Journal of Financial Economics 9 (1981) 1946.

North-Holland Pubhshing Company

MISSPECIFICATION OF CAPITAL ASSET PRICING

Empirical Anomalies Based on Earnings’
Yields and Market Values

Marc R. REINGANUM*
Uniuersrty of Southern California, Los Angeles, CA 90007, USA

Received December 1979, final version received June 1980

This study documents empmcal anomahes which suggest that either the simple one-period
capital asset prlcmg model (CAPM) is misspecified or that capital markets are lneffclent. In
particular, portfohos based on firm size or earnmgs/pruze (E/P) ratios experience average returns
systematIcally different from those predicted by the CAPM. Furthermore, the ‘abnormal’ returns
persist for at least two years. This persistence reduces the hkehhood that these results are bemg
generated by a market inefficIency Rather, the evidence seems to Indicate that the equihbrlum
prlcmg model 1s rmsspeclfied. However, the data also reveals that an E/P effect does not emerge
after returns are controlled for the firm Suze effect, the tirm size effect largely subsumes the E/P
effect. Thus, while the E/P anomaly and value anomaly exist when each variable IS consrdered
separately, the two anomahes seem to be related to the same set of missmg factors, and these
factors appear to be more closely associated with firm size than E/F ratios

1. Introduction

The foundations of current financial theory are being challenged by

empirical research that suggests that corporate earnings and firm size data
can be used to create portfolios that earn ‘abnormal’ returns. The reported
‘abnormal’ returns range from just a few percent per year to almost forty percent.
Such results, if true, are clearly inconsistent with the simple one period
capital asset pricing models of Sharpe (1964), Lintner (1965) and Black
(1972). However, the methodologies used to establish some of these
anomalous results often contain important flaws [see Ball (1978)].
In sections 2 and 3, stock returns after the announcement of quarterly
earnings are analyzed in a framework that avoids these problems. Two basic
results emerge from the analysis of the quarterly data:

*I wish to thank my Ph.D. CommIttee ~ Eugene Fama, Jack Gould, Robert Hamada, Jon
Ingersoll, Roger Kormendl, and Myron Scholes ~ as well as David F DeRosa, Muzhael
Gibbons, Pat Hess, B111 Schwert, Rob Stambaugh, and Ross Watts for theu helnful comments
and suggestions. Partial tinanclal support for th,s research was provided by the Umverslty of
Clucago’s Center for Research m Security Prices and the Graduate School of Busmess.
Naturally, all errors are my responsltnhty.
20 M.R. Reinganum, Anomalies concerning size und P/E rutios

(1) During 1976 and 1977, one could not systematically earn ‘abnormal’
returns by forming portfolios, on the basis of standardized unexpected
earnings as defined by Latane and Jones (1977).
(2) During this same time period, earnings/price ratios could be used to
create portfolios that systematically earned ‘abnormal’ returns of six to
seven percent per quarter. Furthermore these ‘abnormal’ returns persisted
for at least two quarters.

These results, jointly considered, are consistent with Ball’s proposition that
the pricing mechanism of the CAPM is misspecified but that capital markets
are informationally efficient. By construction, the unexpected earnmgs
portfolios are based upon an ephemeral event, so that any ‘abnormal’ returns
on these portfolios most likely would reflect an informational lag. However,
no such ‘abnormal’ returns are detected. In contrast, the persistence of
‘abnormal’ returns in the E/P portfolios is a phenomenon which probably
reflects permanent, underlying factors of equilibrium pricing rather than a
gross market inefficiency.
Section 4 explores the E/P anomaly in greater detail over longer historical
time periods. The analysis is based upon portfolios formed with annual
earnings/price ratios. The historical analysis documents the persistence and
extent of the E/P effect between high and low E/P securities. Indeed,
extremely high E/P securities experienced ‘abnormal’ returns of more than
seven percent per year in the second year after being selected for inclusion in
the high E/P portfolio.
The relationship between the E/P anomaly and value anomaly [see Banz
(1978)] is investigated in section 5. The evidence indicates that the
anomalous return behavior of low market value firms is perhaps even more
astounding than the E/P anomaly. A portfolio of fifty small firms experienced
average ‘abnormal’ returns of nearly fifteen percent per year for at least two
years. Indeed, after controlling ‘abnormal’ returns for an E/P effect, a large
and persistent market value effect is still detected. However, after controlling
‘abnormal’ returns for market value effects, one could not detect an
independent E/P effect. Thus, although an E/P anomaly and value anomaly
are detected when each variable is considered separately, the two anomalies
seem to be related to the same set of factors. Furthermore, these factors
appear to be more closely associated with firm size than with E/P ratios. The
value anomaly largely subsumes any E/P effect.

2. Tests of the CAPM based on standardized unexpected earnings

The history of research that explores portfolio selection based on

unexpected earnings is rather rich. Jones and Litzenberger (1970), Brown and
 M.R. Reingunum, Anomulres concrming size and P/E ratios 21

Kennelly (1972), Latane, Jones and Rieke (1974), Latane and Jones (1977),
Joy, Litzenberger and McEnnally (1977), and Watts (1978), among others,
have claimed that unanticipated earnings forecasts, based on publicly
available information, can be used to systematically predict stock prices.
These studies employ various models to forecast earnings. The forecasts are
then used to derive an unexpected earnings figure. In this study, a model of
Latant, Jones and Rieke (1974) and Latane and Jones (1977) is used. The
reason for the choice is simple. Latane and Jones do not claim that their
model is statistically superior to others or even measures true ‘market’
expectations. Rather, they simply claim that their model produces ‘abnormal
profits. That is, on the basis of their standardized unexpected earnings
(SUE), Latane and Jones report a mean spread between their high and low
SUE portfolios of about forty percent on an annual basis. No other study
reported such remarkable findings. However, a predictive test of the Latane
and Jones technique might not be expected to generate such a large spread.
For insofar as their model of quarterly earnings is misspecified [for example,
see Foster (1977)], tests based on portfolios formed using their estimator of
unexpected earnings could be biased against finding ‘abnormal’ returns.
The major result to be reported here is in marked contrast to the Latani
and Jones findings. The evidence indicates that significant ‘abnormal’ returns
cannot be systematically earned by grouping securities on the basis of their
SUE. The drop from about forty percent to nothing is indeed precipitous.
The major differences between this study and the Latane and Jones work are
as follows. First, the earnings data are collected from the Wall Street Journal
and not the Compustat tapes. Also, announcement dates are collected and
not assumed. Furthermore, the data analyzed ‘in this study are outside the
sample period considered by Latane and Jones. Hence, any benefits of model
selection due to hindsight are eliminated. Finally, the tests in this section are
explicitly formulated in the CAPM framework.

