reverse holds true with respect to the ability
of the models to explain observed stock prices. Moreover, the evidence in
Table 6 is consistent with the expectations embedded in stock prices under
estimating the mean reversion in abnormal earnings. In the tradition of funda
mental analysis, we therefore provide tests of whether observed stock prices tend
to revert toward the ‘fundamental’ or ‘intrinsic’ values implied by particular
models. These tests entertain the possibility of temporary stock mispricing that
can be systematically predicted by particular valuation models. The tests are
constructed by taking the ratio of the intrinsic model values to observed equity
values. Decile portfolios are then formed using the ranked ratios. Lower deciles
consist of stocks that are overpriced relative to intrinsic value, and are therefore
expected to experience lower future stock returns. Higher deciles consist of
stocks that are underpriced relative to intrinsic value, and are therefore expected
to experience higher future stock returns. Note that the ratio formed for the
model using "0 is just the booktomarket ratio, while the ratio formed for
the model using "1 is proportional to the earningstoprice ratio. The
predictive ability of these ratios with respect to future stock returns is already
well documented.
The results are presented in Table 7. Panel A reports the oneyearahead
returns for ratios formed on the valuation models ignoring other information.
P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34 27
Table 7 (continued)
Panel B: Implied values incorporating ‘other information’, computed as
P
R
"b
R
#
1#r!
x
R
#
1#r
(1#r!)(1#r!)
v
R
("1, "0) (", "0)
("0, "0) and and (", "S)
Portfolio ("0, "1) ("0, "S)
1 (Lowest) 0.149 0.157 0.154 0.162
2 0.176 0.145 0.165 0.159
3 0.147 0.154 0.154 0.154
4 0.162 0.177 0.161 0.158
5 0.178 0.179 0.174 0.171
6 0.175 0.173 0.175 0.175
7 0.178 0.181 0.173 0.185
8 0.211 0.210 0.213 0.203
9 0.201 0.208 0.206 0.204
10 (Highest) 0.220 0.210 0.224 0.224
Hedge 0.071 0.054 0.070 0.062
(tstatistic) (1.77) (1.44) (1.71) (1.34)
Notes: Each year, observations are ranked and assigned in equal numbers to deciles based on the ratio
of implied model value to observed market value of equity. Equalweighted buyhold stock returns are
then computed for each decile portfolio over the subsequent 12 months, beginning three months after
the end of the ﬁscal year from which the historical forecast data are obtained. The table reports the
mean of the 20 years of annual portfolio returns. ¹statistics are based on the timeseries standard
errors of the 20 annual portfolio returns. Sample consists of 50,133 observations from 1976 to 1995.
Abnormal earnings for year t is deﬁned as
x
R
"x
R
!r.b
R
,
where x
R
denotes earnings before extraordinary items and discontinued operations for year t,
r denotes the discount rate (assumed to be 12%), and b
R
denotes book value of equity at the end of
year t;
is the ﬁrst order autoregression coeﬃcient for abnormal earnings and is estimated using all
historically available data from 1950 through the forecast year in a pooled timeseries crosssectional
regression;
is the predicted value of from the regression model speciﬁed in Table 2 and estimated using all
historically available data from 1950 through the forecast year;
S is the ﬁrst order autoregression coeﬃcient for the other information variable, v
R
, and is estimated
using all historically available data from1950 through the forecast year in a pooled timeseries cross
sectional regression.
v
R
is deﬁned as
v
R
"f
R
!x
R
,
where the period t consensus analyst forecast of abnormal earnings for the next period is deﬁned as
f
R
"f
R
!r.b
R
f
R
denotes the I/B/E/S consensus forecast of earnings for year t#1 measured in the ﬁrst month
following the announcement of earnings for year t.
28 P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34
The hedge portfolio return, which is the diﬀerence between the return
for portfolio 10 and the return for portfolio 1, summarizes the predictive
ability of each model with respect to future returns. The return interval
begins 3 months after the ﬁscal year end of the year from which the historical
data is obtained. Statistical inference is conducted using the standard error
of the annual mean hedge portfolio returns over the 20 years in the sample
period. The model using " displays the greatest predictive ability,
with a hedge portfolio return of 9.9% (t"2.44). The model using "
is second, with a hedge portfolio return of 9.4% (t"2.39), while the model
using "1 is third with a hedge portfolio return of 7.6% (t"2.24). The model
using "0 displays the lowest predictive ability, with a hedge portfolio return
of 7.2% (t"1.94). The superior predictive ability with respect to future stock
returns of the model using " potentially explains why this model performs
poorly in the pricing tests (Table 5), despite its superior predictive ability with
respect to future abnormal earnings (Table 4). It appears that the expectations
reﬂected in stock prices fail to fully anticipate the rate of mean reversion in
abnormal earnings that is captured by this model. However, this explanation
should be interpreted with caution due to the low statistical signiﬁcance of the
results.
Panel B reports the oneyearahead returns for ratios computed using valu
ation models incorporating the other information in analysts’ earnings forecasts.
The hedge portfolio returns are uniformly lower, ranging from 7.1% (t"1.77)
for the model using ("0,"0) to 5.4% (t"1.44) for the model using
("1,"0). These results contrast sharply with the contemporaneous stock
price results in Table 5. While valuation models incorporating information in
analysts’ forecasts have the greatest ability to explain contemporaneous stock
prices, valuation models ignoring this information have the greatest predictive
ability with respect to future stock returns. Moreover, the valuation model using
("1,"0) is the best at explaining contemporaneous stock prices, but the
worst at predicting future stock returns. These relations are exactly what would
be expected if analysts’ earnings estimates are naively incorporated in stock
prices even when they do not fully reﬂect all information in current abnormal
earnings about future abnormal earnings. However, the results in Table 7 are
indirect and their statistical signiﬁcance is weak. Moreover, Kothari and
Warner (1997) and Barber and Lyon (1997) provide evidence that statistical tests
using long horizon stock returns are poorly speciﬁed. Table 8 therefore reports
results of more direct tests of the hypothesis that investors price predictable
errors in analysts’ forecasts.
The regressions in panel A of Table 8 examine the extent to which each of the
models ignoring other information in analysts’ forecasts detects errors in ana
lysts’ forecasts of oneperiodahead earnings. The regressions in panel B then
examine whether the errors identiﬁed in the analysts’ forecasts appear to be
accompanied by corresponding errors in stock prices. The results in panel
P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34 29
Table 8
Panel A
Analysis of the relation between forecast errors in abnormal earnings predictions from analysts’
consensus earnings estimates and forecasts of abnormal earnings that ignore the other information
in analysts’ consensus earnings estimates.
The earnings forecasts that ignore the other information are generated by the model.
x
GR>¹
"x
GR
#
GR>¹
. Statistics reported are based on the estimates from 20 annual crosssectional
regressions from 1976 to 1995. Sample consists of 50,133 observations from 1976 to 1995. All
variables are measured on a pershare basis.
Regression model estimated is
(x
R>¹
!f
R
)"
"
#
¹
(x
R
!f
R
)#e
R>¹
Coeﬀ. Mean Std. error Min. Q1 Med. Q3 Max
"0
"
!0.03 0.005 !0.08 !0.04 !0.03 !0.03 0.00
¹
!0.13 0.083 !0.63 !0.40 !0.16 0.06 0.71
R` 0.05 0.018 0.00 0.01 0.01 0.08 0.26
"1
"
!0.02 0.003 !0.04 !0.03 !0.02 !0.01 0.00
¹
0.40 0.036 0.13 0.29 0.42 0.45 0.64
R` 0.11 0.017 0.01 0.04 0.12 0.16 0.24
"
"
!0.02 0.004 !0.05 !0.04 !0.03 !0.01 0.00
¹
0.42 0.043 0.14 0.25 0.44 0.51 0.74
R` 0.08 0.024 0.01 0.03 0.07 0.12 0.40
"
"
!0.02 0.004 !0.04 !0.03 !0.02 !0.01 0.00
¹
0.48 0.043 0.16 0.35 0.49 0.58 0.76
R` 0.12 0.022 0.01 0.05 0.10 0.16 0.35
A indicate that the models using "1, " and " all identify
systematic errors in analysts’ earnings forecasts. The results in panel B indicate
that these systematic forecast errors are reﬂected in stock prices, though the
statistical signiﬁcance of these results is weak. Frankel and Lee (1998) also
report that analysts’ earnings forecasts contain predictable errors that are not
rationally anticipated in stock prices. Thus the hypothesis that investors naively
price predictable errors in analysts’ forecasts provides a promising explanation
for the results obtained in this paper.
30 P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34
Table 8 (continued)
Panel B
Analysis of the relation between stock returns in the year following the release of analysts’ consensus
forecasts and forecasts of abnormal earnings that ignore the other information in analysts’ consensus
earnings forecasts.
The earnings forecasts that ignore the other information are generated by the model
x
GR>¹
"x
GR
#
GR>¹
. Stock returns are equalweighted buy—hold returns over the 12 months
beginning three months after the end of the ﬁscal year from which the historical forecast data are
obtained. Statistics reported are based on the estimates from 20 annual crosssectional regressions
from 1976 to 1995. Sample consists of 50,133 observations from 1976 to 1995. All variables are
measured on a pershare basis.
Regression model estimated is
Ret
R>¹
"
"
#
¹
(x
R
!f
R
)#e
R>¹
Coeﬀ. Mean Std. error Min. Q1 Med. Q3 Max
"0
"
0.18 0.037 !0.06 0.11 0.14 0.27 0.56
¹
!0.03 0.096 !0.73 !0.36 !0.08 0.28 0.80
R` 0.01 0.001 0.00 0.00 0.00 0.01 0.02
"1
"
0.18 0.035 !0.05 0.11 0.15 0.27 0.52
¹
0.07 0.066 !0.05 !0.15 0.01 0.28 0.65
R` 0.01 0.002 0.00 0.00 0.00 0.01 0.04
"
"
0.18 0.036 !0.05 0.11 0.15 0.27 0.53
¹
0.10 0.077 !0.44 !0.15 0.01 0.36 0.76
R` 0.01 0.002 0.00 0.00 0.00 0.01 0.03
"
"
0.18 0.036 !0.05 0.11 0.16 0.27 0.53
¹
0.14 0.070 !0.34 !0.15 0.05 0.46 0.88
R` 0.01 0.002 0.00 0.00 0.01 0.01 0.04
Notes: Abnormal earnings for year t is deﬁned as
x
R
"x
R
!r.b
R
where x
R
denotes earnings before extraordinary items and discontinued operations for year t, r denotes
the discount rate (assumed to be 12%), and b
R
denotes book value of equity at the end of year t.
The consensus analyst forecast of abnormal earnings for the next period is deﬁned as
f
R
"f
R
!r.b
R
where f
R
denotes the I/B/E/S consensus forecast of earnings for year t#1 measured in the ﬁrst
month following the announcement of earnings for year t.
