Você está na página 1de 35

Accounting Research Center, Booth School of Business, University of Chicago

Accounting Measurement, Price-Earnings Ratio, and the Information Content of Security


Prices
Author(s): Jane A. Ou and Stephen H. Penman
Source: Journal of Accounting Research, Vol. 27, Current Studies on The Information
Content of Accounting Earnings (1989), pp. 111-144
Published by: Blackwell Publishing on behalf of Accounting Research Center, Booth School of
Business, University of Chicago
Stable URL: http://www.jstor.org/stable/2491068
Accessed: 25/06/2010 13:58
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=black.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Accounting Research Center, Booth School of Business, University of Chicago and Blackwell Publishing are
collaborating with JSTOR to digitize, preserve and extend access to Journal of Accounting Research.

http://www.jstor.org

Journal of Accounting Research


Vol. 27 Supplement 1989
Printed in U.S.A.

Accounting Measurement,
Price-Earnings Ratio,
and the Information Content
of Security Prices
JANE

A.

OU*

AND

STEPHEN

H.

PENMANt

1. Introduction
A number of papers (for example, Beaver, Lambert, and Morse
[1980],Beaver,Lambert,and Ryan [1987],Collins,Kothari,and Rayburn
[1987], and Freeman [1987]) have documented that stock prices lead
accounting earnings. These investigations have led to the conclusions
that prices provide information about earnings ahead of time and that
earnings capture events that affect security prices with a lag. The study
of P/E ratios has supportedthose conclusions and also producedadditional insights. Beaver and Morse [1978] have shown that PIE ratios not
only predict future earnings changes but they also identify transitory
aspects of currentearnings.Investors utilize other informationin setting
prices which provides both a prediction of future earnings and an indication of whether currentearnings are representativeof future earnings.
Thus a comparisonof price to earnings in a P/E calculationcan indicate
the extent to which currentearnings are transitory.1
In this paper we show that information in prices that leads (future)
earnings is contained in financial statements. While accrual accounting
* Santa Clara University; tUniversity of California, Berkeley. The comments of Trevor
Harris, Jim Manegold, and Jim Ohlson are appreciated, along with those of participants in
seminars at Berkeley, Columbia University, University of Illinois, University of Minnesota,
University of Rochester, and the Wharton School.
' The ratio has traditionally been interpreted as an earnings capitalization multiplier.
However, earnings capitalization is generally not valid (Ohlson [1989a]).
1ll
Copyright (, Institute of Professional Accounting 1990

112

INFORMATION

CONTENT OF ACCOUNTING EARNINGS:

1989

rules produce an earnings number which reflects the information in stock


prices with a lag, they also produce a large array of additional numbers
presented in the income statement, balance sheet, and statement of
changes in financial position. We demonstrate that certain of these
numbers can be summarized into one measure that predicts future
earnings and also filters out transitory components of current earnings.
Like P/E, this measure provides a correction to current earnings as an
indicator of future earnings because it distinguishes earnings that will be
repeated in the future from those that are due to temporary phenomena.
We compare the ability of prices and appropriate financial statement
variables to predict future earnings. We find that the price changesthat
Beaver, Lambert, and Ryan [1987] and Collins, Kothari, and Rayburn
[1987] indicate lead earnings are poor earnings predictors relative to
predictors based on financial statement information. However, price-toearnings comparisons are found to be relatively good predictors. The fact
that both financial statement measures and P/E ratios can predict future
earnings (and can identify transitory current earnings) demonstrates
that information about earnings which can be identified by the PIE
comparison is contained in financial statements. We show that crosssectional differences in PIE ratios can be explained by information in
financial statements.
These findings lead us to a different characterization of accounting
measurement than found in the recent literature. Beaver, Lambert, and
Morse [1980] interpret their findings that earnings reflect information
in prices with a lag as an indication that accounting earnings measure
"true earnings" (dubbed "permanent earnings") implicit in prices with
error. This is because accounting earnings are perturbed by transitory
influences. While the development of these constructs may be criticized
on other grounds (see Ohlson [1989a; 1989b]), the idea is that there are
some aspects of accounting earnings that capture the information in
prices about future earnings and others, the transitory error component,
that do not. Since prices reflect only the "permanent earnings," price
changes reflect changes in "permanent earnings," that is, accounting
earnings purged of transitory components. As an apparent demonstration
of this, Beaver, Lambert, and Morse [1980] and Beaver, Lambert, and
Ryan [1987] show that leading price changes predict earnings better than
leading accounting earnings.
Our findings contrast with this measurement error interpretation of
accounting measurement. Although the earnings number itself captures
information in prices with a lag, our results show that other aspects of
financial statements reveal some of this information contemporaneously
with prices. Further, we find that the transitory earnings components
that are identified by financial statement information also affect prices
and thus are not value-irrelevant measurement error. That is, price
changes contemporaneous with earnings reflect these transitory earnings,
which is why price changes are poor predictors of future earnings relative

INFORMATION CONTENT OF SECURITY PRICES

113

to financial statement information. Since these transitory components


are valued, price changes contemporaneous with current earnings garble
information in financial statements about future earnings. Income measurement rules, rather than measuring value attributes of firms with error,
appropriately incorporate both value-relevant transitory components and
income components that will perpetuate into the future. In addition,
financial statements provide information that distinguishes transitory
earnings components which, because of their temporary nature, may be
valued differently. As such, the accounting system provides information
that enhances valuation rather than garbling it. As a practical matter,
the paper describes a financial statement analysis that discovers this
information.
In addition to comparing the ability of PIE ratios and accounting
numbers to predict earnings, the paper also compares their ability to
predict stock returns. A number of papers (Basu [1983] and Jaffe, Keim,
and Westerfield [1989], for example) find that P/E ratios can also predict
stock returns. This so-called P/E effect has been viewed as evidence of
market inefficiency. Other evidence of earnings leading prices comes
from the documentation of "postannouncement drifts" in returns found
in Watts [1978], Foster, Olsen, and Shevlin [1984], and Bernard and
Thomas [1989], for example.
The seeming paradox of PIE ratios leading both prices and earnings
may be explained by PIE being a measure of risk which is related to
expected (future) returns. However, Ou and Penman [1989] find that the
accounting numbers that contain similar information about future earnings as a PIE ratio also predict stock returns. This predictability is not
due to these accounting items capturing differential risk attributes that
are identified with PIE ratios. Rather, prices may capture the information
about future earnings that is in financial statements with a lag.
We find that the predictable returns associated with PIE measures
and those associated with accounting earnings predictors are negatively
related. Thus, while PIE ratios and identified accounting numbers provide similar information about future earnings, they provide different
information about future returns.
We propose two possible explanations for this finding. Either PIE and
the financial statement predictions measure aspects of risk that are
negatively correlated, or the predictable returns indicate market mispricing with respect to information sets that are negatively related. The
latter explanation is plausible since both PIE and accounting information
identify transitory earnings that are negatively correlated with future
earnings changes. Thus, predictable returns associated with accounting
earnings predictors indicate that the market underutilizes information
in accounting statements about future earnings, whereas the PIE anomaly indicates that the market underutilizes information in accounting
statements about current earnings. This suggests that it takes time for
the market to appreciate both the information about transitory current

114

JANE A. OU AND STEPHEN

H. PENMAN

earnings and about future earnings and, thus, to reflect the filtering of
these components that is carried out by accountants.
In section 2 we describe the financial statement analysis that produces
the accounting-based measure that identifies transitory earnings and
provides a prediction of future earnings and future returns. The data are
described in section 3, and in section 4 we compare the ability of our
measure to predict earnings relative to predictions based on PIE and
price changes. Section 5 provides results on a comparison of the ability
of the accounting measure and P/E ratios to predict stock returns. A
summary and conclusions appear in section 6.

2. The Accounting Indicator of Future Earnings


The accounting measure that predicts future earnings and identifies
transitory components of current earnings is the product of the financial
statement analysis in Ou and Penman [1989]. Early empirical work on
the time-series behavior of annual earnings has generally supported the
hypothesis that earnings follow a martingale or some similar process.2
This was taken to imply that earnings changes are permanent. However,
Freeman, Ohlson, and Penman [1982] found that, if earnings are compared to book value of equity, the direction of one-year-ahead earnings
is probabilisticly indicated. Ou [forthcoming] and Ou and Penman [1989]
further widen the conditioning information set to include a large array
of accounting variables that are line items in the financial statements,
or ratios of these items, and find that these jointly indicate the extent to
which current earnings are likely to be perpetuated in the next fiscal
year. These attributes are combined into a scalar which we refer to as
Pr. Pr provides an assessment of the probability of a one-year-ahead
earnings increase, given the accounting attributes, and as such ranges
from zero to unity. Values of Pr away from 0.5 indicate the direction of
future earnings (increase or decrease) while those close to 0.5 indicate
that financial statement variables are unable to predict changes in future
earnings.
The Pr indicator is obtained by weighting accounting variables by
coefficients estimated using logit estimation techniques during a prior
estimation period. These weighted accounting variables are summed and
transformed into an estimated probability as follows:
Pri

[I

+ exp(-O'

X),

(1)

where Xi, is the set of accounting variables in the annual financial


2 See, for example, Ball and Watts [1972], Allbrecht, Lookabill, and McKeown [1977],
and Watts and Leftwich [1977]. Note, however, that Brooks and Buckmaster [1976] have
documented that extreme earnings changes tend to be followed by changes in the opposite
direction.