2.1. The data

Beginning with the fourth quarter of 1975, corporate quarterly earnings

and announcement dates are collected primarily from the Wall Street Journal
for eight quarters. The net income figures for the previous twenty quarters,
which are needed to calculate SUE, are obtained primarily from a 1978
version of the Compustat tapes. Also, most of the common shares data used
to scale earnings are collected from the Compustat tapes. The sample
consists of 566 New York Stock Exchange and American Stock Exchange
stocks with fiscal year ends in December. The sample is a subset of 577
companies that Latani and Jones analyzed in a paper presented at the 1977
22 M.R. Rernganum, Anomalies concerning size and P/E rarios

American Finance Association Meetings.’ The data for their 577 companies
extend through December of 1975.
Only 535 firms survived until the end of the sample period. Table 1
displays the distribution of quarterly earnings announcement dates by month
of release after the quarter, i.e., the + 1, +2, or +3 month. Firms with both
positive and negative earnings announcements are included in table 1. The
vast majority of firms reveal their first, second, and third quarter earnings
within one month of the fiscal quarter end. Except for fourth quarter results,
only a handful of firms delay release until the +3 month.

Table 1
Dtstribution of quarterly earnings announcements by month
of release.

Month of release”

Quarter +1 +2 f3

4175 124 325 105

l/l6 455 109 1
2116 453 103 4
3116 424 128 2
4176 124 316 106
l/77 441 105 3
2171 421 117 1
3/11 426 105 4

‘The + 1, +2, and + 3 months are the first, second, and

thrrd months following the fiscal quarter close, respecttvely.
The + 1, +2, and +3 columns contain the number of firms
in the sample that publicly released their quarterly earnings
during that month.

2.2. Portfolio selection

The heart of the portfolio selection procedure is the earnings per share
(EPS) forecast for each security. Latane and Jones found an extrapolative
trend model with seasonal dummies to be an eflicacious model. Their EPS
forecast is defined as

‘Latane and Jones supplied me with a lrst of the 577 compames that they used in their
analysis. These compames had thnty-five quarters of complete data for earnmgs, dtvidends, and
prices from June 1967 through December 1975 on a quarterly Compustst tape. All compames
had fiscal years ending on December 31. For analysts m this paper 566 of the 577 companies
were used. The numbers differ because ten compames were not contained on the CRSP dally
master and return tapes and because I was not able to tind the earnings announcements for one
multmattonal company, Unilever Ltd.
 M.R. Reingunum, Anomuhes concerning size und P/E ratto. 23

where

ezI zforecasted EPS at time 21 when the world is at time 20;

8 Eleast-squares regression coefficients; 8r is the estimated coefficient on
the time trend, 8, is the estimated coeffkient on the time trend
squared.
S =seasonal dummies; S, = 1 if second quarter, S, = 1 if third quarter, S,
= 1 if fourth quarter.
For each EPS forecast, the previous 20 quarters of EPS data are used to
estimate 8. For example, to compute EZ2 one calculates the regression
coefficients using data from the 2nd through 21st quarters of the sample. The
time variable still only runs from 1 through 20.
With the predicted value, E,, in hand the unexpected EPS is defined as

UE,=E,-i,, (2)

where E, is the reported EPS for quarter t. UE, is not a residual since E, was
not included as a data point in the regression from which ,!I?,was predicted.
The unexpected earnings per share is scaled by the standard error of estimate
from the regression equation associated with the prediction. The resealed
numbers are called standardized unexpected earnings (SUE). Thus,

SUE, = (Ei, -&)/S,, (3)

is the SUE for company i in quarter t.

The + 1 month SUE portfolios are based only on corporate quarterly
earnings released in the + 1 month after the fiscal quarter close. The high
SUE portfolio contains the twenty securities with the highest SUE; the low
SUE portfolio consists of twenty firms with the lowest SUE. Each twenty
security portfolio is subdivided into two equal-weighted portfolios with ten
securities. One portfolio contains the ten securities with the highest estimated
betas, and the other consists of the ten firms with the lowest estimated betas.
Weights are selected for the two ten-security portfolios so that the overall
twenty security portfolio has an estimated beta equal to one.2

2.3. Results

For the + 1 month SUE portfolios, daily portfolio returns from four
slightly different three month holding periods are analyzed; the periods begin
at the end of the months + 1, + 2, + 3, and + 4 following the fiscal quarter
close.3 For example, consider the firms that released fourth quarter earnings
‘This wetghting scheme was independently derived and used by Watts (1978).
‘Appendix 1 in Reinganum (1979) contains an analysis of the +2 month SUE portfohos. The
results for these portfohos were consistent with the + 1 month lindmgs.
24 M.R. Reinganum, Anomalies concerning size and P/E ratios

in January, a + 1 month. The SUE portfolio returns are analyzed in each of

the following four three-month periods: (February, March, April), (March,
April, May), (April, May, June), and (May, June, July). This technique is like
four different trading rules. Under the first rule, the investor assumes the
portfolio positions immediately after the information is known and holds the
position for three months maintaining the initial portfolio weights on a
daily basis. Under the second, third, and fourth rules, the investor only
assumes the position after delays of one, two and three months, respectively,
and then holds it for three months. One can think of this scheme as a way to
detect if ‘abnormal’ returns persist through time.
Since the beta risk of the high and low SUE portfolios is constructed to be
one, the difference in the expected returns between these portfolios should
equal zero under the null hypothesis that the CAPM accurately describes
asset pricing. Table 2 contains the estimated means of the differences in daily
returns between the high and low + 1 month SUE portfolios. Portfolio
weights on individual securities are based upon the betas estimated with the
sixty days of daily data immediately preceding the three month holding
periods using an equal-weighted NYSE-AMEX market index.4 Table 2
reveals that even if one acted immediately after the extreme SUE securities
were identified, the mean ‘abnormal’ returns are not statistically different
from zero. The difference in daily means between the high SUE and low
SUE portfolios for the overall period is only 0.000280 and its standard error
is 0.000295. As might be expected, ‘abnormal’ returns do not appear as one
delays action. Hence, the SUE evidence does not contradict the CAPM.

2.4. Concluding remarks

The results reported in this section indicate that ‘abnormal’ returns cannot
be earned over the period studied by constructing portfolios on the basis of
firms’ standardized unexpected earnings as defined by Latane and Jones.
While these results are consistent with the CAPM, they suggest an
interpretation that extends beyond the CAPM. In particular, the results offer
support for the assumption of market efficiency. The logic behind this
statement stems from the fact that the composition of the high and low SUE
portfolios naturally changes from quarter to quarter. Thus, one does not
expect this technique to identify a group of securities that persistently exhibit
‘abnormal’ returns, because the selection criterion is based on an ephemeral
event. Instead, the SUE tests are designed to detect one-time ‘blips’ in the

4Reinganum (1979) also uses portfolio weights extimated during the portfolio holding periods.
Although these weights cannot be used for trading rules, they can be used for sensitivity
analysis. While the weights differ from those used in table 2, the conclusions are the same. The
mean ‘abnormal’ returns are not statistically different from zero. The SUE do not demonstrate
ability to systematically discriminate between risk-adjusted returns.
 M.R. Reingonum, Anomalies concerning size and P/E ratios 25

Table 2
Mean differences in daily returns between the high and low standardized unexpected
earnings portfolios of identical beta risk.