Ret
R>¹
is the equalweighted, buy—hold, withdividend stock return over the 12 months beginning
three months after the end of the ﬁscal year t.
is the ﬁrst order autoregression coeﬃcient for abnormal earnings and is estimated using all
historically available data from 1950 through the forecast year in a pooled timeseries crosssectional
regression.
is the predicted value of from the regression model speciﬁed in Table 2 and estimated using all
historically available data from 1950 through the forecast year.
P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34 31
" We emphasize the word ‘directly’ in this sentence. Forecasts of the earnings and book
value components of abnormal earnings contain a forecast of future dividend payments through
the clean surplus relation. Thus, the researcher must focus directly on forecasting future
abnormal earnings, rather than on forecasting its components. This simpliﬁcation embodies the
notion that dividend policy is irrelevant to the extent that reinvested earnings generate the cost of
capital.
5. Conclusions
This paper provides an empirical assessment of the residual income valuation
model proposed in Ohlson (1995). We begin by pointing out that existing
empirical applications of the residual income valuation model are generally
similar to past applications of traditional earnings capitalization models. We
argue that the key original empirical implications of Ohlson’s model arise from
the information dynamics that describe the formation of abnormal earnings
expectations. Our empirical tests indicate that while the information dynamics
are reasonably empirically descriptive, a simple valuation model that capitalizes
analysts’ earnings forecasts in perpetuity is better at explaining contempor
aneous stock prices. Subsequent tests suggest that the superior explanatory
power of the simple capitalization model may arise because investors over
weight information in analysts’ earnings forecasts and underweight information
in current earnings and book value.
Despite the ambiguous empirical support for the model, we believe that the
model provides a useful framework for empirical research for several reasons.
First, as shown in this paper, the model provides a unifying framework for
a large number of previous ‘ad hoc’ valuation models using book value, earnings
and shortterm forecasts of earnings. In doing so, the model highlights the
implicit assumptions that previous models make about the relation between
current accounting variables and future abnormal earnings. Second, the model
provides a basic framework upon which subsequent research can build. For
example, Feltham and Ohlson (1995) generalize the model to incorporate
growth and accounting conservatism. Third, the focus of the model on the
relation between current information variables and future abnormal earnings is
heuristically appealing. Previous valuation models based on the dividend
discounting model often make unrealistic assumptions about dividend policy.
For example, Kothari and Zimmerman (1995) assume a 100% payout ratio.
Ohlson’s model illustrates that valuation models focusing directly on forecasting
future abnormal earnings avoid having to forecast the timing of future dividend
payments."
32 P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34
References
Ahmed, S.A., 1994. Accounting earnings and future economic rents: an empirical analysis. Journal of
Accounting and Economics 17, 377—400.
Anthony, J.H., Ramesh, K., 1992. Association between accounting performance measures and stock
prices. Journal of Accounting and Economics 15, 203—228.
Bernard, V., 1995. The Feltham—Ohlson framework: Implications for empiricists. Contemporary
Accounting Research 11, 733—747.
BarYosef, S., Callen, J.L., Livnat, J., 1996. Modeling dividends, earnings and the book value of
equity: an empirical investigation of the Ohlson valuation dynamics. Review of Accounting
Studies 1, 207—224.
Barber, B.M., Lyon, J.D., 1997. Detecting abnormal longrun stock returns: The empirical power
and speciﬁcation of tests statistics. Journal of Financial Economics 43, 341—372.
Barth, M.E., 1991. Relative measurement errors among alternative pension asset and liability
measures. Accounting Review 66, 433—463.
Beaver, W., Lambert, R., Morse, D., 1980. The information content of security prices. Journal of
Accounting and Economics 2, 3—28.
Brooks, L.D., Buckmaster, D.A., 1976. Further evidence of the timeseries properties of accounting
income. Journal of Finance 31, 1359—1373.
Easton, P.D., Harris, T.S., 1991. Earnings as an explanatory variable for returns. Journal of
Accounting Research 29, 19—36.
Fairﬁeld, P.M., Sweeney, R.J., Yohn, T.L., 1996. Accounting classiﬁcation and the predictive content
of earnings. Accounting Review 71, 337—355.
Fazzari, S., Hubbard R.G., Peterson, B., 1988. ﬁnancing constraints and corporate investment.
Brookings Papers on Economic Activity, pp. 141195.
Feltham, G.A., Ohlson, J.A., 1995. Valuation and clean surplus accounting for operating and
ﬁnancial activities. Contemporary Accounting Research 11, 689—732.
Francis, J., Olsson P., Oswald, D., 1997. Comparing the accuracy and explainability of dividend, free
cash ﬂow and abnormal earnings equity valuation models. Working Paper, University of
Chicago.
Frankel R., Lee, C.M.C., 1998. Accounting valuation, market expectation, and crosssectional stock
returns. Journal of Accounting and Economics, (forthcoming).
Freeman, R.N., Ohlson, J.A., Penman, S.H., 1982. Book rate of return and the prediction of earnings
changes. Journal of Accounting Research 20, 639—653.
Kothari, S.P., 1992. Priceearnings regressions in the presence of prices leading earnings: earnings
level versus earnings change speciﬁcations and alternative deﬂators? Journal of Accounting and
Economics 15, 173—202.
Kothari, S.P., Warner, J.B., 1997. Measuring longhorizon security performance. Journal of Finan
cial Economics 43, 301—339.
Kothari, S.P., Zimmerman, J., 1995. Price and return models. Journal of Accounting and Economics
20, 155—192.
Lee, C.M., Myers C.J., Swaminathan, B., 1998. What is the intrinsic value of the Dow? Working
Paper, Cornell University.
Lundholm, R., 1995. A tutorial on the Ohlson and Feltham/Ohlson models. Contemporary Ac
counting Research 11, 749—761.
Malkiel, B., Cragg, J., 1970. Expectations and the structure of share prices. American Economic
Review 60, 601—617.
Ohlson, J.A., 1991. The theory of value and earnings, and an introduction to the Ball and Brown
Analysis. Contemporary Accounting Research 8, 1—19.
Ohlson, J.A., 1995. Earnings, book values and dividends in security valuation. Contemporary
Accounting Research 11, 661—687.
P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34 33
Ohlson, J.A., 1998. Earnings, book values and dividends in equity valuation: An empirical perspect
ive. Working paper, Columbia University.
Palepu, K.G., Bernard V.L., Healy, P.M., 1996. Business analysis and valuation using ﬁnancial
statements. SouthWestern College Publishing Co., Cincinnati, Ohio.
Penman, S.H., 1997. A synthesis of equity valuation techniques and the terminal value calculation
for the dividend discounting model, Working Paper, University of California, Berkeley.
Penman, S.H., Sougiannis, T., 1996. A comparison of dividend, cash ﬂow and earnings approaches
to equity valuation. Working Paper, University of California, Berkeley.
Scherer, F.M., 1980. Industrial Market Structure and Firm Performance. Rand McNally College
Publishing Co., Chicago, IL.
Sloan, R.G., 1996. Do stock prices fully reﬂect information in accruals and cash ﬂows about future
earnings?. Accounting Review 71, 289—315.
Vander Weider, J.H., Carleton, W.T., 1988. Investor growth expectations: Analysts vs. history.
Journal of Portfolio Management 14, 78—82.
Whitbeck, V., Kisor, M., 1963. A new tool in investment decision making. Financial Analysts
Journal 19, 55—62.
34 P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34
2
P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34
1. Introduction A recent paper by Ohlson (1995) has stimulated interest in the residual income formulation of the dividend discounting valuation model. This development has potentially important implications for empirical researchers, as Ohlson’s model speciﬁes the relation between equity values and accounting variables such as earnings and book value. Existing empirical research has generally provided enthusiastic support for the model, and the model is now proposed as an alternative to the discounted cash ﬂow model in equity valuation. Existing empirical research argues that the model breaks new ground on two fronts. First, the model predicts and explains stock prices better than the models based on discounting shortterm forecasts of dividends and cash ﬂows (Bernard, 1995; Penman and Sougiannis, 1996; Francis et al., 1997). Second, the model provides a more complete valuation approach than popular alternatives (Frankel and Lee, 1998). In this paper, we evaluate the empirical implications of Ohlson’s model. Central to our analysis is the incorporation of the residual income information dynamics in Ohlson (1995). Past empirical applications of the residual income valuation model ignore Ohlson’s information dynamics. In many cases, the resulting valuation models are similar to past applications of the dividenddiscounting model that capitalize current or forecasted earnings, but make no appeal to book value or residual income (e.g., Whitbeck and Kisor, 1963; Malkiel and Cragg, 1970; Kothari and Zimmerman, 1995). Consistent with Ohlson’s information dynamics, we ﬁnd that residual income follows a mean reverting process. In addition, we show that the rate of mean reversion is systematically associated with ﬁrm characteristics suggested by accounting and economic analysis. The rate of mean reversion is decreasing in the quality of earnings, increasing in the dividend payout ratio and correlated across ﬁrms in the same industry. We also ﬁnd that incorporating information in analysts’ forecasts of earnings into the information dynamics increases forecast accuracy. This result highlights the importance of information other than current residual income in forecasting future residual income. Our pricing tests indicate that stock prices partially reﬂect the mean reversion in residual income. An important implication of this result is that book value conveys additional information over earnings in explaining contemporaneous stock prices. However, we also ﬁnd that book value provides very little additional information about stock prices beyond that contained in analysts’ forecasts of next year’s earnings. This result is somewhat surprising, because analysts’ forecasts of next year’s earnings do not fully capture the longterm mean reversion in residual income. Further tests help reconcile these seemingly
See Palepu et al. (1996) for a discussion of the application of the model to equity valuation.
P.M. Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34
3
contradictory results by suggesting that observed stock prices seem to display a lagged response to the longterm mean reversion in residual income. We conclude that Ohlson’s formulation of the residual income valuation model provides a parsimonious framework for incorporating information in earnings, book value and earnings forecasts in empirical research. We illustrate how many of the valuation relations implicit in past empirical research can be considered as special cases of Ohlson’s model. However, we also ﬁnd that past earnings and book value convey relatively little information about ﬁrm value beyond that reﬂected in analysts’ forecasts of next year’s earnings. Thus, while the model provides a unifying framework for earningsbased valuation research, our eﬀorts at implementing the model provide only modest improvements in explanatory power over past empirical research using analysts’ earnings forecasts in conjunction with the traditional dividenddiscounting model. Nevertheless, an important shortcoming of past research is that the relation between earnings forecasts and future dividends has been speciﬁed in an ad hoc fashion. By formalizing the information dynamics, Ohlson’s model provides a guiding framework for future valuation research. The remainder of the paper is organized as follows. Section 2 reviews Ohlson’s formulation of the residual income valuation model and identiﬁes the model’s empirical implications. Section 3 describes our research design and variable measurement. Section 4 presents the empirical results and Section 5 concludes.