INFORMATION CONTENT OF SECURITY PRICES

115

statements for firm i in fiscal year t and 0 the set of estimated coefficient
weights applied to those variables. Henceforth we often drop the i and t
subscripts as well as the hat (^) from Pit, although it should be remembered that this is an estimate which is undoubtedly measured with error.
To estimate the coefficients, a binary dependent variable specification is
used to indicate an increase or decrease in the following year's earnings
before extraordinary items.3 Table 8 in Appendix A (the same as table 2
in Ou and Penman [1989]) lists the 68 accounting variables examined
and estimated coefficients on each using data from two estimation
periods, 1965-72 and 1973-77, based on univariate logit estimation. The
table also provides a x2statistic and associatedp-value for each estimated
coefficient relative to zero. Using this set of variables, a parsimonious
model was estimated for each of the two periods. These models are shown
in table 9 in Appendix A (the same as table 3 in Ou and Penman [1989]).
Readers requiring a deeper appreciation of the Pr measure are referred
to our earlier paper. In this paper, Pr is contrasted to the annual earnings
number with which it is reported in annual financial statements. Accounting earnings for a period, t, can be represented as the result of the
calculation:
Earningsit = A cashit - capital contributionsit+ cash dividendsit
+ A receivablesit+ A inventoriesit - Apayablesit
+ Aplant assets, net of additions and disposalsit,

(2)

+ etc.,...,
where A indicates changes over period t. This representation of the
periodic earnings calculation comes from the articulation of the income
statement and balance sheet. Earnings are a combination of the (comprehensive set of) financial statement variables on the right-hand side
of (2) into a scalar according to certain rules. Pr in (1) is also a
combination of accounting variables, Xit into a scalar, by the "rules"
contained in the 0 vector. Thus the financial statements can be depicted
as describing certain variables that project into current earnings (in year
t) and certain ones that project into future earnings (in year t + 1). The
ability of Pr to identify transitory components of current earnings will
become clearer as we proceed.
Three points about the Pr summary measure should be noted. First,
the procedures to obtain Pr values were designed to exclude ex post
selection bias and statistical overfitting. Second, model estimates were
based on a pooling of data over firms and over time. If different operating
characteristics predict earnings differentially across firms (in different
'The dependent variable is earnings change minus a drift estimate. This provides
roughly the same number of earnings increases and decreases.

116

JANE A. OU AND STEPHEN H. PENMAN

industries, for example), the Pr measure obtained from our pooled data
measures that prediction with error. Third, the estimation model employs
only a binary specification of future earnings changes, thus ignoring
information about the magnitude of future earnings changes. These last
two points suggest that calculated Pr values are noisy indicators of future
earnings. If this is so, our results here should be understated.

3. Data
Using the coefficient estimates given in table 9, Pr values were estimated from financial statements published for the fiscal years 1973 to
1983. For years 1973-77, Pr values were calculated using coefficient
values estimated during the 1965-72 period; those for the years 1978-83
were based on updated coefficient estimates from the 1973-77 estimation
period. For each of the 11 years, financial statement data were obtained
from the 1984 Compustat annual report files and the 1984 Compustat
research file. The suppliers of Compustat services claim that these files
are comprehensive with respect to firms whose common stock is traded
on the NYSE or AMEX, and in addition include some utilities and
financial firms traded on other exchanges.
Table 1 gives the number of firms available for the analysis in each
year. For some of these, the accounting variables necessary to estimate
Pr were not available on the Compustat files. The number of firms for
which a Pr value was obtained is given in the third column of the table.
The industry composition of this set is similar to that on the Compustat
files, except that there are few electric and gas utilities and financial
firms. These firms typically do not report the accounting items identified
in the Pr models. The fourth column of the table shows the number of
firms for which earnings data were available in the year following the
financial statements from which Pr is drawn. These are the firms used
in the prediction tests using Pr. The number is less than that in the third
column because some firms ceased to exist prior to the annual earnings
reporting date for the following year.
For each firm for which a Pr value was available, we calculated a PIE
ratio using earnings per share (EPS) reported in financial statements
for a given year and per-share price three months after fiscal year-end.4
We assumed this price reflected the information which is used to estimate
Pr. That is, the information would have appeared in financial statements
assumed to have been published within three months after fiscal yearend. The number of firms for which Pr and PIE ratios could be calculated
4Prices (and shares outstanding for later calculations) were obtained from the Compustat
quarterly file or the CRSP NYSE monthly master file at a point three months after fiscal
year-end. For those (relatively few) AMEX firms on neither of these files, these data were
obtained from the Compustat annual file at the nearest December 31 to fiscal year-end.
Results are not sensitive to the exclusion of firms with prices obtained from the Compustat
annual file.

INFORMATION CONTENT OF SECURITY PRICES

117

AQ

M"o
6?
bj

ce

?
t-

00

00

t-t
00

M 0

CZ CZ

Co

t-

<

Co

LO

LO

~~~~~
~~~
~~
~~
~Q~~
~~
~
4
C2
XS
0

u:

cq~~~~~~~~~~~~~Ci

-4~~~~~~~~~~~-D

CZ

-d

0o
C
o
4~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1

<

~~~~~~o
A

0~~~~~~~~~~~~
CZ.
CZ
9
320

ce

a3

U:

CZ

~~~~s
a

+2

>s

t s

u:

oo

oo

U:

= = w
U N% s % % > > Q Q
I~~~~~~~~~~~~~~~~~~~~~~~~~:
F4

>

;3

cq~~~~~~~~~~
~~~~~~~~~~~~~~~~~~~~~

a~~~o
<
=;

c;

=~~~~~~~~~~~-

U~~~~~~~~~Ao U
U

P
O~~~~~~~~~~~~~~~~0
<

Uo._~~~~~~C
CZ2

CZ
L4-D

4
cq
1

O
c41

CZ

_t
s

00<

t-

_tt m 00

00

"I

00 -t

LO

00 - 00
ssss

HH

t-

LO

LO

00
L

M
CZ

C
b

1:g4
1e-'
> CZ

W
x

W;

118

JANE A. OU AND STEPHEN

H. PENMAN

appears in the second-to-last column of table 1; the last column shows


the number of this set with available year-ahead earnings.

4. Pr, PIE, and Current and Future Earnings


In this section we summarize observed relationships between Pr and
PIE and indicate the ability of each to identify transitory earnings. We
also compare the ability of Pr and price to predict future earnings.
The insights about prices leading earnings have come from observations of the mean reversion behavior of P/E ratios. High (low) P/E ratios
tend to be followed by lower (higher) P/E ratios in later years, as
documented in Beaver and Morse [1978]. The PIE ratio can be interpreted as a comparison of two information sets, the information about
current and future earnings that is summarized in price (the numerator)
and the information in current earnings alone (the denominator). Accordingly, the mean-reverting behavior of PIE ratios has been interpreted
as prices providing indications of future earnings relative to current
earnings. High (low) price-earnings ratios indicate that earnings will be
higher (lower) in the future. When these higher or lower subsequent
earnings are ultimately recorded, observed PIE ratios revert toward the
mean. In support of this, Beaver and Morse [1978] find that PIE ratios
are positively correlated with subsequent earnings changes. They also
find that PIE ratios are negatively correlated with current earnings
changes (incorporated in the denominator of the PIE calculation). Thus,
PIE ratios indicate reversals in the direction of earnings changes. This
is the so-called Molodovsky effect.5 Earnings changes that prices indicate
will be reversed in the future are designated "transitory earnings."
This phenomenon is demonstrated for our sample in panel A of table
2. The estimated correlations were obtained from portfolio values based
on an assignment of firms to ten equal-sized portfolios after ranking
them in each year of the 11-year sample period on decreasing values of
EIP.6 Panel A presents, for each year, estimated rank correlations
between portfolio values of EIP and median percentage earnings changes
for the portfolios in the current year (year 0) and years following.7 The
presentation is very similar to that for an earlier period in Beaver and
Morse [1978]. Like them, the table shows that EIP ratios negatively
rank-order earnings changes as far as three or four years ahead. In
'The ability of the P/E ratio to identify transitory deviations in current earnings from
long-run earnings is characterized in Molodovsky [1953] and has been since referred to as
the "Molodovsky effect." See Cottle, Murray, and Block [1988, p. 348].
6 We rank an E/P rather than a P/E to provide continuity through zero. Also, since the
month of a given year's fiscal year-end may differ over firms, these rankings are not based
on values determined at the same point in time. The cross-sectional comparisons that
follow are, however, similar when only firms with (the same) December 31 fiscal year-end
are analyzed.
'Percentage earnings changes are defined as the change in EPS divided by the absolute
value of the prior year's EPS. Earnings are before extraordinary items.

INFORMATION

CONTENT OF SECURITY PRICES

119

contrast, E/P ratios are positively correlated with current earnings


changes. The price comparison to current earnings identifies transitory
earnings that reflect temporary phenomena: E/P ratios indicate current
earnings that are lower (higher) than in the past but will likely revert to
past levels in the future.
Panel B of table 2 provides a similar analysis, but for firms grouped
into portfolios on Pr values. In each of the 11 years in the sample period,
firms were assigned to one of ten Pr portfolios as follows. Firms with .9
< Pr < 1.0 were assigned to portfolio 1, firms with .8 < Pr < .9 were
assigned to portfolio 2, and so on, so that firms with Pr c .1 were assigned
to portfolio 10. Panel B indicates that Pr performs the same role of
identifying reversals in current earnings changes. (The signs of the
correlations differ over the two panels of the table because the portfolios
in panel A were formed on E/P rather than P/E.) Pr is not only an
earnings predictor (as constructed). It also captures temporary earnings
changes that subsequently reverse, on average. Pr values away from .5
indicate that current earnings are temporarily up or down, while Pr
values close to .5 indicate current earnings levels that will persist in the
future.
The observation that E/P and Pr capture common information is
reinforced by estimation of the following cross-sectional regression equation for each of the 11 years of the sample period:
(E/P)it = oc + 3Prit + vit,
where the i subscripts indicate firms and the t subscripts indicate that
the E/P and Pr measures should be observed at the same point in time,
three months after fiscal year-end. The mean value of 3 over 11 estimates
from all of the sample years is -.44, with a t-statistic (calculated from
the time series of coefficient estimates)8 of -9.21. A negative d3 was
observed in all 11 years, with a range of -.55 to -.34. The mean R' over
all the years was .22, and the R2 ranged from .14 to .36. Thus, on average,
22% of the cross-sectional variation in E/P is explained by the accounting-based Pr measure.
An analysis of the relative ability of Pr and price variables to predict
one-year-ahead earnings-per-share changes, AEPSt+, is reported in table
3. The purpose of the analysis is, first, to demonstrate the lower-bound
ability of financial statements to predict earnings relative to price variables and, second, to show that leading price changes are relatively poor
predictors. We deflate AEPSt+l by the previous year's earnings per share.
This makes our analysis comparable to those in Beaver, Lambert, and
Morse [1980], Beaver, Lambert, and Ryan [1987], and Collins, Kothari,
and Rayburn [1987] who use percentage earnings changes as the target
variable.
3
8The observed first-order serial correlation in
is .035. Hence, in calculating the tstatistic, the f were treated as independent over time.

120

H. PENMAN

JANE A. OU AND STEPHEN

Cl

Cl

0~

COD~~~~~~~~~~~~~~~~C

Cl

0ii

.E~~~~~~~~~~~~~~~~L
C L
Ll

CZ

Co

Y
g~~~~~~~~~~
Cq t o

. >c.
t
g~~~~~C

(~C

Cl

~N~

42

fO
C. L
Lo
t M
C M

~ ~ ~ ~ ~

X4

<

..

>;

co

CoC

CZ

Er' ~ ~~
(D~~~~~~~~~~~~~~~~
~~
0
C9

R3

C
t-

m Ce LO Co

Co

m00

t-

M
LO C

0L

00

0Loo

1-1-1

.?

.0

C s

00 C

CJ2

C9 C9 Lo

M .
.1.

e~LL 00O 00 C0

C
4-i CZ

INFORMATION

CONTENT OF SECURITY PRICES


.2

Cl

Cl

ho

F
.-4
CYD -

CZ

~~~

Cl2

.2

-6

Cl!

C~

cl~~~

(D

;,_

to

CZ2 4--0
C'i'
0~C
0~~~~~~~~~~~~~

LO

o00
II

121

pq

~~0
~~~~~~~~~~~C~~~~~~07l~~~~~~~~~~~
C

4LO

Cl~~~~~~~C

~ ~

0
CZ
(D

~~~

o14a

C
~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
~ Y 0.0~ F-4
CZ0I'~

> (n
;4
0
(
t
LOCOC~~~~~~~~~~~~~~tOLOLO~~~~~~~~~~~~~~l
0
CZ
0
(D
ClL,tC"t-,
-Ct

~~~~~~~~~~~~~~~~C

*
0

000

t0tCO

L-l0)

CO

40

42

0)C~~~~~t0~~~~-lCO'l'lL0L0
Lo

CZ

00w 00

M L

CZ0

00 Cq c.0Lo Lo Cq

0~,
-

CCZ~

-O)C0)CO)

"t

0
kz
~~~~~~~~~~co)
'>2
Cl~~~~~~~~~~~~(

"t

0-

2C3

m0.0

C. "'

CZ)

CZ
0

co

W0

42.

(nC
0

'o

(
Cl)P (n 4-

C)

co ;-, 0
W

00

70

CZ

CZ+'.4

C0

0
co

4Cl

0P;

-,~W4COO0

,'

Ql

CZ

Co Q (n 0

122

JANE A. OU AND STEPHEN

H. PENMAN

4-a
-,

CO

CCCCCt

O.

lC.

~C1O.

1O. O.O

^.CZ
C13

4- .
C)

OO

(O

Ce

cq

~C

CID

00

CZC
O C
0CZ

CO

CCZ

4-0

CZ

.w

CO

OCC?<

0)

CC

Oo

Ez

>

0LC

>

Xt
2;

1<n
IL-4

H
H

O
>

I~sx o.

FN

<

ce

.C *

Cs

Cs

>H

Q
4

e
e

u:

cs

00

oo

CC~~=cO

>

CZ
U

~~0C
~
D

IqIC

~~<LoC

C-)

CZ
I-

E=

4
+~~~~~~~~~~~~~~~~~~Z
t

= ==

0EC
4

C?C=CZ

CZ

0 ..2~~~~~~~~~~~~~~~~~~~~~~

0=
4Z4E
o ~CZ
Q
-eQ
~~~~-o

CC

CC-

0)

CC

~~~~~~~~
~~~~~~~~~CZ.C0(
~ C

CO

=m

.C.*_CC

ce

9~~~~~~~~~~~~~~~~~~~~~~
>~

CZ
CC

EC

-~~~~~~~~~~~~~~~~~~~~~~~
CZ~~~~~~
*wrH~~~~~~~~~~~~~~~~~~~
.

co

CCE

CCD
CC
4 CIS

CC0C

C~~~~4CC CISI
CO ~~~~~~~~~~~~
~ ~ ~ ~~~~~
~~~~
00 CO C-~~~~~~~~~~~I'lW
C-I (N ~~~~~~~~~~~~~~~~
~~~~~~~~
~
cq~~~~~~~~~.

CID~~~~~~~~~
O

CC

CC C .

CCCC.0
U

.4

nO

C~~~~~~~~~~~~~~D
oC'sCC

X4

COCCCC)C
4~~~~~~~~0,

'ACCC.

CZ

4-C

CC

.CZ

CZ C.

COZ 0C -4
CZC

200

CZ

INFORMATION CONTENT OF SECURITY PRICES

123

Table 3 shows the mean coefficient estimates and R2 values from the
estimation of the cross-sectional regressions of percentage EPS changes
in year t + 1 on alternative predictors in each of the 11 years of the
sample period. The reported t-statistics on the mean coefficient estimates
are calculated from the time series of coefficient estimates. The table
also indicates the number of positive coefficient estimates observed in
the 11 years and the total number of firm-year observations in all 11
years.9
The mean coefficient estimates and R2 values in the first line of table
3 illustrate the well-documented inability of earnings changes to predict
future earnings changes. This is the Ball and Watts [1972] result. The
next three lines reflect the ability of leading price-change variables to
forecast earnings. CRit12is the cumulative monthly return for firm i over
the 12 months prior to the end of three months after fiscal year t. This
period excludes earnings announcements for year t + 1 but captures all
earnings reports for year t plus the annual reports that reveal Pr. CUit12
is the same calculation as CRit12but with market-adjusted returns.10
These predictors are similar to those used in Beaver, Lambert, and Ryan
[1987] and Collins, Kothari, and Rayburn [1987].11 Relative R2 values
indicate that these lagged price-change measures do not explain oneyear-ahead earnings changes as well as E/P and Pr.12 Extending the
prior return period to 24 months (using CUit24)only adds noise to the
prediction.
The last two lines of table 3 summarize the ability of prior price
changes (as measured by CUit12) to explain one-year-ahead earnings
changes for cases where the financial statements give little indication of
the future earnings (Pr between .4 and .6), and cases where they do (Pr
values greater than or equal to .6 or less than .4). CUit12is a relatively
good predictor of the future earnings changes for .4 < Pr c .6 but not so
for Pr > .6 and Pr c .4. The reasons for this will become apparent as we
proceed.
Table 4 assesses the incremental explanatory power of each predictor
conditional on alternative predictors. The dependent variable is the
9 These numbers may not tally to totals in table 1 because of rejection of outliers or
unavailability of returns data for the 12-month period. Absolute values of standardized
earnings changes greater than 2.0 and absolute values of (E/P)i, greater than 1.0 were
excluded. Results are not sensitive to cutoff points within a reasonable range of these
figures, including a 1 3.0 I cutoff for standardized earnings changes used in other papers.
10The market adjustment makes no difference in cross-sectional regressions where fiscal
years are aligned, of course. However, cross-sectional regressions are estimated for each
year using firms with different fiscal year-ends.
" Note that the mean estimated coefficients on the price-change variables are close to
that observed by Beaver, Lambert, and Ryan [1987, table 4], assuming successive price
changes are uncorrelated.
12 Strictly speaking, the R2 values are not comparable because of slightly differing
numbers of observations in each regression estimate. However, these inferences hold when
only the same stocks are included in the comparisons.

124

H. PENMAN

JANE A. OU AND STEPHEN

Cnl

= N

o Lo "I

ci

q q)

L
L

"o

Cd

0)
00
+

r.
0(

qqq

P4

C!

C!

CO

cq

C_

t: Cf)

06

Lo ;

0)
0)
0)

Oq W CS
cq L uO

ai
;

m m "It
u

+~~~~~~V

d
C:
c6

c6

m Il

CC

LC

)0 C

!~~~

It
O

e~~~~~~~~coO oO
4~!rE0 ~ ~~0

0C
C)C C)qqqqqqqq

0
co
Ud

m ceCO s 'I
O0
t_ c6 cs6 Ue
C0e
cQ