Portfoho position taken at end of montha

Quarter +1 f2 +3 +4

All 0.280 0.158 0.410 - 0.057

(0.950) (0.533) (1.485) (-0.180)
4175 - 1.795 - 1.013 -0.416 -0.156
(-2.227) (- 1.585) (-0.653) (-0.265)
l/76 0.782 1.178 2.120 0.550
(0.746) (1.024) (2.182) (0.637)
Z/76 0.700 0.279 0212 0.163
(0.823) (0.310) (0 282) (0.218)
3176 0.479 0.537 0.287 0.344
(0.591) (0.735) (0.387) (0.427)
4176 -0.236 0.081 1.125 0.103
(-0.265) (0.099) (1.343) (0.065)
l/77 1.025 0.202 0 213 - 0.970
(1.430) (0.293) (0.310) (-1.372)
Z/77 1.067 0.603 0.151 -0.355
(1.291) (0.682) (0.175) (-0.510)
3177 0.189 - 0.633 - 0.449 -0.106
(0.297) (-0.804) (-0.650) (-0.127)

“The portfohos analyzed in this table only contam firms that released quarterly
earnings in the month lmmedlately followmg the fiscal quarter close. The identical
beta portfolio weights are estimated using the sixty days of dally return data
immediately preceding the three month portfolio holding periods. Betas are market
model estimates using an equal-weighted NYSE-AMEX market index. Reported dally
means are multiplied by 1000. T-values are in parentheses. The + 1 through +4
months refer to the first through fourth months following the fiscal quarter close. The
means for each quarter are based upon three months of trading day returns,
regardless of whether the three-month period begins at the end of the + 1, + 2, + 3, or
+ 4 month.

returns of specific securities. Thus, if extraordinary returns systematically

appeared, this might Suggest either a market disequilibrium or an
equilibrium with informational lags due to transaction and search costs. But
the evidence reported in this section does not support this scenario. Rather,
the evidence, along with the findings of the following sections, suggests
another interpretation. Namely, that when the criteria for inclusion in a
portfolio are based on ephemeral events, and when these events are
accurately measured so that only information actually available to investors
is used, then tests of market efficiency may not be very sensitive to the model
26 M.R. Remganum, Anomaltes concerning SW and PIE rut~o~

of equilibrium employed. That is, market efficiency tests on ephemeral signals

are probably robust with respect to reasonable equilibrium models. Indeed,
tests in subsequent sections indicate that the CAPM may not be an entirely
adequate model of equilibrium.

3. Tests of the CAPM based on quarterly E/P ratios

The proposition that high earnings/price ratio securities outperform low

earnings/price ratio securities dates back at least to Nicholson (1960). As
recently as 1977 Basu claims that the returns of extreme E/P portfolios
reflected a market inefficiency. In this section, the major result to be reported
is that an ‘abnormal’ return of about 0.1 percent per day on average can be
earned by forming portfolios based on E/P ratios. That is, the mean return of
a high E/P portfolio exceeds the mean return of a low E/P portfolio by
about 0.1 percent per day, even after adjusting for beta risk. Ignoring
transaction costs, this mean spread is greater than six percent per quarter,
and it persists for at least two quarters.

3.1. The data

The quarterly earnings data and firm sample are identical to those in
section 2. Earnings/price ratios are computed as the quarterly net income
divided by the value of the common stock. The value of the common stock is
calculated with both pre- and post-earnings announcements prices. The
closing price on the last day of the fiscal quarter is the pre-announcement
price. Thus, prices used in this ratio do not reflect information contained in
the public announcement of earnings. The post-announcement price is the
closing one on the day the earnings announcement appeared in the Wall
Street JournaI. If capital markets rapidly incorporate information into prices,
then rankings based upon post-announcement prices should reflect only the
equilibrium effect between E/P ratios and asset pricing; that is, scaling by
post- rather than pre-announcement prices eliminates any noise due to
unanticipated earnings from the equilibrium relationship between E/P ratios
and asset pricing.

3.2. Portfolio selection

Portfolios are formed on the basis of the ranked quarterly E/P ratios for
firms that released earnings during the + 1 month after the fiscal quarter
end. Hence, portfolios are formed on the basis of earnings information that is
at most one month old. The twenty highest and twenty lowest firms in the
ranking with positive E/P ratios become the high and low E/P portfolios,
respectively. In a manner identical to that described in section 2, each twenty
security portfolio is constructed to have an estimated beta equal to one.
 M.R. Remgonum, Anomalies concernmg size and P/E ratios 21

3.3. Results

Since the high and low E/P portfolios are constructed to have identical
beta risks, the null hypothesis is that the difference in expected returns equals
zero. Table 3 presents the estimated mean differences in daily returns
between the high and low E/P portfolios formed with earnings released in the
+ 1 months.5 Portfolio weights on individual securities are based upon the
betas estimated with the sixty days of daily return data immediately
preceding the three-month holding periods using an equal-weighted NYSE-
AMEX market index.6 Results are reported for portfolio positions assumed
at the ends of months + 1, +2, + 3, and +4 and held for three months in
each case. This technique is designed to test whether ‘abnormal’ returns
persist through time.
For the overall period, the mean ‘abnormal’ returns are positive; the null
hypothesis that the mean difference between portfolio returns is zero is
rejected. The data also reveal that the magnitudes of the ‘abnormal’ returns
across purchase dates do not change very much; even if one waited to act on
the earnings information for three months, a mean ‘abnormal’ daily return of
0.1132 percent could be earned. Table 3 also reveals that one might very well
accept the null hypothesis in any given quarter. In almost all quarters,
though, the estimated means are positive. When one considers all the
quarters together, the positive effect is estimated more precisely than in any
one subperiod.
The differences in mean daily returns between the high and low E/P
portfolios can be interpreted as ‘abnormal’ returns in table 3, because the
two portfolios have equivalent beta risk. Portfolio weights are based upon
betas estimated by regressing daily security returns against daily market
returns. However, work by Scholes and Williams (1977) indicates that the
stochastic process generating daily security returns may differ from this
‘market model’ process. The sensitivity of the results to this possibility can
also be investigated. The University of Chicago’s Center for Research in
Security Prices (CRSP) has computed Scholes-Williams beta estimates with a
value-weighted NYSE-AMEX market index for firms on their daily return
tape each year. CRSP assigns securities to various control portfolios based
upon their Scholes-Williams beta estimates. The daily control portfolio
return is subtracted from the daily security return to get the daily ‘abnormal’
or ‘excess’ security return. The daily ‘abnormal’ returns of the high and low
E/P portfolios can be constructed by just averaging (equal-weights) the
‘abnormal’ returns of the individual securities within these portfolios.