2. Model development This section provides an empirically oriented review of the residual income valuation model developed in Ohlson (1995). Our review emphasizes that the model is a restated and restricted version of the standard dividenddiscounting model. Empirical applications of the model that ignore Ohlson’s restrictions on the timeseries properties of residual income are diﬃcult to distinguish from empirical applications based on the standard dividend discounting model. We illustrate this point with reference to existing empirical research employing the residual income valuation model. We complete the section by outlining the key issues in the empirical implementation of Ohlson’s valuation model. 2.1. Model review The model is comprised of three basic assumptions. First, price is equal to the present value of expected dividends: E [d ] R R>O , P" R (1#r)O O (1)
M. Another way of illustrating the independence of Eq. its usefulness must stem from properties in addition to the clean surplus assumption. (1) yields: E [b #x !b ] R R>O\ R>O R>O . For example. the valuation R>O relation in Eq. and ‘residual income’ or ‘abnormal earnings’ is deﬁned as x "x !r.b R R R\ so that price can be expressed as the sum of book value and the present value of future abnormal earnings: E [x ] R R>O . (4) is assumed to be zero. if accrual accounting is incrementally useful over cash accounting in the valuation process. given a stream of future dividends. . (2) R R\ R R where b is the book value of equity at time t. The resulting variables clearly satisfy the clean surplus relation embodied in Eq. (5) will yield the present value of the dividend stream. (3) to be rewritten as (3) E [x !r. Dechow et al. P "b # (5) R R (1#r)O O Eq. P" R (1#r)O O Simple algebraic manipulation allows Eq. and so the resulting ‘residual cash ﬂow valuation model’ is also a legitimate reformulation of the dividend discounting formula. This assumption allows future dividends to be expressed in terms of future earnings and book values. It is important to note that Eq. Second. / Journal of Accounting and Economics 26 (1999) 1–34 where P is the price of the ﬁrm’s equity at time t.4 P. r is the (assumed constant) discount rate. the clean surplus accounting relation: b "b #x !d . So long as the b s are updated according to Eq. (2) with the dividend discounting model in Eq. P "b # (4) R R (1#r)O (1#r) O The ﬁnal term in Eq. E [ ] is the expected value operator R conditioned on date t information. d is net dividends paid at time R R t. (2). (5) is just a restatement of the dividenddiscounting model which in no way depends on the properties of accounting numbers other than through the clean surplus relation.b ] E [b ] R R>O R>O\ ! R R> . the value of b and the values all the x s could be picked as random R R>O numbers. Thus. (5) is the residual income version of the dividenddiscounting model. and x is earnings for the period R R from t!1 to t. (5) from accrual accounting concepts is to redeﬁne b as the ﬁrm’s cash balance at the end of period t and x as the net R R eﬀect of all nondividend cash ﬂows for period t. Combining the clean surplus relation in Eq. (2).
(1). The point is illustrated by a recent application of the residual income valuation model in Frankel and Lee (1998). R b(i) is b(i!1) #f (i) !d(i) (the period t forecast of book value for period t#i). The redundancy of the residual income valuation model applies more generally to studies that generate explicit forecasts of earnings and book values (and hence dividends) for several periods. and overlooking it can lead empiricists to implement the residual income valuation model by incorporating explicit estimates of future dividends.r (1#r) where f (i) is the period t consensus analyst forecast of earnings for period t#i. (5) by forecasting abnormal earnings for three periods and taking the last period in perpetuity as follows: f (2) !r. (1997). P" R (1#r) (1#r) (1#r). In order to estimate future book values. Penman (1997) demonstrates how some of the more common terminal value Similar terminal value assumptions are used by Francis et al. 1997). R As a matter of algebra.g. the valuation model can be viewed as an application of the dividenddiscounting formula in which explicit forecasts of dividends are provided for the ﬁrst two periods and dividends are assumed to equal the forecast of period t#3 earnings thereafter. the researcher must estimate future dividends. Francis et al. and the researcher may just as well have used the dividenddiscounting model in Eq. the book value and earnings estimates become redundant. It is also noteworthy that the book value of equity drops out of this particular model. and then use a terminal value assumption to complete the valuation (e. R R R R and d(i) is period t forecast of dividends for period t#i. R R# P "b # R R (1#r) (1#r). The valuation relation cannot be implemented without estimates of future book values. . 1998. However. Eq. without realizing that this makes the appeal to the residual income formulation of the dividend discounting model somewhat redundant.b R R# R R. The above point is subtle. / Journal of Accounting and Economics 26 (1999) 1–34 5 From an empirical standpoint.b(1) f (3) !r.M.b(2) f (1) !r.P.r Thus. Lee et al. Frankel and Lee. this valuation expression reduces to d(1) d(2) f (3) R # R # R . Dechow et al. and the appeal to the residual income formulation of the dividenddiscounting model is redundant. (1998) and Penman and Sougiannis (1996). once future dividends are estimated.. The valuation model is readily interpretable in the context of the original dividenddiscounting model. They implement Eq.. (5) leaves the researcher in much the same position as the dividenddiscounting model.
Ohlson’s third assumption is that abnormal earnings satisfy the following modiﬁed autoregressive process: x " x #v # . are readily available and easily measured. while the residual income formulation of the dividenddiscounting model may have intuitive appeal because of its focus on accounting numbers. x and v ) and R R R three parameters ( . Combining the residual income valuation model in Eq. nor does it require additional assumptions about the computation of ‘terminal value’. R R R R where " /(1#r! ) and "(1#r)/[(1#r! )(1#r! )]. mean zero disturbance term. as the diﬀerence between the conditional expectation of abnormal R . Both Ohlson (1995) and Lundholm (1995) emphasize that the original empirical implications of Ohlson’s model depend critically on the third and ﬁnal assumption regarding the abnormal earnings information dynamics. However. (5) with the information dynamics in Eqs. (6a) R> R R R> v " v# . This assumption places restrictions on the standard dividenddiscounting model. and the three parameters are more diﬃcult to measure. This valuation function does not require explicit forecasts of future dividends. (6a) and (6b) yields the following valuation function: (7) P "b # x # v . and r) to be provided as inputs. (6a) and (6b) and the valuation function in Eq. book value (b ) and earnings (x ). and and are earnings. it is well established that prices reﬂect information about R future earnings that is not contained in current earnings. it provides no new empirical implications in and of itself. (7) embody the original empirical implications of Ohlson (1995). The R R remaining variable. GR ﬁxed persistence parameters that are nonnegative and less than one. / Journal of Accounting and Economics 26 (1999) 1–34 assumptions employed in the residual income valuation model are readily interpretable in the context of the standard dividenddiscounting framework. (7) requires three variables (b . Eq. (6a) indicates that Ohlson deﬁnes his other information variable. (1980). R Turning ﬁrst to v . From a theoretical perspective. the third assumption speciﬁes the nature of the relation between current information and the discounted value of future dividends. v . v . Attempts to incorporate this other information into valuation analyses date back at least as far as Beaver et al.2. Empirical implementation Empirical implementation of the information dynamics in Eqs. 2. Thus. (6a) and (6b) along with the valuation function in Eq. The ﬁrst two variables.M. the ﬁrm is still being valued by discounting future dividends.6 P. The information dynamics in Eqs. (6b) R> R R> where v is information about future abnormal earnings not in current abnormal R is the unpredictable. Dechow et al.
it is well established that nonrecurring special items. We measure the period t conditional expectation of period t#1 earnings using the consensus analyst forecast of period t#1 earnings. First.P. First. The estimation procedure is described in more detail in Section 3. The persistence of abnormal earnings is a function of the persistence of the abnormal accounting rate of return and the growth rate in book value. Details of the estimation procedure are again provided in Section 3. / Journal of Accounting and Economics 26 (1999) 1–34 7 earnings for period t#1 based on all available information and the expectation of abnormal earnings based only on current period abnormal earnings: v "E [x ]! x . denoted f . values for the three parameters . 1998). so we expect that will be lower for ﬁrms with extreme levels of operating accruals. variables that forecast the persistence of accounting rates of return and the growth rate in book value will determine . Second.M. We use the average historical return on equities to measure r. and r. must be established. 1996). Economic analysis points us to two factors that are expected to relate to the persistence of abnormal earnings. v can then be measured as R v "f ! x . Dechow et al. (1982) provide evidence that extreme levels of earnings and extreme accounting rates of return mean revert more quickly. The characteristics that we use are described in more detail below. The extant accounting literature has identiﬁed a number of factors aﬀecting the persistence of accounting rates of return. Firms with growth opportunities tend to have lower payout We are grateful to Jim Ohlson for suggesting this procedure for measuring v (see Ohlson. we expect that will be smaller for ﬁrms with extreme abnormal accounting rates of return. We measure and using their historical unconditional sample estimates. such as restructuring charges and asset writedowns. R R> R R R The other information.g. Sloan (1996) establishes that accounting rates of return are less likely to persist for ﬁrms with extreme levels of operating accruals. dividend policy serves as an indicator of expected future growth in the book value of equity. respectively.b .. We refer to these estimates as and S. Third. so we expect that will be lower for ﬁrms with extreme levels of special items. are less likely to persist (e. R R R> R Note that the conditional expectation of period t#1 abnormal earnings is equal to the conditional expectation of period t#1 earnings less the product of period t book value and the discount rate.. which we refer to as . Thus. Thus. Fairﬁeld et al. R . Brooks and Buckmaster (1976) and Freeman et al. R R R Finally. We also develop a conditional forecast of using characteristics suggested by accounting and economic analysis. so that R E [x ]"f "f !r.
we are able to rule out several of the combinations of assumptions about the parameters and . (e. / Journal of Accounting and Economics 26 (1999) 1–34 ratios. 1994). 1988. numerous studies suggest a link between industry structure and ﬁrm proﬁtability (e. The columns of Fig.1. We do this because several of R the conditioning variables relate to shortterm mean reversion in abnormal . The rows of Fig. This is because we estimate S from a v R autoregression. . 1 each summarize valuation models that make alternative assumptions about the value of the abnormal earnings persistence parameter. Our empirical analysis focuses on the improvements provided by Ohlson’s model over these simpler and more restrictive models. Anthony and Ramesh. Dechow et al. R A priori. resulting in a higher . Scherer. Note that we superscript by the abnormal earnings persistence parameter. In particular..g.. Research design 3. 1.8 P. Second. v . 1 each summarize valuation models that make alternative assumptions about the other information variable. We therefore expect that the persistence of abnormal earnings should be increasing in the historical persistence of abnormal earnings for ﬁrms in the same industry. . respectively. S. Ahmed. The remaining three columns summarize valuation models that incorporate the other information variable into the valuation analysis. The ﬁrst column ignores other information altogether.M. 1980. Thus. Fazzari et al. The four rows consider values for of 0. 3. we rule out the use of with models incorporating the other information variable. First. v . 1992). The additional restrictions range from ignoring the ‘other information’ in analysts’ forecasts of earnings altogether. we predict that a variety of industryspeciﬁc factors should inﬂuence the persistence of abnormal earnings. to setting the persistence parameters and to their polar extremes of 0 and 1. The columns diﬀer with respect to the assumed value of the other information persistence parameter.. we expect that ﬁrms with low payout policies will experience growth in the book value of equity in the future. Model evaluation We evaluate the empirical implications of Ohlson’s residual income valuation model relative to several competing accountingbased valuation models. R and is therefore restricted to valuation models based on past abnormal earnings alone. The competing valuation models are summarized in Fig. 1 and the unconditional estimate.g. . We assume that the eﬀect of industry speciﬁc factors should be relatively stable over time. and we show that they can all be considered as special cases of Ohlson’s model. and the measurement of v depends on the value used for . 1. The three columns consider values for of 0. respectively. The competing valuation models generally correspond to valuation models that have been used in previous empirical research. the unconditional estimate ( ) and the conditional estimate ( ).