~~~~~~~~~I

,9
C

C<co
Q1

co

u
0
ulQ

u:

III

+0CO1)

0)s
QCOD

tT

O0

0C

~~ ~

I0)

C*0

C4)

CI~~~~~~~~
0
:t "l~!

.':t

INFORMATION CONTENT OF SECURITY PRICES

125

portion of one-year-ahead earnings changes that is orthogonal to a given


predictor. Xit denotes the explanatory variable in the first-stage regression
and X'it that in the second-stage regression, as indicated at the head of
the table. Pr and E/P have explanatory power beyond that in the pricechange variable, CUa12. But CUa12 also has some explanatory power
incrementally to E/P and Pr. Indeed, all the predictors appear to capture
some aspect of the available information about future earnings that the
others do not. However, the explanatory power of CUia12given Pr is
largely restricted to cases where .4 < Pr c .6, that is, where the financial
statements do not give a strong indication of the direction of future
earnings. Similarly, while E/P and Pr both provide incremental predictive ability over each other, Pr does so only in cases where Pr > .6 or Pr
c .4. In contrast, E/P performs well incrementally in both cases where
Pr> .6 or Pr c .4. and .4 < Pr c .6.
The superiority of Pr values away from .5 as predictors relative to
prior price changes is explained by an inspection of table 5. For each Pr
portfolio described earlier, the table presents (from a pooling over years)
the number of stocks in each portfolio,13 the mean Pr value (a result of
portfolio construction procedures), mean relative E/P, mean relative
percentage change in earnings reported concurrently with Pr, and oneyear-ahead percentage earnings changes. The mean relative values are
means over the 11 years of the portfolio median minus yearly grand
medians. This calculation produces an appropriate pooled value in the
presence of year-to-year median shifts. These values reflect the observations made above regarding relationships among Pr, PIE, and current
and future earnings. The comparisons of current and future earnings
changes highlight Pr as a variable that identifies the earnings reversals
that are indicative of transitory earnings. Note that although Pr was
constructed to identify the sign of future earnings changes, it also
discriminates among the magnitudes of those changes. Similarly, it
discriminates, negatively, among magnitudes of current earnings changes.
With the exception of the last column, the remainder of the table
describes the pricing of Pr and current earnings in the market. For each
set of returns,14we report the return to a zero-investment position, which
we refer to as the hedge portfolio return. This position involves going
short in stocks with Pr c .4 at the beginning of the relevant period
indicated by the column head and investing the proceeds in stocks with
Pr > .6. As indicated in the footnotes to the table, the return to this
position is the mean of market-adjusted returns to stocks on the long
side minus the mean of market-adjusted returns to stocks on the short
13 The total number of stocks here (17,757) is less than the 19,579 in the third column
of table 1 because returns data were not available for some firms. Stocks are included in
table 5 if they had returns data at the end of three months after fiscal year-end.
14 The table presents market-adjusted returns without any further adjustment for risk.
A full analysis of risk of Pr portfolios (which does not change the inferences here) is
contained in Ou and Penman [1989]. Similar results are obtained with size-adjusted returns.

126

H. PENMAN

JANE A. OU AND STEPHEN

*0

bBc

4-)

0~~~~~~~~~~~~

CZ
O0CZ
I
~~000VI

CZ

-~~~~~~~~~~

O +L

1I
0

o0
~~~~~~~~~~~~~~~~~~~~
&~~~~~~.
?

.
4-j4)

0
0

40+
0

~~~~~~~~~~~~~~~~~~
00.

000
-CT

-~

-CZ

I
'O

I0

CZ

?0CZ
~~~~~~~0
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Oo
0

>
l - 04-??

00

O,00

0
0

0~~~~~~~~~~~~~l
0
0

r-lr-lC

10

C,~~~~~~~~~~~~~~~~~~
00~~~~~~~~~~~0

~~~~~~~~~~~~~~~~~~~~~
~~~~~~~~d

0~~~
Z0

00~~~~~~~~~~~~~~~
-,,,C
~~~~~~~~~~~~~~~~~~~~~
0

Q)

0k

00~~~~~~~~~~~~~~i

CZ

L0

0
co

c0

co
i2-bn

o0

0)

0.

bn

00"lCeC

0?0
00

0a4
O

~0

~o~i'cooEn

0
?