‘Appendix 2 in Reinganum (1979) presents the Iindmgs for the +2 month E/P portfohos.
These results are consistent with the dlscusslon of the + 1 month findings.
6Reinganum (1979) shows that using portfolio weights based upon betas estimated with the
daily return data from the three-month hotdmg periods does not significantly alter the results
reported m table 3.
 28 M.R. Reinganum, Anomalres concerning size and P/E ratios

Table 3
Mean ddferences in daily returns between the high and low E/P portfohos of Identical
beta risk.

Portfolio postttons taken at end of month”

Quarter fl +2 +3 +4

All 1.204 1.284 I.298 1.132

(3.50) (3 52) (3.65) (3.34)
4175 1.591 0.198 0.300 0 038
(1.76) (0.23) (0 36) (0.04)
l/76 0.692 2.082 2.122 1.599
(0.66) (1.53) (1.51) (1.68)
2176 -0.013 0.039 0.992 1.356
(-0.01) (0.04) (0.81) (1.11)
3116 2.038 2.934 1.779 0.732
(1.47) (2.24) (1.70) (0.73)
4176 1.708 1.608 1.965 2.458
(2.39) (2.46) (2.61) (2.53)
l/77 1.989 0.989 -0.164 -0.319
(2.30) (0.97) (-0 19) (-0.35)
2177 0.334 0.834 0.381 1.415
(0.38) (0.85) (0.42) (1.54)
3177 1.361 1.506 3.038 1.853
(1.39) (1.69) (3.65) (2.23)

“The portfohos analyzed m thts table only contain tirms that released quarterly
earnings in the month immedtately following the tiscal quarter close. The tdenttcal
beta portfolio weights are estimated using the stxty days of daily return data
tmmedrately preceding the three month portfolio holding periods. The t-values are m
parentheses. Reported dally means are multiplied by lO@O. Earnings are scaled by
closing prices on the last day of the liscal quarter. The + 1 through +4 months refer
to the tirst through fourth months following the fiscal quarter close. The means for
each quarter are based upon three months of trading day returns, regardless of
whether the three-month period begins at the end of the + 1, +2, +3, or +4 month.

The differences in the CRSP mean daily ‘abnormal’ returns between the
+ 1 month high and low E/P portfolios are displayed in table 4. While the
point estimates of the overall mean difference seem to tail .off as one
postpones the portfolio purchase date, the difference in ‘abnormal’ returns
still ranges in the six percent to seven percent per quarter vicinity.
Furthermore, the point estimates for the overall period are easily within one
standard error of the point estimates contained in table 3. In addition, the t-
values for the overall period are greater than three, and thk mean differences
within each subperiod are almost always positive. The results in table 4
indicate anomalous returns for at least six months; the high E/P securities
outperform the low E/P securities even after beta risk adjustment.
 M.R. Reingunum, Anomalies concerning sire and PIE rutlo.\ 29

Table 4
Mean differences m excess returns of the high and low E/P portfolios.

Portfolio positions taken at end of month’

Quarter +1 +2 +3 +4

All 1.202 1.103 1.066 0.937

(3 81) (3.56) (3.45) (3.08)
4175 1.763 0.038 - 0.067 0.142
(1.59) (0.05) (-0.08) (0.20)
l/76 0.491 1.443 1.633 0.867
(0.54) (1.53) (1.64) (0.90)
2176 -0.043 0.040 0.933 1.589
( - 0.50) (0.04) (1.02) (1.58)
3176 2.595 2.569 1.101 0.657
(2.70) (2.66) (1.20) ‘(0.75)
4176 1.734 1.778 2.300 1.900
(2.31) (2.46) (3.02) (2.47)
l/77 1.454 0.421 -0.320 -0.427
(2.03) (0.57) ( - 0.43) (-0.55)
2177 0.343 1.022 0.178 0.990
(0.41) (1.09) (0.20) (1.13)
3177 1.345 1.538 2.693 1.862
(1.47) (1.68) (3.04) (2.25)

“The portfolios analyzed m this table only contain firms that released quarterly
earnings in the month immediately followmg the fiscal quarter close. Excess security
returns used to construct portfolio excess portfolio returns are those computed by
CRSP for their Beta Excess Return Tape. The r-values are in parentheses. Reported
daily means are multrplied by 1ooO. Earnings are scaled by closmg prices on the last
day of the fiscal quarter. The + 1 through +4 months refer to the lirst through fourth
months following the fiscal quarter close. The means for each quarter are based upon
three months of trading day returns, regardless of whether the three-month period
begins at the end of the + 1, +2, f3, or +4 month.

Although the conclusions drawn from table 4 do not differ from those
drawn from table 3, the experimental technique does. In fact, the
computation of ‘abnormal’ returns in table 4 involved a different market
index (value-weighted versus equal-weighted NYSE-AMEX returns), different
beta estimates (Scholes-Williams versus ‘market model’), and different
security weighting procedures. Thus, the evidence in table 4 not only
corroborates the findings in table 3 but strongly suggests that the E/P
anomaly is not an artifact of methodology in this time period.
The numerical results in table 3 can be pictured vividly and perhaps better
illustrated with the aid of graphs. Figs. 1 and 2 show the performances of the
 30 M.R. Rerngunum, Anomul~rs concernrng size and P/E ratros

+ 1 month high and low E/P portfolios purchased at the end of months + 1
and +4, respectively. Each graph plots the cumulative average ‘abnormal’
returns (CAAR) for the first fifty trading days after each portfolio is bought.
Since the high and low E/P portfolio are constructed to have betas of one,
CAAR can be constructed for each portfolio as well as their difference. The
market return is subtracted from the high and low portfolio returns to
calculate their ‘abnormal’ performances separately. Under the null
hypothesis, all the CAAR should fluctuate randomly about zero. If there is a

I 10 20 3.0 4b 50
Days After Purchase

Fig. 1. Dtfference in cumulattve average ‘abnormal’ returns between the high E/P and low E/P
portfolios (- ). Cumulatrve average ‘abnormal’ return of the htgh E/P portfoho (-v v--),
Cumulative average ‘abnormal’ return of the low E/P portfoho (’ *. * ‘) Smce the estimated beta
of each E/P portfolio is 10, the drfference between the daily portfoho return and the dally
market return can be viewed as an ‘abnormal’ return. Portfolio returns are tracked for etght
three-month pertods The mttral event date (r=O) for each perrod ts the last tradmg day of the
month Immediately following the fiscal quarter close. The ‘abnormal’ portfolio return for day t,
averaged over the eight perrods, is calculated as follows:

The cumulattve average ‘abnormal’ portfoho for day T IS just the summatron of the daily
average ‘abnormal’ returns. That IS,