Dechow et al. it makes little sense to apply the low conditional persistence of this period’s abnormal earnings to the expectation of next period’s abnormal earnings. Thus. The intuition for including reinvested period t earnings is that they will increase the book value base that is available to generate earnings in the next period. This combination of assumptions implies that abnormal earnings are nonstationary. it makes little sense to apply the conditional persistence parameter for this period’s abnormal earnings to the conditional expectation of next period’s abnormal earnings. as it suggests that abnormal proﬁt opportunities will never be competed away. we rule out cases where one of the persistence parameters is assumed to be 1 and the other persistence parameter is assumed to be strictly positive. "1.g. . Cells in Fig. if current abnormal earnings consist of a large negative special item.g. "0. expected future abnormal earnings are zero and price is equal to book value. list both the expectation of next period’s abnormal earnings and the valuation function implied by the corresponding valuation model. ignore other information This model assumes that expectations of future abnormal earnings are based solely on current abnormal earnings and that abnormal earnings persist indeﬁnitely. We ﬁnd this implication unappealing from an economic standpoint. ignore other information This model assumes that expectations of future abnormal earnings are based solely on information in current abnormal earnings and that abnormal earnings are purely transitory. 1991). This restricted version of Ohlson’s model corresponds to valuation models in which accounting earnings are assumed to measure ‘value creation’ (e. Barth. / Journal of Accounting and Economics 26 (1999) 1–34 9 earnings that is not necessarily expected to persist beyond the next period. we do not expect that a corresponding special item will be reported in next period’s earnings.. 1996). 1991). This special case of Ohlson’s model corresponds closely to the popular earnings capitalization valuation model in For example. The remaining cells. we brieﬂy discuss each of the valuation models and provide examples of prior research using the valuation models... These assumptions imply that expected abnormal earnings equal current abnormal earnings and price equals current earnings capitalized in perpetuity plus any reinvested period t earnings. However.. Fairﬁeld et al. Variants of this valuation model are implicit in many ‘levels’ studies in which market values are regressed on book values (e. Thus. Easton and Harris. Below. resulting in a low conditional persistence parameter. Second. . Consequently.M. 1 corresponding to one of the valuation models that we rule out above are labeled ‘Not considered’.P. 1982. This implication is also inconsistent with past empirical evidence suggesting that accounting rates of return are mean reverting (Freeman et al. then we would expect earnings to be temporarily low this period.
10 P. Dechow et al.M. / Journal of Accounting and Economics 26 (1999) 1–34 .
is the ﬁrst order autoregression coeﬃcient for the other information variable. R R R where x denotes earnings before extraordinary items and discontinued operations for year t.b R R R f denotes the I/B/E/S consensus forecast of earnings for year t#1 measured in the ﬁrst month following the announcement of earnings for year t. v . b denotes book value of equity at the end of year t. Each cell contains the forecast of next period’s abnormal earnings (E [x? ]) and the contemporaneous forecast of stock price (P ) for the respective model. R R> R Abnormal earnings for year t is deﬁned as x "x !r.b . 1.M. Summary of the implications of the accountingbased valuation models examined in the empirical tests. Dechow et al. R is the ﬁrst order autoregression coeﬃcient for abnormal earnings and is estimated using all historically available data from 1950 through the forecast year in a pooled timeseries crosssectional regression. R r denotes the discount rate (assumed to be 12%).Fig. and is estimated using all historically available data from 1950 through R the forecast year in a pooled timeseries crosssectional regression. is the predicted value of from the regression model speciﬁed in Table 2 and estimated using all historically available data from 1950 through the forecast year. v is deﬁned as R v "f ! x . / Journal of Accounting and Economics 26 (1999) 1–34 f "f !r. R R R where the period t consensus analyst forecast of abnormal earnings for the next period is deﬁned as P. R 11 .
"0 Unlike the prior model. An important feature of the model is that book value does not enter the valuation function.M. The relative weight on book value (abnormal earnings) is decreasing (increasing) in the persistence parameter. Variants of this model have long been popular in empirical applications of the dividenddiscounting model. For example. Note that expected abnormal earnings equal the consensus analyst forecast of abnormal earnings by construction for all of the models incorporating the other information variable. / Journal of Accounting and Economics 26 (1999) 1–34 which earnings are assumed to follow a random walk and the future dividend payout ratio is assumed to be 100% (e. Thus. 1995). This model corresponds to Penman and Sougiannis’ (1996) application of the residual income valuation model with a one period horizon and no terminal value. "1. this model combines elements of the two preceding models. "0. . Price is equal to book value plus the discounted value of the forecast of next period’s abnormal earnings. Expected abnormal earnings are equal to the consensus analyst forecast of abnormal earnings. Price is a linear function of book value and current abnormal earnings. " . 1992. except that the unconditional estimate of the persistence parameter ( ) is replaced by the conditional estimate of the persistence parameter ( ). Kothari and Zimmerman. Expected abnormal earnings equal current abnormal earnings multiplied by the persistence parameter. The model is considered in more detail in Ohlson (1991). "0 This valuation model incorporates the other information in the conditional forecast of next period’s abnormal earnings. Whitbeck and Kisor (1963) and Vander Weider and Carleton (1988) model the ratio of price to the consensus analyst forecast of next period earnings as a function of . and that abnormal earnings mean revert at their unconditional historical rate.. current book value receives a heavy weighting in the valuation function.12 P. Consequently. this model assumes that forecasted abnormal earnings persist indeﬁnitely. Dechow et al. " . because forecasted abnormal earnings are assumed to be purely transitory. but assumes that both abnormal earnings and the other information variable are purely transitory. Kothari. ignore other information This model assumes that expectations of future abnormal earnings are based solely on current abnormal earnings. Abnormal earnings have no implications for ﬁrm value beyond next period. " . so price is equal to the forecast of next period’s earnings capitalized in perpetuity.g. ignore other information This model is identical to the model discussed directly above.
" . "0). that this model does not allow for mean reversion in abnormal earnings. Note.P. and so does not place any weight on current book value in the valuation function.M. Frankel and Lee (1998).g. 1998). The relative weight on book value (forecasted abnormal earnings) is decreasing (increasing) in the persistence parameter. or grow at some nominal rate (e. "1 This valuation model is identical to the model obtained by assuming that ( "1. 1 is the substitution of S for in the valuation function. "0. " . Dechow et al. 4%). (1997) all compute terminal value by assuming that abnormal earnings in the terminal year either remain constant in perpetuity. Thus. " . this model combines elements of the two preceding models. this model has formed the basis for the computation of terminal values in empirical applications of the residual income valuation model using ﬁnite horizon data. however. v captures the entire expectation of next period’s abnormal earnings. The diﬀerence is reconciled by noting that when "0. This model represents our best attempt to implement the residual income valuation model proposed by Ohlson (1995). / Journal of Accounting and Economics 26 (1999) 1–34 13 the dividend payout ratio and expected growth in earnings. More recently. Assuming that "1 then has the eﬀect of allowing the expectation of next period’s abnormal earnings to persist indeﬁnitely. Allowing both abnormal . While this model is appealing in that it combines analysts’ forecast data with information in book value. note that the valuation function is always symmetric in and (see Ohlson.. it has received little attention in the empirical literature. Penman and Sougiannis (1996) and Francis et al. Lee et al. More generally. (1998). "0). "0. "0 This model allows for gradual mean reversion in next period’s expected abnormal earnings by assuming that equals its historical unconditional value. Bernard (1995) captures the spirit of this model by regressing price on book value and shortterm forecasts of abnormal earnings. which is discussed above. " S The symmetry of the valuation function implies that this valuation model is identical to the model obtained by assuming that ( " . The intuition for this result is that "0 implies that v measures the complete expectation of next period’s abnormal earnings. One apparent diﬀerence between the two models that is evident from Fig. which is discussed above. so that S reﬂects the persistence of next period’s abnormal earnings. " S The ﬁnal valuation model sets both and equal to their historical unconditional values. Price is a linear function of book value and the forecast of next period’s abnormal earnings.
2. An abnormal earnings autoregression is estimated using all available observations from previous years. Thus. and the 1% most extreme observations are winsorised so that they do not have an undue inﬂuence on the regressions. Our primary empirical analysis uses annual ﬁnancial statement data from 1976 to 1995.133 observations. Stock return data are obtained from the CRSP daily ﬁles. The relative rankings of the models in the empirical tests are robust to discount rates ranging from 9% to 15%. Combining the three databases gives us a total of 50. The resulting estimate of the autoregressive parameter.14 P. The empirical analysis is conducted using pershare data. current abnormal earnings and the other information variable. Our analysis requires a measure of the discount rate. Historical accounting data are obtained from the COMPUSTAT ﬁles.M. Strictly speaking. current abnormal earnings and the other information embedded in the forecast of next period’s abnormal earnings all contain incremental information about price. and our objective is not to evaluate alternative methods for estimating discount rates. / Journal of Accounting and Economics 26 (1999) 1–34 earnings and the other information variable to each have their own persistence parameters produces a valuation function in which price is a linear combination of book value. we use a discount rate of 12%. All of our empirical tests employ withdividend stock returns and buyandhold returns. attempts to document predictable variation in expected returns that are consistent with the predictions of asset pricing models have met with limited success. Moreover. r. is used to implement the unconditional version of Ohlson’s model. extraordinary items are nonrecurring. The unconditional value of used in the Ohlson valuation model is estimated separately for each ﬁscal year. Note that the discount rate enters all of the models in a similar fashion. Data and variable measurement The empirical analysis uses three data sources. . . All of our tests use earnings measured before extraordinary items. This is the procedure that we adopt for special items. Dechow et al. All variables are scaled by market value of equity to control for heteroscedasticity. from a practical perspective. excluding extraordinary items from earnings violates the clean surplus assumption underlying the theoretical development of the residual income valuation model. and so their inclusion is unlikely to enhance the prediction of abnormal earnings. which approximates the longrun average realized return on US equities. . An alternative procedure would be to use a deﬁnition of earnings that incorporates extraordinary items and to then incorporate the lower persistence of the comprehensive earnings number in the persistence parameter. This valuation model implies that book value. 3. However. going back as far as 1950. Analyst forecast data is obtained from the I/B/E/S ﬁles.