~ ~

~~0
~

(n

CZ

CZCn

?0

0
~ '
u~~~~~~~~~'-

>k

->

CZ

CZ~0
0
~~~~~~~~~~~~~~~~~~

0'

0~0

~~~~~~~~~~-~~
0
0)
0~~~~~~~~~~~~~~~~~~
~~~~~~ 00 ~~~~~~~~~~~~~~~~~~~~~~~~
.0

'CC-)

0~~~~~~~~~~~~~~~~~
'~~~~~~~~~~~~~~~~~~~
0

l0

>1

O'C

~ ~

INFORMATION CONTENT OF SECURITY PRICES

127

side. The table indicates that just as Pr is negatively correlated with


current earnings changes, it is also negatively correlated with returns
over the 12 months during which these earnings are published (in four
quarterly installments), FYE - 9 to FYE + 3 months. (FYE indicates
fiscal year-end.) This return is CUWi2. We see from the table that the
negative relationship between mean Pr (in the third column) and mean
CUa12for portfolios is almost monotonic. This is based on observations
pooled over years. The mean estimated rank correlations over years of
portfolio mean Pr values and mean CUWt12
is -.72 (t = -7.43). Thus,
while the accounting statements provide indications of future earnings
increases (for high Pr) or decreases (for low Pr), contemporaneous prices
move in the opposite direction. The inference is that these price changes
are reflecting the current earnings which Pr has filtered out as transitory.
Even though these transitory earnings changes will probably be reversed
(for Pr values away from .5), they are interpreted by the market as valuerelevant (that is, they are thus priced).15
This is the reason prior price changes measured by CUit12and CRt12
are not good predictors of future earnings changes. Whereas Pr and
E/P identify transitory components of current earnings, contemporaneous price changes incorporate them. For Pr values close to .5, transitory
components are small, in which case current earnings and the price
changes contemporaneous with current earnings are relatively good predictors of future earnings (as indicated in table 3). This is not so for
more extreme Pr values. Table 5 indicates that these price changes are
in the opposite direction to the forecasted (and, on average, actual) oneyear-ahead earnings changes.
To reinforce this point, we observed that the sign of the CUit12variable
in tables 3 and 4 for .4 < Pr < .6 correctly predicted the sign of the oneyear-ahead earnings changes 57.8% of the time over all observations in
the sample. However, for Pr> .8 or Pr < .2, where the direction of future
earnings was strongly indicated in the financial statements, the prediction success rate for CUit12was only 43.8%. For Pr > .8 and Pr < .2,
estimates of b for the regression equation at the top of table 3 with CUit12
as the predictor variable were .03 (t = .33). To provide further insights,
a two-stage regression was estimated for each of the 11 years, with the
first stage being:
CUit12= yo + yl AEPSit/EPSit_1

Wit,

and the second stage:

AEPSit+1/EPSit= zy'o + y'iwit + Xit+i.


The mean value of 'i over the 11 years was .23 (t = 8.02) and the mean
R2 was .14. These values are similar to those observed by Beaver,
15 An alternative explanation is that prices overreact to current earnings. This explanation is rejected in Ohlson and Penman [1989].

128

JANE A. OU AND STEPHEN

H. PENMAN

Lambert, and Ryan [1987] for their ungrouped data. The return variable,
wit, takes out the portion of contemporaneous returns that is explained
by contemporaneous earnings. The mean value of "'1 was .34 (t = 6.80),
and the mean R2 for the second equation was .04. This R2 is twice that
reported for CUit"2 in table 3. For .4 < Pr < .6, the sign of witpredicted
the sign of one-year-ahead earnings changes 56.5% of the time, which is
about the same as for CUit2. However, for firms with Pr > .8 or Pr c .2,
this figure was 60.9%, compared to a 43.8% success figure for CUit12 for
these Pr values. (The success rate for Pr was 73.1%.) Purging the
contemporaneous return of the implications of current earnings improves
its forecasting ability for cases where high transitory components are
identified, that is, where earnings reversals are more likely. This contrasts
to the perspective (in Beaver, Lambert, and Ryan [1987], for example)
where price changes capture "permanent earnings" purged of transitory
components. Note, further, that grouping on price changes (as in Beaver,
Lambert, and Morse [1980]) will involve grouping on the price effects of
transitory earnings.
The returns over FYE - 9 to FYE + 3 in table 5 clearly indicate that,
over the 12-month period prior to FYE + 3, the effects of current earnings
with its transitory components seem to swamp any effects of the Pr
values on stock prices. However, the full financial statements containing
much of the information in Pr are not released until after FYE. The
eighth and ninth columns of table 5 also summarize returns over the
three months from FYE + 1 to FYE + 3. Since Pr and current earnings
changes are negatively correlated and since current earnings changes are
priced, there is a control for the direction of these earnings changes (with
predicted earnings drift subtracted). Returns over the 3-month period
are positively related to the Pr information occurring in annual reports
during the period. It appears then that the market reflects at least some
of the Pr information. This is the result observed by Ou [forthcoming].
Thus, the Pr information is reflected in P/E ratios at FYE + 3, contributing to the observed correlation between Pr and PIE.
The preceding analysis supports the following conclusions. Some of
the information about future earnings and transitory components of
current earnings that is implicit in prices is also captured in financial
statements. While that information is apparently incorporated in accounting income with a lag, it is available contemporaneously with prices
in other numbers produced by the accounting process. In essence, the
accounting statements contain two earnings numbers: the current earnings number which is a projection from certain financial statement
variables, as in (2), and a future earnings indicator which is also a
projection from financial statement variables, as in (1). The current
earnings number contains transitory components which make it a poor
indicator of future earnings. However, these transitory components are
value-relevant and are appropriately captured by accounting calculations.
In addition, the accounting process produces information that identifies

INFORMATION CONTENT OF SECURITY PRICES

129

these transitory components and provides projections to future earnings.


In sum, the valuation of future earnings and the identification of transitory current earnings are contained in both prices and in financial
statements.

5. Pr, P/E, and Future Price Changes


Ou and Penman [1989] found that, in addition to predicting earnings,
the Pr measure also predicts stock returns. They discovered that a
significant portion of the return to a Ball and Brown [1968] hypothetical
investment strategy based on perfect foreknowledge of earnings can be
predicted by the Pr measure based on "publicly available" financial
statement information about future earnings. This indicates that accounting information about future earnings is incorporated into prices
with a lag.
Indeed, a number of studies (for example, Basu [1983] and Jaffe, Keim,
and Westerfield [1989]) indicate that the earnings predictor, PIE, also
predicts stock returns, after controlling for beta, "size," and "January"
effects in returns. In this section, we compare the ability of Pr and PIE
to predict stock returns. Our purpose is to see whether the predicted
returns are capturing similar phenomena. In the previous section we
proceeded as if prices contain all relevant information about earnings
and attempted to discover the extent to which that information was
captured in the accounting statements. Here we concede the possibility
of market inefficiencyand investigate relationships between Pr, PIE, and
future stock returns.
The last column of table 5 summarizes cumulative returns for Pr
portfolios over the 24 months following FYE + 3.16 Pr predicts future
stock returns, suggesting that prices do not capture all the information
in the financial statements about future earnings. This result, summarized in the return to the zero net investment hedge portfolio, is the
finding of Ou and Penman [1989].17 This cannot be explained by differences in beta risk, standard deviation of return, firm size, leverage, or
market-to-book premiums. Most telling, however, are the positive associations of Pr and price changes over the period during which Pr is
reported in financial statements, FYE + 1 to FYE + 3. The market
16 Again we assume that the accounting items on which Pr is based were published in
annual reports during the three months after FYE.
17 The hedge portfolio return summarizes the future returns for all stocks in the sample
with Pr > .6 or Pr c .4. However, it does not represent the return to investment strategies
that could have been carried out at the time because security weights in portfolios cannot
be calculated until the end of the sample period. To calculate a return to such a strategy,
we calculated the 24-month return to investing in the hedge position at each April, in the
sample period for firms with December 31 fiscal year-ends only. The mean 24-month return
to these hedge positions over the 11 years was .1256. The relative frequency of observing
this figure or greater in 2,000 random replications of the strategy was .000. Other issues
regarding the calculation of returns are covered in Ou and Penman [1989].

130

H. PENMAN

JANE A. OU AND STEPHEN

CO
CO

D
=0.00
mE
;J0

+
o

L-

C<)/>
q
00

cq

C-4

Cl

U 5m

> > 4
000000000
0 0 CD. . O. O. O. O. O. O. O. O. O._
) ;
C

coo
.0

a) Cn 4- -4

%nn

nte

Cf)

E -Z o ?-W M
--~~~~
I X coX0 U D t0 t O
e IL
D Q,

CZ

C:U

w~o=

EH,

> C) CZ4
(n a4
cn

.!

Ill

IDnH

ssnn

s>c

~ca

CZ
0

-1

kmM,C~

> .0

e~O?

La)Cfm

otr

T0t-C

.-

O."4

0~~~~~~~~~~~~
Co~~~~~~~~~~~~~~~C
a)~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

INFORMATION CONTENT OF SECURITY PRICES

131

interprets Pr when it is reported as an earnings predictor not as a risk


attribute. If Pr were a risk measure, we should observe a negative
correlation between Pr and returns during the announcement period.
Table 6 summarizes returns associated with different levels of E/P.
Nine equal-sized portfolios were formed from a ranking of firms with
positive E/P ratios in each year, with firms with negative E/P (loss
firms) placed in portfolio 10.18 This is done not only because it is
conventional in E/P studies but also to discover if there is anything
distinctive about negative E/P ratios compared to low positive E/P
ratios. For each portfolio, the table summarizes (in a similar format to
table 5) values of Pr, contemporaneous earnings changes, one-year-ahead
earnings changes, returns prior to FYE + 3 and returns following FYE
+ 3.
The relationship between E/P and Pr and between E/P and current
and future earnings changes is again evident from this table. E/P, like
Pr, identifies transitory earnings components. However, whereas the
returns over FYE - 9 to FYE + 3 for Pr portfolios in table 5 are
negatively correlated with Pr, this is not the case for P/E ratios (the
inverse of E/P) in table 6. In table 5, the seventh column shows that
high (low) Pr values are associated with relatively low (high) returns
over this interval. In contrast, the sixth column of table 6 shows that,
for nonnegative earnings (portfolios 1-9), P/E ratios are positively associated with returns over the same period. Further, the returns for Pr
portfolios are consistent with the direction of current earnings changes,
whereas those for the E/P groups are in the opposite direction (but in
the same direction as the one-year-ahead earnings changes). This indicates that PIE ratios incorporate information about future earnings that
is not incorporated in Pr (or current earnings).
The last column in table 6 summarizes returns over the 24 months
following FYE + 3. The return to a zero-investment position, calculated
in the same way as that in the final column of table 5, is shown at the
bottom of the last column. These stock returns are size-adjusted because
of evidence that PIE effects in part reflect size-effects in stock returns.
For each stock in each month, these returns were calculated as the
difference between its monthly return and that of a size portfolio in
which it was a member. It is clear that E/P ranks these returns, and the
zero-investment position based on E/P yields significant returns. This is
the so-called PIE effect.'9 A similar result is observed for market-adjusted
18 The number of stocks in this analysis is less than that in column 5 of table 1 because
only firms with prices available at FYE + 3 were included. See n. 4.
" The hedge portfolio return is based on all stocks in the sample, but it is not an
implementable strategy. The mean 24-month size-adjusted return from the zero-investment
strategy executed at each April 1 in the sample period using firms with fiscal years ending
December 31 was .0699. The relative frequency in observing this return in a random
strategy repeated 2,000 times was .000. Banz and Breen [1986] claim that the PIE effect
documented by others is due to biases in sampling procedures. No such biases exist here.
The analysis in Jaffe, Keim, and Westerfield [1989] indicates that the Banz and Breen
findings occur only in their limited sample period.

132

JANE A. OU AND STEPHEN

H. PENMAN

returns. These predictable returns are negatively correlated with those


over FYE - 9 to FYE + 3, indicating price reversals associated with
PIE ratios.
To what extent are the "P/E effect" and the documented returns to
Pr positions in table 5 capturing the same phenomenon? One aspect of
the data which dismisses this question immediately is the basic observation of the paper that E/P and Pr are negatively correlated. The
direction of the returns to the Pr strategy and the E/P strategy summarized in tables 5 and 6 respectively indicates that the returns to these
strategies are not closely related. High PIE ratios are associated with
high Pr values (reflecting their abilities as earnings predictors), but the
returns to high PIE positions are negative while the returns to high Pr
positions are positive. If these returns reflect market inefficiency, they
reflect different aspects of market inefficiency; similarly, if they reflect
the pricing of risk attributes, they do so for different risk attributes.
Table 7 makes this point clearer. This table presents size-adjusted
returns over the 24 months following FYE + 3 for five Pr portfolios
within each of the ten E/P groups, thereby allowing an assessment of
the Pr strategy holding E/P constant (across each row) and an assessment of the E/P strategy for a given level of Pr (down each column).
Returns to the hedge positions described in tables 5 and 6 are also shown
for each column and each row. Since the number of firms in some Pr
groups for some E/P portfolios is small, the results are summarized in
the lower panel of the table for three E/P groups formed from nine
positive E/P portfolios.
It is clear from reading across rows that, with the exception of the
negative E/P portfolio (portfolio 10), Pr discriminates incrementally
among realized future returns for each E/P level. The reported Spearman
rank correlations between E/P and Pr within each E/P portfolio and the
mean differences in median E/P over the two sides of the Pr hedge
positions indicate that these returns are not capturing residual E/P
effects within E/P portfolios. E/P, as we have observed, is an earnings
predictor. The ability of the Pr earnings predictor to predict stock returns
for a given level of E/P indicates that it captures future earnings information or other value-relevant attributes that are not captured in the PI
E predictor. However, the inferred failure of PIE to reflect all predictive
information does not explain the PIE effect. Reading down columns in
table 7 we can observe an EIP effect holding Pr constant. For .4 < Pr c
.6, where there is neutral future earnings information in Pr, the return
to the EIP strategy is evident.
Given the negative correlation between returns to the EIP and Pr
strategies, the PIE may be due to unspecified risk differences across
firms or the market's failure to incorporate information that is negatively
correlated with predictive earnings information in Pr. Since we know
from correlations reported earlier that current earnings changes are
negatively correlated with Pr and PIE, and we find it difficult to conceive

INFORMATION CONTENT OF SECURITY PRICES

133

of other value-relevant predictive information that is negatively correlated with Pr (and PIE), we conjecture that the PIE effect captures
underreaction to current earnings, the denominator of the PIE calculation. Indeed, Kim [1987] shows that E/P ratios are positively correlated
with "unexpected (current) earnings" and the returns to an E/P strategy
are positively correlated with those based on these "unexpected earnings."
Many papers (e.g., Foster, Olsen, and Shevlin [1984] and Bernard and
Thomas [1989]) have documented "abnormal" returns following the
announcement of (unexpected) earnings that are in the direction of those
earnings. This implies that the market is slow to reflect the information
in current earnings. Kim's work indicates that the PIE effect may be
capturing this same phenomenon. This interpretation, along with the
results on the mispricing of Pr, would suggest that the market underreacts
to both the information in financial statements regarding current earnings and information in financial statements regarding future earnings.
The alternative explanation is that both EIP and Pr are proxies for
risk. Ball [1978] suggests that EIP may be a risk measure, but, if this is
so, the negative correlation between EIP and Pr indicates that Pr risk
must be some aspect of firm risk that is negatively correlated with EIP
risk.
From an investment point of view, the returns to EIP and Pr positions
are offsetting. For example, if we invested long in stocks with high EIP
to gain positive returns for "low PIE" stocks we may in the process be
investing in stocks with low Pr values which will generate negative
returns. Table 7 reveals that returns from long positions in stocks with
low PIE and high Pr and short positions in stocks with high PIE and
low Pr dominate returns to positions in Pr or PIE alone. With respect
to positions based on current earnings changes, postearnings drifts may
also be understated, at least for annual earnings changes. Since there are
reversals in earnings changes that are identified by Pr, some long (short)
positions based on positive (negative) earnings changes will yield negative
returns because they will pick up subsequent negative (positive) earnings
changes (in the opposite direction to current earnings changes) that are
identified by Pr. Indeed, Ball, Kothari, and Watts [1988] observe significant negative returns to long positions based on extreme positive annual
earnings changes. Extreme earnings changes have been identified with
earnings reversals in Brooks and Buckmaster [1976].

6. Conclusion
The evidence presented in this paper demonstrates that accounting
statements provide more than a historical perspective on the firm.
Financial statements, like prices, are also prospective. The information
about future earnings that is reflected in PIE ratios is also contained in
financial statements. Financial statements reveal two earnings numbers-current earnings that are explicitly identified in the income state-

134

JANE A. OU AND STEPHEN

H. PENMAN

c0a

o
.E

"

Q ,X

C',C'
~~~~~t-i0