CAAR(T)= ;. AAR(t)
,=I

The high and low E/P portfohos only contam firms that released quarterly earmngs mformatton
m the month tmmedrately followmg the fiscal quarter close.
 M.R. Remganum, Anomalies concernmg Alze and P/E ratios 31

10 20 30 40 50
1

Days After Purchase

Fig 2. Ddference m cumulative average ‘abnormal’ returns between the high E/P and low E/P
portfohos (- ) Cumulative average ‘abnormal’ return of the high E/P portfolio (-v---)
Cumulative average ‘abnormal’ return of the low E/P portfoho (‘. . . ‘) Smce the estimated beta
of each E/P portfoho 1s 1.0, the difference between the daily portfoho return and the dally
market return can be vlewed as an ‘abnormal’ return. Portfolio returns are tracked for eight
three-month periods. The imtlal event date (t =0) for each period 1s the last tradmg day of the
fourth month followmg the fiscal quarter close. The ‘abnormal’ portfoho return for day t,
averaged over the eight periods, is calculated as follows.

The cumulative average ‘abnormal’ portfolio for day T LS Just the summation of the daily
average ‘abnormal’ returns. That IS,

CAAR(T)= ; AAR(t).
I= I

The high and low E/P portfohos only contam firms that released quarterly earnmgs mformatlon
m the month lmmedlately followmg the fiscal quarter close.

one-time effect on prices, the CAAR should adjust to the new level and then
vacillate about that level. But the graphs do not demonstrate either of these
patterns. Instead, one sees the CAAR of the difference in portfolio returns
steadily trending up throughout the portfolio holding periods. The upward
trend appears even after action is delayed for three months. The CAAR of
 32 M.R. Reingunum, Anon&es concernmg Sue and P/E ratios

the low E/P portfolios exhibit a persistent negative trend in both graphs. The
CAAR of the high E/P portfolios trend upward, but not as markedly as the
CAAR of the low E/P portfolios drift down in this time period. The graphs
clearly illustrate that the mean ‘abnormal’ returns of the E/P portfolio
positions reported in table 3 are not due to extraordinarily large ‘abnormal’
returns earned during the first couple of days the portfolios are held. Rather,
the evidence indicates that the model is consistently misspecilied; on average
‘abnormal’ returns can be earned on a day-to-day basis.
All previously reported results are based upon portfolios formed using E/P
ratios computed by dividing quarterly net income by closing prices on the
last day of the fiscal quarter. However, by scaling earnings by a post- rather
than pre-earnings announcement price and re-ranking securities, the stability
of the E/P proxy can be further investigated. Although not shown, the
differences in mean daily ‘abnormal’ returns between the high and low E/P
portfolios based on announcement day prices are virtually identical to those
in table 3.’ One can still earn significant ‘abnormal’ returns for at least six
months. In fact, a closer examination of the similarity between the results
reveals that the composition of securities within the high and low E/P
portfolios is virtually identical in each case. That is, for the two sets of high
and low E/P portfolios in any given quarter, about eighteen (out of twenty)
firms are common to the corresponding component portfolios. That scaling
by post-announcement prices does not alter the anomalous findings probably
reflects two things: first, that E/P ratios do indeed proxy for a variable
omitted from the simple CAPM; and secondly, that the pre-announcement
prices incorporate, to a large extent, the quarterly earnings information of
the extreme E/P securities.
In addition to persistent ‘abnormal’ returns, positively correlated E/P
rankings over the time would be additional evidence that E/P ratios proxy
for determinants of equilibrium not considered by the CAPM. Under the
null hypothesis of independent rankings, the test statistic, the Kendall Tau
coefficient, assumes values between - 1 and + 1 and is symmetrically
distributed about its mean value of zero. A tau equal to + 1 would imply
that the rankings perfectly coincide; a tau of - 1 would mean that
one ranking is the exact inverse of the other. Table 5 presents the Kendall
Tau coefficients for the E/P rankings of each quarter compared with the
rankings of every other quarter, using rankings based on earnings scaled by
pre-announcement prices. All the tau values in table 5 are positive.
Furthermore, one would reject the null hypothesis of independent rankings
at the 0.001 level in favor of the alternative of positive association.
Another measure of portfolio composition constancy is frequency with
which individual firms are in the top fifty (high E/P) and bottom fifty (low

‘Results based on month-end closmg prices reported in appendix 3 in Reinganum (1979) do

not alter this finding.
 M.R. Reinganum, Anomalies concernmg size and P/E ratios 33

Table 5
Kendall Tau coeftictents for E/P rankmgs between quarters a

Quarter

Quarter 4175 l/76 2176 3176 4176 l/71 2177 3177

4175 1.00 0.42 0.46 0.42 0.52 0 29 0.31 0.30

t/76 0.42 1.00 0.50 0.36 0.33 0.49 0.33 0.25
2176 0.46 0.49 1.00 0.53 0.38 0.39 0.52 0.35
3176 042 0.36 0.53 1.00 0.39 0.32 0.42 0.51
4116 0.52 0.33 0.38 0.39 1.00 0.34 0.32 0.31
l/77 0.29 0.49 0.39 0.32 0.34 1.00 0.38 0.29
2177 0.31 0.33 0.52 0.42 0.32 0,38 1.00 0.45
3171 0.30 0.25 0.35 0.51 0.31 0.29 0.45 1.00

‘The Kendall Tau coefictent assumes values between - 1 and + 1 and IS symmetrtcally
distrtbuted about tts mean value of zero. A tau equal to + 1 would Imply that the E/P rankmg
of securities between quarters coincide perfectly. A tau of - 1 would mean that one ranking IS
the exact inverse of the other

Table 6
Number of firms whrch appeared m the extreme E/P groups m exactly n quarters.”

Number of quarters, n

E/P group 1 2 3 4 5 6 7 8

Htgh E/P 69 48 21 19 8 3 2 3
Low E/P 96 56 17 13 9 5 2 0
Expected
number 201 73 15 2 <1 il <l <l
under H,

“Thts table contams the actual and expected frequency wtth whrch mdrvrdual firms are
m the top 50 (high E/P) and bottom 50 (low E/P) part of the E/P ranking durmg the
eight qiarters of the sample

E/P) portion of the E/P ranking during the eight quarters of the sample.
Table 6 displays these frequencies. For example, three firms appeared in the
high E/P group in all eight quarters; on the other hand, sixty-nine firms were
selected for inclusion in the high E/P group in only one quarter. Table 6 also
contains the expected number of firms that should appear in a fifty firm
portfolio in exactly n quarters under the null hypothesis that each firm in the
sample has equal probability of being selected. As can be seen, with random
selection, one expects many more firms to be selected one or two times for
inclusion in the extreme E/P groups than actually are. However, almost no
firms are expected to appear live or more times, even though many do. Not
only does this benchmark indicate a significant core of firms being selected
for inclusion in the extreme E/P groups four or more times, but it also
34 M.R Rernganum, Anomalies concerning sue and PIE ratios

indicates that many firms are systematically excluded. Thus, this evidence
supports the contention that the composition of E/P portfolios tends to be
stable over time.