because a payout ratio greater than one cannot be sustained indeﬁnitely. The ﬁrst variable (q1) measures the magnitude of abnormal earnings. The third variable (q3) measures the magnitude of operating accruals.M. One complication that arises in the estimation of S is that the measurement of the other information variable. is estimated in a similar manner. Recall from Fig. then is set equal to . and is computed as the absolute value of the ratio of abnormal earnings to lagged book value.. The regressions use all available observations from 1950 through the previous year. 1996). If the dividend payout ratio is negative due to negative earnings. through an ‘other information’ autoregression. / Journal of Accounting and Economics 26 (1999) 1–34 15 The conditional value of the autoregressive parameter. using the same procedure that we used to estimate . The fourth variable (div) measures the dividend payout policy and is computed as the ratio of dividends to earnings over the most recent ﬁscal year. Next. with each regression using all available observations in the sample from previous years. R\ R\ R\ R\ R A separate regression is estimated for each ﬁscal year in the sample. and is computed as the absolute value of the ratio of special items to lagged book value. we use the ratio from the most recent previous year in which the ﬁrm reported positive earnings.P. q2.g. Sloan. is estimated via an abnormal earnings autoregression in which each of the ﬁve determinants of are included as interactive eﬀects: x " # x # (x q1 )# (x q2 )# (x q3 ) R R\ R\ R\ R\ R\ R\ R\ # (x div )# (x ind )# . going back as far as 1950. A pooled industryspeciﬁc abnormal earnings autoregression is used to estimate the historical persistence parameter for each SIC grouping. div and ind: A" # q1 # q2 # q3 # div # ind . Dechow et al. we set it to one. The ﬁfth variable (ind) measures the historical persistence of abnormal earnings for ﬁrms in the same industry. We use twodigit SIC codes to measure industry membership. q3. . R R R R R If one of the variables required to compute is missing. Operating accruals are computed in the standard way (e. A ﬁner partitioning results in an unsatisfactorily low number of observations for some industries. Finally. 1 that we only require . depends on an assumed value of . and is measured as the absolute value of the ratio of operating accruals to lagged total assets. we estimate S. We ﬁrst construct the ﬁve variables that are hypothesized to be associated with crosssectional variation in the persistence of abnormal earnings. If the ratio is greater than one. The estimate for each ﬁrmyear is then computed using the parameter estimates from this regression and the ﬁrmyears actual values of q1. The second variable (q2) measures the magnitude of special items. v.
Third. we examine whether the ﬁrstorder autoregressive process is suﬃcient for abnormal earnings by adding additional lags of abnormal earnings. The plots in Fig. The plots compare the predictive ability of timeseries models setting equal to 0.M. / Journal of Accounting and Economics 26 (1999) 1–34 a measure of in the situation where " . Panel A reports the results from a ﬁrstorder abnormal earnings autoregression. because mean reversion in abnormal earnings is almost complete after four years. it does a relatively good job of tracking longterm abnormal earnings. Dechow et al. The ﬁgure is constructed by ﬁrst ranking all sample observations on deﬂated abnormal earnings and equally assigning the ranked observations to deciles. the hypotheses that "1 and "0 respectively are both strongly rejected. The ﬁrst three tests are presented in Table 1 and employ pooled timeseries and crosssectional regression analysis. Second. We test ﬁve aspects of the timeseries behavior of abnormal earnings. is 0. Hence. First. Finally. we estimate by allowing the autoregressive coeﬃcient on abnormal earnings to vary as a function of our conditioning variables. diﬀers reliably from the polar extremes of 0 and 1. we examine whether the autoregressive coeﬃcient. Note also that while the model using "0 does a poor job of predicting shortterm abnormal earnings.62 with a tstatistic of 138. " S). we need only estimate the other information autoregression with v measured using . We also consider the model with ( "0. we examine whether the autoregressive coeﬃcient. Fourth. It can be readily seen that the model using " tracks subsequent abnormal earnings the most closely. Empirical results 4. 1 and .16 P.31. 2 illustrate the superior forecasting ability of a timeseries model that incorporates the gradual mean reversion in abnormal earnings. S measures the persistence of abnormal earnings. which is given by . . We measure v using the estimate of obtained from all data available through the end of R period t!1. . . The mean values of abnormal earnings for the highest and lowest deciles are then plotted over the next four years. Thus. In this case. 4. we relax the constraints that the autoregressive process places on the earnings and book value components of abnormal earnings. along with the values implied by each of the models.1. respectively. Timeseries behavior of abnormal earnings We begin our empirical analysis by evaluating how well the evolution of abnormal earnings is described by Ohlson’s information dynamics. . The autoregressive coeﬃcient. diﬀers reliably from the polar extremes of 0 and 1.
but the coeﬃcient magnitude is only 0.59 (68.04) 0.09 (!77.36) 0.01 (!12. Abnormal earnings for year t is deﬁned as: x "x !r.P.86) 0.35. .02 (!29. (1996) investigate the appropriateness of the single lag information dynamic in a more general framework and ﬁnd that a second lag achieves modest statistical signiﬁcance.133 annual observations from 1976 to 1995.50) 0.34 17 Panel B: Pooled analysis with four lags x " # x # x # x # x # GR> GR GR\ GR\ GR\ GR> !0. increasing the explanatory power from 0.01 (1. Finally.31) " # x # GR> GR GR> R 0. / Journal of Accounting and Economics 26 (1999) 1–34 Table 1 Autoregressive properties of abnormal earnings Panel A: Pooled analysis with one lag x !0. Panel C reports the BarYosef et al.62 (138. Figures in parentheses are tstatistics.07 (7.M. Thus.47 (80.34 to 0.12) !0.59 on the ﬁrst lag.35 Panel C: ºnconstrained estimation with one lag x " # x # b # GR> GR GR\ GR> 0.59) R 0. Inclusion of three additional lags of abnormal earnings has a trivial impact.b R R R\ where x denotes earnings before extraordinary items for year t. r denotes the discount rate (assumed R to be 12%).07 versus 0. R Panel B of Table 1 reports results including additional lags of abnormal earnings to examine whether the ﬁrstorder autoregressive process is suﬃcient.16) 0. Only the second lag is statistically signiﬁcant (t"7. All variables are scaled by the market value of equity at the end of year t.64) R 0.02 (17. the ﬁrst order autoregressive process appears to provide a reasonable empirical approximation.40 Notes: Sample consists of 50.31) 0. and b denotes book value of equity at the end of year t.50).01 (0. Dechow et al.
we see that book value loads with a signiﬁcantly negative coeﬃcient and that the inclusion of book value leads to a decline in the coeﬃcient on abnormal earnings. . . Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34 Fig.133 observations from 1976 to 1995. If the ﬁrstorder autoregressive process is appropriate. However. then the additional book value term should not load in the regression.M.18 P. However. unreported tests reveal that this unconstrained speciﬁcation is not signiﬁcantly helpful in forecasting future abnormal earnings and so it is not pursued further. results of regressions of abnormal earnings on lagged abnormal earnings and the book value component of lagged abnormal earnings. 2. The sample consists of 50. Feltham and Ohlson (1995) suggest that the negative loading on book value can be interpreted as ‘aggressive’ accounting. Comparison of the actual timeseries properties of abnormal earnings with the properties predicted by a ﬁrstorder autoregressive process with alternative values for the autoregressive coeﬃcient. The graph is formed by taking observations in the extreme deciles of abnormal earnings performance in year 0 and plotting the mean level of abnormal earnings performance for each decile over the following four years.
133 observations from 1976 to 1995. Table 2 reports results from allowing each of the hypothesized determinants of persistence to enter as interactive variables in the abnormal earnings autoregression.21 (!35.97) ? 0.10) Notes: Sample consists of 50.M. q3 is deﬁned as the absolute value of accounting accruals divided by total assets at the beginning of R year t. . q2 is deﬁned as the absolute value of special accounting items divided by book value of equity at the R beginning of year t.77) ! # 0. q2 and q3.68) ! !1. All of the interactive eﬀects enter with their hypothesized signs and are statistically signiﬁcant.80) (8.22) ! !0. div is deﬁned as dividends paid during year t divided by earnings before extraordinary items and R discontinued operations for year t. Abnormal earnings are scaled by market value of equity at the end of year t. These results conﬁrm that the persistence of Table 2 Determinants of the persistence of abnormal earnings x " # x # (x q1 )# (x q2 )# (x q3 )# (x div ) R R\ R\ R\ R\ R\ R\ R\ R\ R\ # (x ind )# R\ R\ R Predicted sign ? !0. R r denotes the discount rate (assumed to be 12%).59) ! !0. .P. Recall that this coeﬃcient measures the persistence of abnormal earnings and is hypothesized to have ﬁve determinants. Dechow et al.37 (!28. ind is deﬁned as the ﬁrst order autoregressive coeﬃcient from an abnormal earnings autoregression R for all ﬁrms in the same two digit SIC code as the observation.02 (!30.11 0.34 to 0. measured by the empirical constructs q1.b R R R\ where x denotes earnings before extraordinary items and discontinued operations for year t. Persistence is hypothesized to be lower when earnings contain more transitory accounting items. Inclusion of the ﬁve interactive eﬀects increases the explanatory power of the regression from 0.61 (!7. Figures in parentheses are tstatistics.61 (13. / Journal of Accounting and Economics 26 (1999) 1–34 19 Table 2 analyses variation in the autoregressive coeﬃcient. and b denotes book value of equity at the end of R year t.17 (!3. q1 is deﬁned as the absolute value of abnormal earnings for year t divided by book value of equity at R the beginning of year t. The autoregression is conducted using all available ﬁrms on the COMPUSTAT annual tapes in the same two digit SIC code from 1950 through year t.40. Abnormal earnings for year t is deﬁned as x?"x !r. Persistence is also hypothesized to be decreasing in the dividend yield (div) and increasing in the historical level of industrywide abnormal earnings persistence (ind).40 R !0.
is 0.b R R R and abnormal earnings for period t is deﬁned as x "x !r.08 Notes: Sample consists of 50. However.94) R 0. the other information embedded in analysts’ forecasts of next R period’s abnormal earnings Pooled analysis with one lag v " # v# R> R R> 0.94. x denotes earnings before extraordinary items for year t. Thus. the other information mean reverts at about twice the rate of abnormal earnings. Dechow et al. .b R R R\ f denotes the I/B/E/S consensus forecast of earnings for year t#1 measured in the ﬁrst month R following the announcement of earnings for year t. Figures in parentheses are tstatistics. The other information variable is deﬁned as v "f ! x R R R where the period t consensus analyst forecast of abnormal earnings for the next period is deﬁned as f "f !r. is the ﬁrst order autoregression coeﬃcient for abnormal earnings and is estimated using all historically available data from 1950 through year t in a pooled timeseries crosssectional regression. Table 3 examines the autoregressive properties of the other information variable.133 annual observations from 1976 to 1995. Finally.20 P. The estimate of the ﬁrstorder autoregressive coeﬃcient on the other information. .01 (38. is also signiﬁcantly diﬀerent from the polar extremes of 0 and 1 that are implicitly assumed in many of the valuation models used in past research. . v. the conditional estimates of that we use to implement Ohlson’s valuation model should oﬀer additional predictive ability with respect to future abnormal earnings. Consequently.M. R r denotes the discount rate (assumed to be 12%). All variables are scaled by the market value of equity at the end of year t. R abnormal earnings varies in a systematic and predictable manner. b denotes book value of equity at the end of year t.79) 0.32 (57. we expect that incorporating more precise estimates of this coeﬃcient should improve our ability to forecast future abnormal earnings and hence predict contemporaneous stock prices. Thus. / Journal of Accounting and Economics 26 (1999) 1–34 Table 3 Autoregressive properties of v .32 with a tstatistic of 57.