- O
C;

Cl 10
M X,

1, L-

4.4)c

'-4

r 2

CO

co

co

ES

0o

At

. oo

.z

Al
Al

C'

All

CeD

>
t ~~~,4
~~~~~~~~~
to

*1

I *

I *

VI

A)

>

>

CO

V wA;:

v)

CCD

?1X

co Al .-

00

oL- _

.4
co

s II

00

C%

I*

*0*

*e

0Id

C~

Cl

J?

-z

*
*
C
r-- r-- L oo L L-

c
q

c:?

00

-0

o
r

ce

11 C1

INFORMATION CONTENT OF SECURITY PRICES

135

o
L
r-i
CYD
L
Dl

0
00

oCl D 0

C
q C

Cl rj
O-- 00

t
0

cX
co

o
0

w
bD

n4
=

'Qf

0
o~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~c
1
*A8
40
o

Cli

C!

0
0

4-)

I_
C

Q C))
D-

co
0

)
CY

00

)aDO

w,4

C1?

-4

zSD

H CO ?O

00

CS

>

S~~~~~~~~~CD

;,

co

?;

O
4

co~~g
Cl

~ ~ ~
~~~~~~~~~~~~

~ ~~~~0
~

11

4-

t-

~ ~

.?toXX
CYD

40

0
t0
cq

00

r-- "t

00 -tItil

0
'-4

0
0

0~a

r0

3?
bn

1:0a)
$- +

0-

ment and future earnings that are indicated by the financial statement
measure Pr. The latter can be estimated by our financial statement
analysis.
Rather than supporting the view that accounting rules garble information about earnings and that prices contain information that provides
a correction to the accounting calculation, our evidence indicates that
accounting reports contain information that filters out transitory components. As a result these reports provide, in part at least, a distinction
between current and future earnings that are implicit in prices and in
price-to-earnings comparisons. We also find that the transitory components of accounting earnings that can be identified by financial statement
analysis are value-relevant aspects of operations and not measurement
error.
Although there is adequate evidence here and elsewhere that prices
lead earnings, we find that price changes predict earnings changes relatively poorly when accounting statements indicate a high transitory
component to current earnings. The reason is that prices necessarily
capture both transitory and long-run earnings components. While price
changes reflect information about future earnings, they also reflect
transitory elements of current earnings that are negatively correlated
with future earnings changes.
The paper also finds that stock returns predicted by the financial
statement measure, Pr, are negatively correlated with stock returns
predicted by PIE ratios. Although both Pr and P/E provide similar
predictions of future earnings, they provide different predictions of future
stock returns. This can be explained either by P/E and Pr measuring
different aspects of risk or by the market pricing not only the future
earnings information in financial statements with a lag but also pricing
current earnings in financial statements with a lag. This contrasts with
the more popular view that accountants capture information in prices
with a lag.

137

INFORMATION CONTENT OF SECURITY PRICES

>

Ce

I c3s
O Co

CqL

Ce

. n.

r--C cO

r- )

cy r4 t-

4-)

to

-,

. o.

q0-imc)w0
--d
CYD
'm0
ri0

. o.

000

o.

4q00o6 r- (:

. o.

CY

to

-ti

. o.

olI m
It
II
I 00 N

.m

U~~~~~~~~~~~~~~~t
_w

t-

CSH
N r-0
L-

n
Ce
L-

r.-

CD cq
L-

rr-

o-- 0
rL

r->

CeD

r-

r-

r-

Ce

r-

q co

L-

4o
C

LCS

Lo

CS~~~~~~~~~~
CeD Cw

C)
*,i,~~~~~~~~~~~~~~c

cq

r-

R3~~~~~~~~~~~~~~~c

. o, .