3.4. Concluding remarks

The evidence presented in this section seems to indicate that something is

wrong with the capital asset pricing model. On the basis of earnings/price
ratios, one can form portfolios that earn ‘abnormal’ returns of between six
and seven percent per quarter. That is, the mean spread between high and
low E/P portfolios of equivalent beta risk is about 0.1 percent per day. The
most striking feature of the ‘abnormal’ returns is that they persist and do not
diminish for at least six months. This persistence reduces the likelihood that
these results are being generated by informational inefficiencies. Rather, the
evidence seems to suggest that the equilibrium pricing model is misspecified.
To reconcile the fact that E/P ratio portfolios detect a model
misspectficatton with the fact that standardized unexpected earnings
portfolios do not detect a model inconsistency requires an interpretation of
the evidence which is consistent with both findings. The E/P evidence
indicates that either the CAPM is wrong or that capital markets are
inefficient or both. However, the persistence of the ‘abnormal’ returns, along
with the unexpected earnings results, seems to rule out a market inefftciency
explanation. The interpretation to emerge from the data is that capital
markets are mformationally efficient, but that the simple capital asset pricing
model incorrectly specifies the equilibrium pricing mechanism. Thus, the
evidence is consistent with Ball’s conjecture that E/P ratios proxy for a
determinant of equi!ibrium omitted from the two-parameter model. However,
as the evidence from section 5 demonstrates, even E/P ratios do not matter
after security returns are controlled for the market value of common stock
effect.

4. Tests of the CAPM based on annual E/P ratios

The evtdence of the previous section shows that during 1976 and 1977
high E/P securities experienced higher average returns than low E/P
securities of similar beta risk. This portion of the study extends the E/P
analysis back through time and analyzes the performances of intermediate as
well as extreme E/P firms. The historical data corroborate the major finding
of the quarterly E/P data. In particular, a high E/P portfolio experienced
average returns of about thirteen percent a year more than the average
returns of low E/P portfolios durmg the second year after each was
identified, even though the low E/P portfolio had a higher estimated beta.
The methodology in this section differs from the one described in
 M.R. Reinganum, Anomahes concernmg sze and P/E rattos 35

section 3. First, earnings are gathered from the Compustat tapes and
announcement dates are not collected at all, because the quarterly findings
indicated that the E/P anomaly did not appear to be an information effect.
Secondly, annual earnings rather than quarterly earnings are used to
compute E/P ratios. The reasons for the change are twofold: (1) the use of
annual earnings should reduce the seasonality embedded in quarterly
earnings; and (2) by using the annual tape, E/P ratios can be computed back
until 1962, which permits full utilization of the daily CRSP NYSE-AMEX
data bases. The final difference between procedures is that portfolio
compositions are updated annually rather than quarterly.

4.1. The annual data

Data for the historical analysis were gathered from two sources. Corporate
annual earnings for the years 1962 through 1975 came from a 1978 version
of the Compustat Merged Annual Industrial Tape produced by CRSP. The
merged tape includes Compustat’s research file, so that firms not currently
doing business can nonetheless be analyzed in earlier periods. Stock prices,
returns, and common share data are collected from the CRSP daily master
and return tapes.
To qualify for inclusion in the sample a firms had to meet the following
requirements. First, the firm’s fiscal year end month must be December.
Secondly, the firm’s annual earnings, year-end price and common shares data
must be available on the Compustat and CRSP data bases, respectively. No
other requirements on availability of past data or future data were used to
select the sample. The number of firms which met these requirements in any
given year ranged from about seven hundred in the mid-1960s to about
twelve hundred in the mid-1970s. Firms are continually dropping out of the
sample while others are entering the sample, for reasons such as mergers,
bankruptcies, new listings, and data availability.

4.2. Portfolio selection

Each year all firms within the sample are ranked on the basis of their E/P
ratios. Earnings/price ratios are computed as annual income after
extraordinary items and discontinued operations divided by the value of the
common stock at the fiscal year end. That is, annual earnings are divided by
the value of the common as of December 31. As with the quarterly data,
only firms with positive E/P ratios are considered for inclusion in the E/P
portfolios. The distribution of annual E/P ratios is then broken down into
deciles. The daily returns of firms with E/P ratios in the highest decile are
combined to form the daily return of portfolio EPIO. Similarly, the daily
returns of firms in the other E/P deciles are comgined to create the daily
36 M.R. Remganum, Anomalies concerning size and P/E ratios

returns of the other nine E/P portfolios, EPl through EP9. Although the
number of firms in each portfolio varies from year to year (because the
eligible sample changes yearly), over the fourteen year period there are about
ninety-five firms in each portfolio on average. No special weighting scheme is
employed to risk adjust E/P portfolios for beta; equal weights are applied to
all securities. Preliminary analysis of the data revealed that ‘market model’
beta estimates were close to one. Thus, in this section, the equal-weighted
NYSE-AMEX market index will serve as the control portfolio against which
E/P portfolio returns are compared.’

4.3. Annual results for ten E/P portfolios

To understand the presentation of results in this section, some basic

nomenclature must be explained. Let Y denote the fiscal (and calendar) year
from which E/P ratios are computed. For example, when 1964 annual
earnings are used in the numerator of the E/P ratio, Y is 1964. Let Y + 1
denote the twelve month period beginning on April I of the calendar year
following year Y. Let Y +2 stand for the twelve month period starting on
January 1 of the second calendar year after year Y. Portfolio returns in this
section are tracked and analyzed in both the Y + 1 and Y +2 years.
For table 7, the daily returns in years Y + 1 of each E/P decile portfolio
are stacked into one long time-series vector. The results of this table can be
interpreted as illustrating the average E/P effects for the fourteen year overall
period. The danger of such an interpretation is that the statistical
underpinnings of the tests may be violated. In particular, one worries
whether the parameters of the distributions can possibly be stationary over a
fourteen year period with securities shifting in and out of the E/P portfolios.
However, it may be the case that rotation of securities in and out of the E/P
portfolios actually tends to preserve the parameters of the portfolios over
time. Furthermore, differences between the high and low E/P portfolios
consistently appear in the year-by-year results.g
For each of the E/P portfolios in the Y + 1 years, five basic pieces of
information are presented in table 7: (1) the mean daily excess return; (2) the
estimated portfolio beta; (3) the average number of firms (in percent) within
the portfolio that are listed on the American Stock Exchange; (4) the median
E/P ratio for each E/P class averaged over the fourteen years; and (5) the
daily autocorrelations of the excess returns through lag three. For table 7 an
excess return is computed as the daily E/P portfolio return minus the equal-
weighted NYSE-AMEX market return. The results show that the average

8All tests were also conducted usmg the value-wetghted NYSE-AMEX marhet Index These
results are m appendrx 4 m Reinganum (1979). The evidence led to conclusions similar to those
drawn m the text.
‘The year-by-year E/P portfoho results are contamed m appendix 5 m Relnganum (1979)
 M.R. Reingunum, Anomalies concernmg size nnd P/E ratios 37

Table 7
Mean excess daily returns of ten E/P portfolios in years Y+ I (based on equal-weighted NYSE-
AMEX index).’