029 0. Abnormal earnings for year t is deﬁned as x "x !r.P.006 !0.052 0. b denotes book value of equity at the end of year t. computed as E [x ]"f R R> R !0. is the ﬁrst order autoregression coeﬃcient for abnormal earnings and is estimated using all historically available data from 1950 through the forecast year in a pooled timeseries crosssectional regression.b R R R where x denotes earnings before extraordinary items and discontinued operations for year t.087 0. is the predicted value of from the regression model speciﬁed in Table 2 and estimated using all historically available data from 1950 through the forecast year.032 0.033 0.133 observation from 1976 to 1995.006 Mean absolute forecast error 0. The forecast error for year t is computed by subtracting the forecast of abnormal earning for year t#1 from the realized abnormal earnings for year t#1.032 0.M.b R R R and the period t consensus analysts’ forecast of abnormal earnings for period t#1 is deﬁned as f "f !r. R f denotes the I/B/E/S consensus forecast of earnings for year t#1 measured in the ﬁrst month R following the announcement of earnings for year t. .2. while the mean absolute forecast error and the mean square forecast error measure forecast accuracy. Dechow et al.081 0. The mean forecast error measures forecast bias.030 0.015 Notes: Sample consists of 50. Forecast errors are scaled by the market value of equity at the end of year t. computed as E [x ]" x R R> R Mean forecast error "0 "1 " " !0.028 Panel B: Prediction for models incorporating ‘other information’.076 Mean square forecast error 0.077 0. Prediction of next period abnormal earnings Statistics on the bias and accuracy of the predictions of next period abnormal earnings generated by each of the valuation models are reported in Table 4. All forecast errors are deﬂated by market value and the extreme 1% of the forecast errors are Table 4 Relative forecasting ability of alternative modes for predicting next year’s abnormal earnings Panel A: Predictions for models ignoring ‘other information’. R r denotes the discount rate (assumed to be 12%). / Journal of Accounting and Economics 26 (1999) 1–34 21 4.008 !0.
respectively. recall from Fig. " and " . the model using "1 S is considerably less accurate than the other three models. 4. reﬂecting overoptimism in analysts’ forecasts.22 P. The measures of forecast accuracy are similar for the models using "0. this model therefore generates relatively poor forecasts of stock price. / Journal of Accounting and Economics 26 (1999) 1–34 winsorised. 2 that the model using "1 model generates poor forecasts of longrun abnormal earnings. Dechow et al. We measure the analysts’ earnings estimates using the I/B/E/S mean consensus earnings estimates provided in the month immediately following the announcement of the annual earnings data used in the timeseries models. The mean forecast error is close to zero for the models using "1. The model using has only slightly smaller forecast errors than the model using . and is slightly negative for the model using "0 (!0. Hence. v. Explanation of contemporaneous stock prices The relative ability of the competing valuation models to explain contemporaneous stock prices is evaluated in Table 5. on average.291) and greatest for the model using "1 (forecast error"0. Panel A of Table 5 reports results for the four models ignoring the other information.3. The measures of forecast accuracy indicate that the predictive abilities of the models ignoring the other information in analysts’ forecasts are all very close. This result highlights the important role of the other information embedded in analysts’ forecasts in predicting future abnormal earnings. and the model using "0 is the least accurate of all. All of these models generate large positive mean forecast errors. ﬁrms fell slightly short of achieving a return on equity equal to the assumed cost of capital of 12%. The mean forecast error is also negative using the consensus analyst forecast. " and " . while panel B reports forecast errors for the models that incorporate v.M. The undervaluation is smallest for the model using "0 (forecast error"0.378). This latter result indicates that. The model using "1 is slightly less accurate than the two versions using estimates of . However. This ensures that all of the forecasting variables are measured at similar points in time. Since expectations of longrun abnormal earnings are included in the computation of stock price. Panel A reports forecast errors for each of the models that ignore the other information variable. indicating that they undervalue equities relative to the stock market. Recall that the forecast of next period abnormal earnings is equal to the consensus analyst estimate of abnormal earnings for all of the models that incorporate v.029). The results in panel B indicate that analysts’ forecasts of abnormal earnings are much more accurate than the forecasts generated by the historical timeseries models. we only report one set of forecast errors for these models. indicating that our eﬀorts to conditionally estimate the persistence parameter add relatively little to the forecasting ability of the model. To understand this result. The mediocre showing of the .
.266 0.241 Notes: Sample consists of 50. / Journal of Accounting and Economics 26 (1999) 1–34 Table 5 Relative forecasting ability of alternative modles for explaining contemporaneous stock prices Panel A: Price estimates for models ignoring ‘other information’. v is deﬁned as R v "f ! x R R R where the period t consensus analyst forecast of abnormal earnings for the next period is deﬁned as f "f !r.b R R R where x denotes earnings before extraordinary items and discontinued operations for year t. and is estimated R using all historically available data from 1950 through the forecast year in a pooled timeseries crosssectional regression.278 0.320 0. "0) and ( "0.232 0. is the predicted value of from the regression model speciﬁed in Table 2 and estimated using all historically available data from 1950 through the forecast year.326 Mean absolute forecast error 0.419 0. " S) 0. S is the ﬁrst order autoregression coeﬃcient for the other information variable.427 0.291 Panel B: Price estimates for models incorporating ‘other information’.227 0.363 0.461 0. "1) " . "0) and ( "0. and b denotes book value of equity at the end of R year t. v .284 0. "0) "1.259 0.284 0. R r denotes the discount rate (assumed to be 12%).248 0.133 observations from 1976 to 1995.519 0. computed as P "b # x R R 1#r! R Mean forecast error "0 "1 " " 0. Abnormal earnings for year t is deﬁned as x "x !r. Dechow et al.378 0. Forecast errors are scaled by stock price at the end of year t The forecast error for year t is computed by subtracting the forecast stock price for year t from the observed market stock price at the end of the month following the announcement of earnings for year t.P. " S) " .461 0.285 0.402 0.445 0.465 23 Mean square forecast error 0.M. is the ﬁrst order autoregression coeﬃcient for abnormal earnings and is estimated using all historically available data from 1950 through the forecast year in a pooled timeseries crosssectional regression. computed as 1#r P "b # x# v R R 1#r! R (1#r! )(1#r! ) R ( ( ( ( "0.291 0.b R R R f denotes the I/B/E/S consensus forecast of earnings for year t#1 measured in the ﬁrst month R following the announcement of earnings for year t.
the undervaluations are not as large as they were for the models ignoring other information. Panel B of Table 5 reports results for the models incorporating the information in analysts’ forecasts.r)/(1#r! )"1. the poor showing of this model in the pricing tests raises the possibility that stock prices do not reﬂect rational expectations of future abnormal earnings.24 P. Recall from Fig. the model using ( "1. 670). we investigate the ability of the information variables used in the valuation models to explain stock prices without imposing the restrictions implied by the valuation models. These results are consistent with the superior predictive accuracy of analysts’ forecasts with respect to future abnormal earnings. because book value contains additional information about longrun abnormal earnings that should be rationally reﬂected in stock prices. Thus. Dechow et al. Thus. This result is surprising. However. and "( # . The fact that book value loads in addition to earnings indicates that book value contains value relevant information beyond that already in earnings. /(1#r! )"0.39.M.62 (historical average from (Table 1) gives: "1!r. "0) provides the most accurate forecasts of stock prices.85.074) and "3.88 (0. We can obtain further insights from the regressions by comparing the estimated coeﬃcients to values implied by Ohlson’s model in conjunction with representative parameter values. We explore this issue in more detail later in the paper.40 (0. the strong showing of the model using ( "1. The corresponding mean values (standard errors) on the empirical regression coeﬃcients are "0. All of the models incorporating the other information have lower forecast errors than the models using historical data. "0) in the pricing tests again raises the possibility that stock prices do not reﬂect rational expectations of future abnormal earnings.262). 1 that this model simply capitalizes the forecast of next period’s earnings in perpetuity and ignores information in book value. Thus. Using r"12% (longrun historical average) and "0. The formulae for the predicted valuation coeﬃcients on book value and earnings are taken from Ohlson (1995). In Table 6. / Journal of Accounting and Economics 26 (1999) 1–34 model using " is somewhat surprising. These two explanatory variables are the information variables used in the valuation models that ignore other information. Panel A of Table 6 reports results of annual crosssectional regressions of stock price on historical book value and earnings. stock prices appear . (p. Of the models incorporating other information. The undervaluations are surprising. because the results in Table 3 indicate that the analysts’ forecasts of future abnormal earnings are overoptimistic. The mean forecast errors indicate that these models also undervalue relative to the market. Both book value and earnings load positively and signiﬁcantly in the regressions. Table 4 illustrates that this model generates the most accurate forecasts of next period’s abnormal earnings among the four models ignoring the other information.
07 0.85 0.72 0. 0.015 Med.05 5.133 observations from 1976 to 1995. The explanatory power of these .26 0.06 !0. 9.408 0.25 0. Err.262 0.61 4. In addition to book value and earnings.68 4. because investors overestimate the persistence of abnormal earnings.256 0.53 0.40 3.18 2.64 !0.035 0. R f denotes the I/B/E/S consensus forecast of earnings for year t#1 measured in the ﬁrst month R following the announcement of earnings for year t.05 0.M. Q1 Med. / Journal of Accounting and Economics 26 (1999) 1–34 Table 6 Unconstrained regressions of stock price on the variables used in the valuation models Panel A: Regressions of price on book value and earnings P" # b# x# R R R R Min. x denotes earnings before extraordinary items and discontinued operations for year t.88 0. An alternative interpretation is that stock prices do not reﬂect rational expectations.24 0. Sample consists of 50.85.97 0.74 Max 7.67 R P 1 anel B: Regressions of price on book value.82 3.56 3.27 0.81 6.92 0.019 1.69 Q3 5.39 0. Dechow et al.03 5.40 Std.07 0. Mean 9.56 6.59 Max 13.89 0.64 0.42 1. Q1 7.51 3. The regressions reported in panel B of Table 6 employ the information variables used in the valuation models incorporating other information. R b denotes book value of equity at the end of year t.63 0.74 0. 0.86 R Notes: Statistics reported are based on the estimates from 20 annual crosssectional regressions from 1976 to 1995.40 8.074 0.53 Q3 10. Min.57 0.79 0.09 0. All variables are measured on a pershare basis.07 0.65 !0.353 0.68 Coeﬀ.43 0.150 0.00 0. Mean 4. P denotes the stock price measured at the end of the month following the announcement of earnings R for year t. err. to place too low a weight on book value and too high a weight on earnings.P.05 3.09 !0. One interpretation of these results is that they are driven by a misspeciﬁcation in Ohlson’s valuation model.68 0.34 8. these regressions also include the consensus analyst forecast of next period’s earnings. earnings and the consensus analyst forecast of next year’s earnings P" # b# x# f# R R R R R Std.51 25 Coeﬀ. The value of required to justify the empirical regression coeﬃcients is approximately "0.53 4.