>~~~~~~~~~~~~~,

LO

. H

--

ooq

Ce
t-

It
't

co

co

C o

Ocl Cq
Cq
1

00L

o co C' 0O
I

w 0

CS
w
mI
CeD CeDC

o-Co

O
c
c6 o6

IIL

mi

Co

. o. o. o. o. o. o. o. o.

Ce
LO

CCOrL-

Cr-OCO
co
00
O

LOLO tI I

I- Iq I

o.

<3?c) m ooir-i 'o 000 r-i w <3,w ? ?c


'o8 o
O
O
O
00
CN m CeD
T to
It
q "t
It

o o
O

C
CO
C00O
c6 o6 ce
c6

*ce~~~~~~~~~~~~~~~~~~~~~~~~~e
0 tlc
OriII L
D 0" coI - I" 0"d
ceco

cq

o
Ce

00

0
o

o. o. o. o.

00

C.)
4.41

7o

z4

01

-Q

*O

q~~~~~~~~~~~~~~~~~Y
0 0 O O
0 0
O O~~~~~~~~~~~~>

<

<

L
1m

cq
K

_N

_~

Cl

O)

wL0C0

CD

w
C"

1o

w
1>D

C-0

CeD w

D
n 00

tn

00

00

'o

'o

00

'o

m 00 tw
Cs
COm Cq L-

C
-

zD:

'o

CO

CO
w

'o

--I,

~. LO

"I

90

.n

C.)

.A

-11

>)~~~~~~~~~~~~~~C

Ci)~~~~~~~~~~~~~~~e

q~~~~~=.
AQ

CS CO t

e C)

Gwe~~~~~~~~~~~C
3
0

Cci)C.

7l4:

<

<

>

L6

<

>

<

>

Cl)

<

<

<

ir

0s

XC

SC

0so

Q D

..

<

H H
.=;

H
.

OiC

N~~~~~~~~~~~C
N N

<

00

._

co

C ci

..
4~~~~

Nl O

co
-

c)

CD

0O

O . C!
4~~

L o

~C

..
0

00

00
04

0
to

> c i)

Y-

C
0

00

D
00

..0

Y-

oo

t-!clCmCo
Y-

to

ci
c-O6-

1-

t'

'0

co

00
..

C4

!>

..

t>

..

C;

00

SH

Ht

ci
ci

'0

Cm

. 0 0
Co
00.. 0

12-

COL-

cl
co
CCCH

Cq

..

CD

cD

?L

)CC

00

"t

O01
!>

r-

m -4L to
m<: 0 O'
CC

'.coc'l
m<:

OC

c6

.S

i))
c00

....

oCCrY-

00 0
0

"

"Id CD M

t l Cl
Cl
C m
)O CD 00 Cl CDC!
CD LO O O O O O Y1- 0 Y- 0 00 CO
)t1"It O
CD CD CD CD to CD
O.DO O
O O

iOt
'0 co

co

0oHtoo

1..

't
r-

Lo

Cq

Y-

Cq

Cl '.0ci
m L
..

m:

M t0 t>

CC
l 0

-I
0

lC

Cm Y-

0
..O

to

to

Clt>
0

.1

CCm -I Cl
0

.CCO

C4

Cl

Cl
tO Ct

r-

cq

Cl0ci)!

CC0 0
0
1
..

--'0 r-

-iC
c

0)
0>CC,

CC C CC CC CC CC CC CC CLo
00 0
0 0
c)
00 cti
00

CeD CeD Y-

0o

o
Cl

M>

O ' 00 t

CC
!> 0)!>

CeD CD

It
..

Co t
m
Co m ''l
>C0 C
O cO
O "t 00 1- 00 00 rI co O
CD t- Oo O LO O
O "IO

.'0

Co
O
12~~~~~~~~~~~~~
~~~~~~CD
CO
H
eO
t~~~~C

t
CCt
CCt C
Cl 0)

C'

-40)Cl

Lo

c00

..

CD

Co

a0

to>

Lo)0

CIl in

t-

...

CeD Cq

CC . Cl

M
cO

CC!>

Cq
CC Cl
Lo0

!LO)'0!Lo
-L C

c
NH

i
R

o '0C

l~C
0)O Cs L)0
0)?)D1-0
CCoc)
co!>!!>'0
t

C.

>

Lo99999

CYD Cq

.2

Cq
cs

..
00

C C L
l 0)
O0)t

.0

,,d

!>

>t Co CO m ! >
O "t O Co O
CD
CD
O
O O

;b

>4

C;

.Re

H. PENMAN

JANE A. OU AND STEPHEN

138

-t

Y--

-4
C6 i6

00

'-

C'l
; * <

omc
~~~~~~o
0) LO 0 ~
C

C11

C- 2

Q
|-l

-IrIrIm
00
o Y0Cl 'I~~O

LO-lLOL
0)Cl0
CC ~ 1 to
0) CC
co
Ccio C
~~~~~~~~~~~~~~~~~o
to

t-

cs

LC O t Cl Cl O
C l LO "t O 0
LO C LO to LO
Y-4 ,-

0
Cl 0 to
CDto
CDto
in in to in
-

ooI

s m

CC 00 CC
c
CC
oO
to C to
-4 , -4

'o

ms

00
LO
to
, -4

0
0

Cl
t!0m2

CC >
CDto
LO to
, -4 , -4

L
t-

o
Y- 00

Cl 0

CC 00
CDto
LO LO
-

Loo CD
o 0000!>tt0
CC

wrIL
0
'IO
C11

r-

0
Cl
'IO
-

0
It
CC
-

00 00 t CD CC CC
0
LO LO!>
c>
0 t-C
CC CC
Cl Cl '
LO Cl Cl Cl
- 'O
-

1-0

NO-i2
E1G

bD0~~~~~~~~~~~~~~~~~~~~~~~N
?

< s
?o
q
_4 q
+C5

-e

U21

ag ;

O<
O

~
e>

E--

C-2-'-21

4
C12~~2

0nt e

eD
C

CD

(x

CQ ? Q OX:

mCmC
4

e z e
a4%

C12C42C42

eln

e>

e el

L-

'I'd,

LO

LO
M
L-

cq
0
mO 0
m O

cq
O

co

LO
L-

m
00o
CD

..I
LM

139

CONTENT OF SECURITY PRICES

INFORMATION

CD
It

in

14>.4

LO
LO o4
00
M L-o

10

cq~~~ct
0

.
'-4

oo L- to
O cO
r- cO
O

cs

It

co

It

co

co

teC
0

t-

cq

CO
m

r-

CD

CD
O

LO

CD
CO

cm oo LO
C) CO
H

m co
OlooOCl'4
L- oO O
m 00

It
O
Cl

O
O

CS
tcC
O
COS-Cl
HCOCOL
CO~

..

CD
cO

It

~L-O OL--L-cq
LOLOCOCOCO>
tT.

It
O=

- It

C1

~~~~~~~~

'-4
Cl
CD CD CD CD CD
O
00 cO O C>

cq

LO LOm m

cs

'I'd,

CD O

tc1
CO CS

cq Loo
00

H~~

cD L- Lo Co M Co C

c~~ici

ci
cD CeD

ci
00

CO

c
CD CD CD

cq L-

m CD

6
10

c6

>

o
4
C)

C)

)w

....

a-C)C)

O
O c
COOL-

9 9

-,

~ ~~

m L?O

LO CD C

^
c~~~~~~~~~~~~~~~~
o6
C
c
C

ai ci
CD

i6

cq

L-

L-

H~4-

4C'la)

Ma

CO

l
M Co

0
f
~~~~~~~~~--;
'-~~~~~~~~~~~~~~~~~
a

c) 5

O O

rD.=

99

C)

Co CeD y-

00 0

0 C
Cq 00
0
Lo 44t CD C

0 m

Cq

Lo L- cD Cm

4 C CD to
CO

000

t1 O

LO L-

It
LO O

m
CO

Lo

Cm

Cq

-C

co oO OO *r
=.
L-

co 0
O

to~~~~~~~~~~~~~~~~~~~~~~~~~~~~~C
C

;4C))

eD4

CI

C4

~~~0

C
o- CL Cq 4 4
oo r co
tC o~~a~a)LOLOC-4LOC-4COCO
CLO

CO
C
O
4 C
44 Lo 44 Cq Cq

t~

>L~t
COCO;-4CO~~LO1-COC0
bM

t-1--cl

"
CO<
-C4C4-

C4C-4

-e

CC

'

,m

oO

C;?$sO

.CC)
<

CO

t4

'4

'

CO

CO

L0-L
(

I<

'

C.)t\Q

)
I 00CZ
C0'
CD

a)

CZCC

0 CO
5 .C4-~C

?
a)0

oLO

COC
O

~-e-

CO~

a ))

C)