Autocorrelations
Mean Average Average of excess returns
excess percent median
Portfolio returns Beta on AMEX E/P I 2 3

EPl -0.124 1 I2 24.10”‘/0 0.032 0.09 0.03 0.02

(-1 80) (131.7)
EP2 -0.176 17.94 ;/A 0.050 0.13 0.06
(-3.44) (15:.;
EP3 -0.227 0.96 16.00’~ 0.064 0.16 0.07 0.06
(-477) (161.1)
EP4 -0.209 0.90 15.32% 0.076 0.16 0.08 0.10
( - 4.44) (157 5)
EP5 -0.109 0.90 13.91% 0.086 0.13 6.07 0.11
(-2.47) (1694)
EP6 -0.147 0.83 13 44 ‘:/, 0.096 0.20 0.12 0.10
(-3.13) (160 I)
EP7 - 0.070 0.86 17.58”’/0 0.107 0 I3 0.08 0.07
(- 1.48) (157.4)
EP8 0.058 0.82 16.50% 0.121 0.21 0.09 0.08
(1.15) (145.5)
EP9 0.103 0.88 20.27 “/, 0.139 0.14 0.04 0.06
(2.00) (142.2)
EPIO 0.165 0.95 35.86 “/, 0.181 0.03 0.04 0.05
(3.02) (138.8)

“Mean excess returns reported above are multlphed by 1000. T-values are in parentheses. An
excess return 1s detined as the daily portfolio return less the equal-weighted NYSE-AMEX
market return. Betas are market model estimates usmg this index. EPI is the lowest E/P
portfolio; EPfO IS the highest E/P portfolio. The return statistics are based on 3505 dally
observations from 1963 to 1977. The table also contains the percentage of lirms within each
portfoho, averaged over I4 years, that are hsted on the American Stock Exchange. The average
me&an E/P column contams the value of the median E/P ratio for reach portfoho averaged over
the I4 years of the study.

return of portfolio EPlO is more than four percent per year greater than the
average return implied by its beta risk. In fact, the mean return of EPlO is
almost ten percent per year higher than the mean return of EP3 which has
equivalent estimated beta risk. It is also interesting to note that AMEX firms
comprise the greatest proportion of firms for any portfolio in EPIO. Hence,
the portfolio with the largest mean ‘abnormal’ return contains more AMEX
firms than any other portfolio. Thus, in part, the E/P effect may be
correlated with a stock exchange effect. The most obvious difference between
the two exchanges is that smaller firms, in general, are traded on the AMEX.
So the E/P anomaly may be related to the market value of firms.
JFt C
38 M.R. Reinganum, Anomaltes concemmg size and P/E ratios

Although not shown, the same excess return statistics were calculated
using Y +2 year data. Portfolio EPlO still possesses a positive mean
‘abnormal’ return of about 4.5 percent per year. Furthermore, the fifty
highest E/P firms within this portfolio still have a mean ‘abnormal’ return of
slightly more than seven percent on an annual basis m the Y +2 years.
Indeed, EPlO still earns an average return of about eight percent per year
more than EP5, which has identical estimated beta risk. Thus, the ‘abnormal’
returns persist through time.

4.4. Concluding remarks

The annual data bear out the initial conclusions drawn from the quarterly
data. In particular, the average returns of high E/P securities are greater than
the average returns of low E/P securities of equivalent beta risk.
Furthermore, these ‘abnormal’ returns persist for at least two years from the
portfolio formation date. The persistence of ‘abnormal’ returns for two years
reduces the likelihood that the anomaly is due to a market inefficiency.
Rather, the evidence indicates that the CAPM incorrectly specifies the
equilibrium pricing mechanism.
In light of Roll’s criticism of empirical tests of the CAPM, one might argue
the evidence only proves that the equal-weighted market index is not on the
efficient frontier. But this critique does not go to the heart of the matter,
because there is no a priori reason to expect that the inefficiency should be
systematically related to an E/P effect. Furthermore, the preliminary evidence
from work being done to expand the market index indicates that results
based on broader indices do not greatly differ from those based on NYSE-
AMEX indices [for example, see Stambaugh (1979)]. Thus, one may feel
confident in concluding that the E/P anomaly arises because the true
equilibrium pricing mechanism is not well approximated empirically by the
simple one-period CAPM.

5. Relationships between the E/P and value anomalies

In a recently completed dissertation, Banz (1978) reported a nonlinear

relationship between the aggregate market value of a firm’s common stock
and the stock’s mean return. In particular he found that firms with very
small market values (relative to the rest of the market) had large and positive
residual returns. The problem to be explored here is whether the value
anomaly and the E/P anomaly are two independent effects or whether both
anomalies proxy for the same missing factors.
As a first step of the analysis, the value effect should be established on the
sample of firms used to discover the E/P anomaly. To this end, the entire set
of stocks in the E/P sample are ranked on the basis of their December 31
 M.R. Reingunum, Anomuhes concermng ~LZ~ and P/E ratios 39

market values each year. The daily returns of securities with market values in
the lowest decile of this ranking are equal-weighted to form the daily return
of portfolio MVJ. Daily returns of securities in each of the other value
deciles are also equal-weighted to form the daily returns of the remaining
MV portfolios, MV2 through MI/JO. The ten MV decile portfolios are
analyzed in a fashion identical to the ten E/P decile portfolios.