"[!(1#r). For example. An alternative interpretation is that stock prices do not reﬂect rational expectations. suggesting that analysts’ forecast of next year’s earnings subsume value relevant information in current earnings. . We investigate this possibility in the next section.62 (historical average from Table 1) and "0. ]/[(1#r! )(1#r! )]"!0. Book value loads positively and signiﬁcantly. a ( . One interpretation of these results is that they are driven by a misspeciﬁcation in Ohlson’s valuation model.150) and "5. indicating that the analysts’ forecasts contain incremental information about ﬁrm value. Dechow et al. The formulae for the predicted valuation coeﬃcients on book value and earnings are from Ohlson (1998) (p.26 P. we consider whether the values implied by the competing models are able to predict future stock returns. Using r"12% (longrun historical average) and "0. Earnings loads with a small and statistically insigniﬁcant coeﬃcient.55. The corresponding mean values (standard errors) on the empirical regression coeﬃcients are "0. the results in Table 4 . / Journal of Accounting and Economics 26 (1999) 1–34 regressions are considerably higher than in panel A.80.035). the analysts’ forecast of next year’s earnings loads with a positive and statistically signiﬁcant coeﬃcient.24 (0.M.256). 14). We can again obtain further insights from the regressions by comparing the estimated coeﬃcients to values implied by Ohlson’s model in conjunction with representative parameter values.79 (0. These additional tests are motivated by the apparent inconsistencies between the abnormal earnings prediction results in Table 4 and the valuation results in Tables 5 and 6. stock prices place too low a weight on book value and too high a weight on the analysts’ forecast of next year’s earnings. and "(1#r)/[(1#r! )(1#r! )]"2. 4. ) combination of approximately (0. This result indicates that book value contains some value relevant information beyond that in analysts’ forecasts of next year’s earnings. Thus. In particular. In this section.95. Finally.4. "0.00) would be required to approximate the empirical regression coeﬃcients.05 (0.32 (historical average from Table 3) gives: "[(1#r)(1! )(1! )]/[(1#r! )(1#r! )]"0.0. our pricing tests have focused on the ability of the competing valuation models to predict contemporaneous stock prices. though the coeﬃcient is much smaller than in the regressions excluding the analyst forecast variable.72. Prediction of future stock returns Thus far. because investors tend to overestimate the persistence of shortterm earnings forecasts.
197 0.169 0. the results in Table 5 indicate that the reverse holds true with respect to the ability of the models to explain observed stock prices.160 0.174 0.M. and are therefore expected to experience higher future stock returns.162 0. Note that the ratio formed for the model using "0 is just the booktomarket ratio.215 0. In the tradition of fundamental analysis. while the ratio formed for the model using "1 is proportional to the earningstoprice ratio.235 0.165 0. Higher deciles consist of stocks that are underpriced relative to intrinsic value.143 0.136 0.143 0.181 0.212 0. Panel A reports the oneyearahead returns for ratios formed on the valuation models ignoring other information.203 0.166 0.206 0.234 0.076 (2.161 0. we therefore provide tests of whether observed stock prices tend to revert toward the ‘fundamental’ or ‘intrinsic’ values implied by particular models. Decile portfolios are then formed using the ranked ratios.44) .P. Moreover.235 0. and are therefore expected to experience lower future stock returns.196 0.187 0. The predictive ability of these ratios with respect to future stock returns is already well documented. Table 7 Predictive ability of ratios of implied model values to observed market values with respect to stock returns over the following year Panel A: Implied values ignoring ‘other information’.159 0.94) "1 0.24) " 0.170 0. However.140 0.202 0.159 0.072 (1.194 0. These tests entertain the possibility of temporary stock mispricing that can be systematically predicted by particular valuation models.159 0. The results are presented in Table 7.153 0.158 0.180 0.39) " 0. computed as P "b # x R R 1#r! R Portfolio 1 (Lowest) 2 3 4 5 6 7 8 9 10 (Highest) Hedge (tstatistic) "0 0.152 0.171 0.094 (2.182 0.170 0. Dechow et al.181 0.173 0. Lower deciles consist of stocks that are overpriced relative to intrinsic value. / Journal of Accounting and Economics 26 (1999) 1–34 27 indicate that the model using " provides more accurate forecasts of future abnormal earnings than the models using " and "0.222 0. The tests are constructed by taking the ratio of the intrinsic model values to observed equity values.099 (2. the evidence in Table 6 is consistent with the expectations embedded in stock prices underestimating the mean reversion in abnormal earnings.175 0.191 0.
154 0.133 observations from 1976 to 1995.175 0.44) ( " .062 (1.b R R R f denotes the I/B/E/S consensus forecast of earnings for year t#1 measured in the ﬁrst month R following the announcement of earnings for year t.071 (1. Dechow et al.M. Abnormal earnings for year t is deﬁned as x "x !r.181 0.171 0.149 0. " S) 0.224 0.161 0.210 0.175 0.28 P.206 0.162 0.157 0.210 0. S is the ﬁrst order autoregression coeﬃcient for the other information variable.175 0.147 0. "1) 0.211 0. computed as 1#r x# v P "b # R (1#r! )(1#r! ) R R R 1#r! ( "1.179 0. observations are ranked and assigned in equal numbers to deciles based on the ratio of implied model value to observed market value of equity.154 0. v . is the predicted value of from the regression model speciﬁed in Table 2 and estimated using all historically available data from 1950 through the forecast year. "0) and ( "0.185 0.154 0.177 0.213 0.178 0. R r denotes the discount rate (assumed to be 12%). R R R where x denotes earnings before extraordinary items and discontinued operations for year t. and b denotes book value of equity at the end of R year t. . "0) Portfolio 1 (Lowest) 2 3 4 5 6 7 8 9 10 (Highest) Hedge (tstatistic) Notes: Each year.77) 0.204 0. and is estimated R using all historically available data from 1950 through the forecast year in a pooled timeseries crosssectional regression. beginning three months after the end of the ﬁscal year from which the historical forecast data are obtained.176 0.154 0. R R R where the period t consensus analyst forecast of abnormal earnings for the next period is deﬁned as f "f !r.070 (1.165 0.220 0.173 0. "0) and ( "0.162 0.178 0.145 0. ¹statistics are based on the timeseries standard errors of the 20 annual portfolio returns. / Journal of Accounting and Economics 26 (1999) 1–34 Table 7 (continued) Panel B: Implied values incorporating ‘other information’.34) ( " .159 0.201 0.203 0.224 0. v is deﬁned as R v "f ! x .b . is the ﬁrst order autoregression coeﬃcient for abnormal earnings and is estimated using all historically available data from 1950 through the forecast year in a pooled timeseries crosssectional regression.173 0. Sample consists of 50.208 0. Equalweighted buyhold stock returns are then computed for each decile portfolio over the subsequent 12 months. The table reports the mean of the 20 years of annual portfolio returns.174 0.054 (1.71) 0.158 0. " S) ( "0.
but the worst at predicting future stock returns. while the model using "1 is third with a hedge portfolio return of 7. Table 8 therefore reports results of more direct tests of the hypothesis that investors price predictable errors in analysts’ forecasts. "0). which is the diﬀerence between the return for portfolio 10 and the return for portfolio 1.M.6% (t"2. / Journal of Accounting and Economics 26 (1999) 1–34 29 The hedge portfolio return. The model using "0 displays the lowest predictive ability. These results contrast sharply with the contemporaneous stock price results in Table 5.44). Statistical inference is conducted using the standard error of the annual mean hedge portfolio returns over the 20 years in the sample period. with a hedge portfolio return of 7. the results in Table 7 are indirect and their statistical signiﬁcance is weak. Moreover. summarizes the predictive ability of each model with respect to future returns. However.4% (t"2.2% (t"1.94). ranging from 7. The regressions in panel A of Table 8 examine the extent to which each of the models ignoring other information in analysts’ forecasts detects errors in analysts’ forecasts of oneperiodahead earnings.4% (t"1. The superior predictive ability with respect to future stock returns of the model using " potentially explains why this model performs poorly in the pricing tests (Table 5). It appears that the expectations reﬂected in stock prices fail to fully anticipate the rate of mean reversion in abnormal earnings that is captured by this model.1% (t"1. this explanation should be interpreted with caution due to the low statistical signiﬁcance of the results. valuation models ignoring this information have the greatest predictive ability with respect to future stock returns. the valuation model using ( "1.9% (t"2.39). The return interval begins 3 months after the ﬁscal year end of the year from which the historical data is obtained. Panel B reports the oneyearahead returns for ratios computed using valuation models incorporating the other information in analysts’ earnings forecasts. Dechow et al. These relations are exactly what would be expected if analysts’ earnings estimates are naively incorporated in stock prices even when they do not fully reﬂect all information in current abnormal earnings about future abnormal earnings.24). despite its superior predictive ability with respect to future abnormal earnings (Table 4). "0) to 5. The regressions in panel B then examine whether the errors identiﬁed in the analysts’ forecasts appear to be accompanied by corresponding errors in stock prices. with a hedge portfolio return of 9. While valuation models incorporating information in analysts’ forecasts have the greatest ability to explain contemporaneous stock prices. with a hedge portfolio return of 9. Moreover.P. However. The model using " is second. The hedge portfolio returns are uniformly lower. "0) is the best at explaining contemporaneous stock prices. The model using " displays the greatest predictive ability.77) for the model using ( "0.44) for the model using ( "1. Kothari and Warner (1997) and Barber and Lyon (1997) provide evidence that statistical tests using long horizon stock returns are poorly speciﬁed. The results in panel .