4-C '-4
'C
C
CCOCO4CO
e - CO
0 CO&q
E-1

C4-C

co

00 c

>0
F-

~~i

LO
to
4-4~~~~~~-

a)

0
S

CC0C

Lo

c0

0O"~C~ Ca)

0 bM

-4-5

C)8

'-4

xCL0

4-4C-2

i)

It

C
00
e

m .4-C0

)0 LO
COL
oo

OCC'
cOl

CD

CeD

LOslO

r
JCC-c

cl'oJC'tJC'

4
COC.OLOLOCOO'-4
4-4-CC
CC-

LOo)

CD Om

Lo CD CO
to

LO'CO

--

C
CC-)

0O0

~ ~ ~~~~~
'-~~

'C~~LCel*

co

co

Cl

co

c~~~

co

CD

a)

CD

CD

CD

CD

C)C
~~~~0C

H. PENMAN

JANE A. OU AND STEPHEN

140

co

t-: CD

t0

om0

?
CZ:

>!)CD10

r-

00

cD

CeD

L-

Cl

0
*

*Y
t =

,4,4

CI

P4
? o

<0

cl

L-

00

0 >

OWQ

CD

cq

D'XC''CO'
C11

>

Go
q C)

CD

O0

00o

1.

'I'd

CO

Cl

I
o

LO
-

~I

CZ

0~~~~~~~~~~~

E-

O~~c

.
o>

0~~~~~~~

0
0

~~~*;

~ ~ ~

-~~~~~~~~~~~~~~~~~~~~~-

F-

: O1

c~0
0

*-

>H

. ~. ~ ~. ~ . ~ .s .s .s *

F-4
CS

oo>-

o-

FH-

CS

C-

cq

~~~~~~~~~~~~

00

oo

rI

>

Oo

cq

00ms

cqO

oo

s
Cl,

CD

t-

LO

LO

00

*
C'S

CD

cO

cO

CD

C'~

~ ~ ~~~~~~~~~~
LOs

:
ail

00

C
cq

c
..

~~~
I
I

~~~~~~~~
I

C
4

L6

cs

00

cq

>0L

CC

cq

cl~

;-4

zs)

*
Cl

--

C2-

CI

~Cl

CO

Cl

0O

.*.

C-cS
cC
~~~~

&

lC

c-

.
I

Cl

C
CCc)oC1C

CC

CC

CS
bfl~~~~~~~~~~~~~~~c
-

Cl

CC

CC

CO

c12

142

H. PENMAN

JANE A. OU AND STEPHEN

-~~~~~~~~~~~~~

.
~~~~~~~0

0~~~~.
-~~~~~~~~~~~~~~~~~

,-.4

-4
t-

CD

.
10

co

10

>

",44

o6

C12~
.-

-,

oS

.~
C;~~~~~~~~~~*0
-,~~~~~~~~~~~~~~~~~~~
1

cqO

00

~~
0~~~~~40

4Q,
0

f
o~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

~ ~

-41

CZ

00

'~

~ ~ ~~~~~~~~~

Q-

0
0

0~~~~~~~~~~~~~~~~~
~~~~~~~~~0

> -%

0
~~~~~~~~~~~~~~~~~0
Obfl

O~

o
o

~ ~ ~ ~ ~ ~ ~ ~ ~

.1.6 .

C0

-4

L0

CO

M
to

C-100
0
c"I

>

C-C

CD

D C

LO C)

-d

bf

0~~~~~~~~U
bfl

-o

o-~

4-,

cci

L0O-

~ ~~~~~~~~~~
O~~~~~~~~~~~~~~~~~~~~~~~~~~~

00

0
0~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~0
*~
o,

~0O ~ ~

>

~C ~ ~
0
0
~~~~a
-~~~~~~~~~~~~~~~~~~~~
~~~~ r *~~~~~~~~0
c

C-

CO

+'+'

0
0

c0U)
0

ICI
4
0
-d

0
"0
4-4

4-)

~~~~~000C

0
0

+4~~~~~~~~L

-d
0

cf0

0
gj~~~~;,
~~~~~~~~~~~~
-40
0

O.

~~0-

0a)
-~~~~~~~~~~~~~~~~~
U,
3

0
4J~~~~0

0
4-~0
~ o,,~~~~~~~~:
oI
'~~~~~~~~

4-J

"
0
0~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

0
~ 0~~~~~~~~~~~~~~~.
~ ~ ~ ~ ~ 3
~~~~~~~~~~~~

INFORMATION

CONTENT OF SECURITY PRICES

143

REFERENCES
ALBRECHT, W. S., L. L. LOOKABILL, AND J. C. McKEOWN. "The Time-Series Properties

of Annual Earnings." Journal of Accounting Research 15 (Autumn 1977): 226-44.


BALL, R. "Anomalies in Relationships between Securities' Yields and Yield-Surrogates."

Journal of Financial Economics 6 (June/September 1978): 103-26.


, AND P. BROWN. "An Empirical Evaluation of Accounting Income Numbers."
Journal of Accounting Research 6 (Autumn 1968): 159-78.
, AND R. WATTS. "Some Time-Series Properties of Accounting Income." Journal of
Finance 27 (June 1972): 663-82.
, S. P. KOTHARI, AND S. L. WATTS. "The Economics of the Relation Between
Earnings Changes and Stock Returns." Working paper, University of Rochester, 1988.
BANZ,R. W., AND W. J. BREEN. "Sample-Dependent Results Using Accounting and Market
Data: Some Evidence." Journal of Finance 41 (September 1986): 779-93.
BASU, S. "The Relationship Between Earnings Yield, Market Value and Return for NYSE
Stocks: Further Evidence." Journal of Financial Economics 12 (June 1983): 129-56.
BEAVER, W. H., AND D. MORSE. "What Determines Price-Earnings Ratios?" Financial
Analysts Journal (July/August 1978): 65-76.
BEAVER, W. H., R. A. LAMBERT, AND D. MORSE. "The Information Content of Security
Prices." Journal of Accounting and Economics 2 (March 1980): 3-28.
BEAVER, W. H., R. A. LAMBERT, AND S. G. RYAN. "The Information Content of Security
Prices: A Second Look." Journal of Accounting and Economics 9 (July 1987): 139-57.
BERNARD, V. L., AND J. K. THOMAS. "Post-Earnings-Announcement Drift: Delayed Price
Response or Risk Premium?" Journal of Accounting Research (Supplement 1989): 1-36.
BROOKS, L., AND D. BUCKMASTER. "Further Evidence of the Time Series Properties of
Accounting Income." Journal of Finance 31 (December 1976): 1359-73.
COLLINS, D. W., S. P. KOTHARI, AND J. D. RAYBURN. "Firm Size and the Information
Content of Prices with Respect to Earnings." Journal of Accounting and Economics 9
(July 1987): 111-38.
COTTLE, S., R. F. MURRAY, AND F. E. BLOCK. Graham and Dodd's Security Analysis. 5th
ed. New York: McGraw-Hill, 1988.
FOSTER, G., C. OLSEN, AND T. SHEVLIN. "Earnings Releases, Anomalies, and the Behavior
of Security Returns." The Accounting Review 59 (October 1984): 574-603.
FREEMAN, R. N. "The Association Between Accounting Earnings and Security Returns for
Large and Small Firms." Journal of Accounting and Economics 9 (July 1987): 195-228.
, J. A. OHLSON, AND S. H. PENMAN. "Book Rate-of-Return and Prediction of
Earnings Changes: An Empirical Investigation." Journal of Accounting Research 20
(Autumn 1982): 639-53.
JAFFE, J. J., D. B. KEIM, AND R. WESTERFIELD. "Earnings Yields, Market Values and
Stock Returns." Journal of Finance 44 (March 1989): 135-48.
KIM, J. "The E/P Effect and the Earnings Forecast Error Effect: A Comparison of Two
Stock Market Anomalies." Ph.D. dissertation, University of California, Berkeley, 1987.
MOLODOVSKY, N. "A Theory of Price-Earnings Ratios." Financial Analysts Journal (November 1953): 65-80.
OHLSON, J. A. "Ungarbled Earnings and Dividends: An Analysis and Extension of the
Beaver, Lambert and Morse Valuation Model." Journal of Accounting and Economics 11
(July 1989a): 109-15.
. "A Synthesis of Security Valuation Theory and the Role of Dividends, Cash Flows,
and Earnings." Working paper, Columbia University, 1989b.
, AND S. H. PENMAN. "Evidence on Market Over- or Underreaction to Earnings."
Working paper, University of California, Berkeley, 1989.
Ou, J. A. "The Information Content of Non-Earnings Accounting Numbers as Earnings
Predictors." Journal of Accounting Research (forthcoming).

Journal of Financial Economics 6 (June/September 1978): 127-50.


, AND R. W. LEFTWICH. "The Time Series of Annual Accounting Earnings." Journal
of Accounting Research 15 (Autumn 1977): 253-71.

Você também pode gostar