40 !0.01 0.03 !0.05 0.02 0.48 0. Table 8 Panel A Analysis of the relation between forecast errors in abnormal earnings predictions from analysts’ consensus earnings estimates and forecasts of abnormal earnings that ignore the other information in analysts’ consensus earnings estimates.06 0.01 !0. The earnings forecasts that ignore the other information are generated by the model.03 !0. All variables are measured on a pershare basis.13 0.04 0.04 0.40 0.12 0.63 0.03 !0.01 !0.004 0.08 !0.01 !0.03 0. Q1 Med.043 0.01 0.05 !0.133 observations from 1976 to 1995.02 0.01 0. Regression model estimated is (x !f )" # ( x !f )#e R> R R R R> Coeﬀ. Thus the hypothesis that investors naively price predictable errors in analysts’ forecasts provides a promising explanation for the results obtained in this paper. error Min.04 !0.017 !0.03 0.03 0.29 0. The results in panel B indicate that these systematic forecast errors are reﬂected in stock prices.03 0.01 !0.005 0.44 0.35 !0. "0 R "1 R " R " R !0.14 0.74 0. though the statistical signiﬁcance of these results is weak.11 0.49 0.16 0.M.30 P. Statistics reported are based on the estimates from 20 annual crosssectional GR> GR GR> regressions from 1976 to 1995.04 !0.64 0.71 0.12 !0.01 !0.02 0.00 0.58 0.42 0. Q3 Max . Dechow et al. / Journal of Accounting and Economics 26 (1999) 1–34 A indicate that the models using "1.024 !0.00 0.16 0.24 !0.25 0.05 0.12 0.00 0.018 !0. " and " all identify systematic errors in analysts’ earnings forecasts.043 0.16 0. Sample consists of 50.42 0.08 0.76 0.083 0.07 !0.00 !0.00 0.10 !0.40 0.02 0.004 0.35 0.26 Mean Std.022 !0.036 0.51 0.16 0. Frankel and Lee (1998) also report that analysts’ earnings forecasts contain predictable errors that are not rationally anticipated in stock prices.08 0.02 0.04 0.003 0.45 0.13 0. x " x # .
and b denotes book value of equity at the end of year t.53 0. Regression model estimated is Ret Coeﬀ.16 0.b R R R where x denotes earnings before extraordinary items and discontinued operations for year t.44 0. / Journal of Accounting and Economics 26 (1999) 1–34 Table 8 (continued) 31 Panel B Analysis of the relation between stock returns in the year following the release of analysts’ consensus forecasts and forecasts of abnormal earnings that ignore the other information in analysts’ consensus earnings forecasts.27 0.11 !0. r denotes R the discount rate (assumed to be 12%). Statistics reported are based on the estimates from 20 annual crosssectional regressions from 1976 to 1995.037 0.88 0.01 0.01 0. error 0.53 0. Q3 Max Std.46 0. is the predicted value of from the regression model speciﬁed in Table 2 and estimated using all historically available data from 1950 through the forecast year.036 0.52 0. Stock returns are equalweighted buy—hold returns over the 12 months GR> GR GR> beginning three months after the end of the ﬁscal year from which the historical forecast data are obtained.096 0.04 Notes: Abnormal earnings for year t is deﬁned as x "x !r.11 !0.11 !0.036 0. Q1 Med.28 0. All variables are measured on a pershare basis. Sample consists of 50.15 0. R The consensus analyst forecast of abnormal earnings for the next period is deﬁned as f "f !r. .11 !0.07 0.133 observations from 1976 to 1995.00 0.27 0.02 0.00 0.00 !0.01 0.001 0.00 0.00 0.15 0.01 0.01 0. buy—hold.05 0.01 0.P.05 !0.01 0. is the ﬁrst order autoregression coeﬃcient for abnormal earnings and is estimated using all historically available data from 1950 through the forecast year in a pooled timeseries crosssectional regression.M.00 0. withdividend stock return over the 12 months beginning R> three months after the end of the ﬁscal year t.76 0. Ret is the equalweighted.00 0.b R R R where f denotes the I/B/E/S consensus forecast of earnings for year t#1 measured in the ﬁrst R month following the announcement of earnings for year t.15 0.08 0.18 0.15 0.00 !0.002 0.56 0.18 !0.65 0.28 0.27 0.01 0.00 !0.01 0.14 !0.36 0.03 0.070 0. The earnings forecasts that ignore the other information are generated by the model x " x # .01 0. "0 R "1 R " R " R Mean " # ( x !f )#e R> R R R> Min.80 0.077 0.73 0.00 0.27 0.10 0.066 0.04 0.05 0.34 0.035 0.002 !0.36 0.05 !0.01 0.06 !0.05 !0.002 0.18 0.14 0.03 0.00 0.15 0. Dechow et al.18 0.
Despite the ambiguous empirical support for the model. For example. Second. the model provides a basic framework upon which subsequent research can build. Third. Dechow et al. We begin by pointing out that existing empirical applications of the residual income valuation model are generally similar to past applications of traditional earnings capitalization models. a simple valuation model that capitalizes analysts’ earnings forecasts in perpetuity is better at explaining contemporaneous stock prices. rather than on forecasting its components.M. Subsequent tests suggest that the superior explanatory power of the simple capitalization model may arise because investors overweight information in analysts’ earnings forecasts and underweight information in current earnings and book value. Kothari and Zimmerman (1995) assume a 100% payout ratio. First. as shown in this paper. Conclusions This paper provides an empirical assessment of the residual income valuation model proposed in Ohlson (1995). . we believe that the model provides a useful framework for empirical research for several reasons. We emphasize the word ‘directly’ in this sentence. Our empirical tests indicate that while the information dynamics are reasonably empirically descriptive. the model highlights the implicit assumptions that previous models make about the relation between current accounting variables and future abnormal earnings. For example. earnings and shortterm forecasts of earnings. Feltham and Ohlson (1995) generalize the model to incorporate growth and accounting conservatism. In doing so. the model provides a unifying framework for a large number of previous ‘ad hoc’ valuation models using book value. Thus. the researcher must focus directly on forecasting future abnormal earnings. Forecasts of the earnings and book value components of abnormal earnings contain a forecast of future dividend payments through the clean surplus relation. Previous valuation models based on the dividenddiscounting model often make unrealistic assumptions about dividend policy. Ohlson’s model illustrates that valuation models focusing directly on forecasting future abnormal earnings avoid having to forecast the timing of future dividend payments. the focus of the model on the relation between current information variables and future abnormal earnings is heuristically appealing. / Journal of Accounting and Economics 26 (1999) 1–34 5.32 P. We argue that the key original empirical implications of Ohlson’s model arise from the information dynamics that describe the formation of abnormal earnings expectations. This simpliﬁcation embodies the notion that dividend policy is irrelevant to the extent that reinvested earnings generate the cost of capital.
S.. K. (forthcoming). Review of Accounting Studies 1. B. Journal of Financial Economics 43. Association between accounting performance measures and stock prices. J. Dechow et al. C. J. Lee... 301—339. S.. 377—400. 19—36. Journal of Financial Economics 43. 1995.P. 1359—1373. 141195. market expectation. 1988. Accounting Review 66. Easton.. 3—28. Contemporary Accounting Research 11. 1998. B. Accounting Review 71. Valuation and clean surplus accounting for operating and ﬁnancial activities. The theory of value and earnings.P. Comparing the accuracy and explainability of dividend.. 1970. P. Kothari... Contemporary Accounting Research 11. 173—202. Contemporary Accounting Research 8.. Journal of Accounting and Economics 17. Fairﬁeld. Accounting valuation. American Economic Review 60.. 1992. Feltham. Relative measurement errors among alternative pension asset and liability measures. 341—372....H. free cash ﬂow and abnormal earnings equity valuation models.. BarYosef. 1997. 1991. Kothari. J.. Yohn.. Journal of Accounting and Economics. Lee.B. Journal of Accounting and Economics 20. 601—617.. Anthony. Oswald.P. earnings and the book value of equity: an empirical investigation of the Ohlson valuation dynamics. Freeman. 155—192. S..C.M. 337—355. The information content of security prices. Penman. 1996.. 207—224. 1991. D. J. 749—761. Harris. Lyon. The Feltham—Ohlson framework: Implications for empiricists.... and crosssectional stock returns.D. S. Frankel R. Journal of Accounting Research 29. Contemporary Accounting Research 11. 1995. Journal of Accounting Research 20. Earnings as an explanatory variable for returns. Swaminathan.S. Barber.L.D. 661—687. R.. D. Ohlson. J.. Barth.. J.. S. S. Journal of Finance 31. L. 1997. 433—463. ﬁnancing constraints and corporate investment.. 1995.M. M.E. B.. Sweeney.D. V.M.J. R. Measuring longhorizon security performance. P. Detecting abnormal longrun stock returns: The empirical power and speciﬁcation of tests statistics... Ramesh.N. Journal of Accounting and Economics 15. T. R.. J. Book rate of return and the prediction of earnings changes.. Peterson. 1976... pp.M. 1980. Fazzari. Hubbard R. Warner. 639—653.A. Earnings. 1998. Further evidence of the timeseries properties of accounting income.A. D. Morse. Beaver. Lambert. Callen. Cornell University. and an introduction to the Ball and Brown Analysis.A.A. Francis.G. 1995. Accounting classiﬁcation and the predictive content of earnings. What is the intrinsic value of the Dow? Working Paper.. J. S. C.J.. 1994.. Ohlson.. Journal of Accounting and Economics 2. University of Chicago. A tutorial on the Ohlson and Feltham/Ohlson models...P. Ohlson. Buckmaster.A. 689—732. J. 1991. Myers C. Cragg. Olsson P.A..L. J. J. Ohlson. 1996. G. W. Zimmerman. R.M. 203—228.. 1992.. 733—747. T. 1997. J. Malkiel. Accounting earnings and future economic rents: an empirical analysis..A. Bernard. Livnat. 1995. 1—19. 1982. Lundholm. Kothari. Expectations and the structure of share prices. Modeling dividends. Brooks. B. .. / Journal of Accounting and Economics 26 (1999) 1–34 33 References Ahmed. book values and dividends in security valuation.H. Working Paper. Brookings Papers on Economic Activity. Contemporary Accounting Research 11. Price and return models. Priceearnings regressions in the presence of prices leading earnings: earnings level versus earnings change speciﬁcations and alternative deﬂators? Journal of Accounting and Economics 15..
. 1996. 1963.. A comparison of dividend. Chicago... Berkeley.. Working Paper. Working paper.M.M. A synthesis of equity valuation techniques and the terminal value calculation for the dividend discounting model. F.. Journal of Portfolio Management 14. history.. 78—82. 55—62. V. S. University of California. R. Ohio. Whitbeck. Cincinnati. Berkeley. / Journal of Accounting and Economics 26 (1999) 1–34 Ohlson. T..G. Industrial Market Structure and Firm Performance. Penman. J. IL.34 P.. . Sloan.T.H.G. Sougiannis.A. 289—315.. 1996.H. S. 1996. P. Investor growth expectations: Analysts vs..H. Working Paper. Columbia University.. K. A new tool in investment decision making. Kisor. 1980. Business analysis and valuation using ﬁnancial statements. Healy. 1988. Palepu. Financial Analysts Journal 19... 1998. Earnings. M.L.. W. book values and dividends in equity valuation: An empirical perspective. University of California. SouthWestern College Publishing Co. Vander Weider. cash ﬂow and earnings approaches to equity valuation. Rand McNally College Publishing Co. Dechow et al. J. Do stock prices fully reﬂect information in accruals and cash ﬂows about future earnings?. Carleton.M. 1997. Penman. Bernard V. Accounting Review 71. Scherer.