5.1. Annual results for ten MV portfolios

In table 8 live pieces of information are presented for the ten MV

portfolios in years Y + 1: (1) the mean excess daily returns; (2) the estimated
portfolio beta; (3) the average percentage of firms within the portfolio listed
on the American Stock Exchange; (4) the median value of the firms within
the portfolio averaged over the fourteen years; and (5) the first three daily
autocorrelations of the excess returns. Excess returns ari calculated by
subtracting the daily return of the equal-weighted NYSE-AMEX index from
the daily portfolio return. Betas are market model estimates using this
index.‘” The results in table 8 are computed by stacking the fourteen years
of daily portfolio returns into one time-series vector.” The potential pitfalls
of such a summarization of results were discussed earlier.
In table 8 one observes that for the lowest market value portfolios, MVJ
through MV4, the estimated portfolio betas are virtually one. Hence, the
excess returns of these portfolios can be interpreted directly as ‘abnormal’
returns. Portfolio MVJ possesses a mean daily ‘abnormal’ return of 0.05
percent or more than twelve percent on an annual basis in the Y + 1 years.
For the next decile portfolio, MV2, the ‘abnormal’ daily return has dropped
to only 0.02 percent or just slightly over four percent per year. For all other
MV portfolios, the point estimates of the excess returns are negative. Thus,
positive anomalous return behavior is detected only for firms in the lower
two value deciles and is especially pronounced in the firms of portfolio MVJ
with the smallest market values. Portfolios MVJ and MV2 stand out not
only because of their positive ‘abnormal’ returns but also because each
portfolio is heavily populated with firms that trade on the American Stock
Exchange. In portfolio MVJ, on average more than eighty percent of the
firms within the portfolio are listed on the AMEX. For portfolio MV2, the
percentage is just about fifty. As with the high E/P portfolio, positive
‘abnormal’ returns seem to be associated with the AMEX. Of course, as table
8 indicates, the stock exchange effect might more precisely be described as a
value effect.

“In appendix 4 in Reinganum (1979) resu!ts based on a value-welghted NYSE-AMEX mdex

are presented. Interpretations of the evidence are not altered by the change of index.
“The market value portfoho results on a year-by-year are reported in appendix 5 in
Remganum (1979)
40 M.R. Reinganum, Anomalies concerning size and P/E ratios

Table 8
Mean excess daily returns of ten MV portfolios in years Y + 1 (based on equal-weighted NYSE-
AMEX index).’

Autocorrelations
Mean Average Average of excess returns
excess percent medran
Portfolio returns Beta on AMEX value 1 2 3

Ml’1 0.500 1.00 82.61 % 8.3 0.06 0.03 0.06

(6.42) (101.7)
MV2 0.193 1.02 48.35 % 20.0 -0.05 0.01 -0.00
(3.47) (144.9)
MV3 - 0.033 1.00 23.81 % 34.1 0.01 -0.00 0.02
(-0.71) (171.3)
MV4 - 0.050 1.00 11.29% 54.5 0.05 - 0.02 0.00
(-1.11) (177.1)
MVS -0.115 0.94 8.59 % 86.1 0.09 0.04 0.01
(-2.60) (170.3)
MV6 -0.193 0.88 4.42 % 138.3 0.20 0.07 0.11
(-4.18) (160.9)
MV7 -0.189 0.90 4.35 % 233.5 0.27 0.17 0.16
(-3.99) (156.8)
MV8 -0.214 0.83 2.71 % 413.0 0.37 0.22 0.17
(-4.00) (135.9)
MV9 - 0.292 0.83 2.46 % 705.3 0.38 0.23 0.21
(-5.14) (126.3)
MVlO - 0.343 0.82 1.60 % 1,759.0 0.37 0.25 0.21
(-4.79) (96.3)

‘Mean excess returns reported above are multiplied by 1000. T-values are in parentheses. An
excess return is detined as the daily portfolio return less the equal-weighted NYSE-AMEX
market return. Betas are market model estimates using this index. M1’1 is the lowest MV
portfolio; MVZO is the highest MV portfolio. The return stattstics are based on 3505 daily
observations from 1963 to 1977. Median values are stated in terms of millions of dollars. Only
the average of the median values during the 14 years of the study is presented in the table.
Similarly, the percentage of firms within each portfoho that are traded on the American Stock
Exchange, averaged over 14 years, is presented.

The daily returns of the MV portfolios during the Y + 2 years were also
analyzed. The similarities between the Y + 1 year and Y + 2 year results are
striking. Even a year after the portfolios have been identified, the small firms
in MVI continue to earn ‘abnormal’ returns of about 0.05 percent per day.
Not only do ‘abnormal’ returns for small firms persist, but they persist at
about the same level in the second year as in the first year. The return
behaviors of the larger ML’ portfolios are also very similar to the results
reported in table 8. Thus, the conclusions drawn from the Y +2 year results
are almost identical to those drawn from the Y + 1 year data. Namely, one
 M.R. Reinganum, Anomalies concerning size and PIE ratios 41

can earn ‘abnormal’ returns that persist for at least two years by forming
portfolios based on the market value of the stock.

5.2. Interactions between the E/P and value effects

With the value anomaly replicated in the E/P stock sample, attention can
now be directed to the issue of whether the E/P and value anomalies are
related or independent. As a first descriptive step in the analysis, firms are
classified by both the market value of the common stock and E/P ratios.
Firms are classified according to which value decile and E/P decile they
jointly belong. Table 9 contains the two-way classification scheme averaged
over the fourteen portfolio formation periods. Casual observation of the table
reveals a slightly positive correlation between low E/P ratios and large
market values. But perhaps the most striking characteristic of the
classifications is the paucity of firms with large market values and high E/P
ratios. Firms with high E/P ratios tend to be classified in the low market
value deciles. One plausible hypothesis is that E/P ratios indirectly proxy for
the same factors that generate the value anomaly.

Table 9
Average number of firms in value by E/P decile categortes from 1962 through 1975!

E/P dectles

Value deciles
in dollars 0.042 0.058 0.070 0.081 0.091 0.101 0.114 0.128 0.153 0.916

13,885,509 12.4 7.1 5.9 5.7 66 6.5 7.6 8.4 12.0 21.9
26,362,480 8.0 6.5 6.0 7.0 7.6 8.1 11.1 11.3 12.1 16.1
43,859,952 8.1 4.9 7.1 75 7.8 8.0 11.9 10.9 12.9 149
68,086,944 5.3 7.1 8.2 7.6 8.9 9.0 10.0 12.3 13.3 12.1
108,477,280 6.5 9.1 7.6 10.0 9.1 112 8.9 11.8 10.6 9.0
174,785,568 6.3 9.5 10.1 9.5 11.9 113 10.2 110 8.8 5.0
309,664,512 7.7 9.9 129 12.6 11.6 9.4 8.3 8.0 7.7 5.4
530,984,960 8.3 10.8 10.9 12.2 11.0 11.1 10.0 7.4 7.9 3.7
956,984,832 10.1 11.4 13.4 12.1 10.9 11.2 9.7 6.9 4.9 2.5
34,765,234,200 214 17.6 119 9.6 8.5 7.6 5.9 5.4 2.8 2.4

“Firms are classified according to which value decile and E/P dectle they jomtly belong. The
upper cutoff points for the value and E/P dectles are also presented.
42 M R. Rernganum, Anomalzes concernmg sze and PIE ratros

To test whether E/P ratios can generate any additional ‘abnormal’ returns
above and beyond those detected in small firms, one might like to subdivide
the M V portfolios into sub-portfolios based on E/P ratios. The classification
scheme selected for this paper is to form twenty-five portfolios based on E/P
ratios and market values. The placement of a firm in a given portfolio
depends on the firm’s E/P ratio and the market value of its common stock.
The cutoff point for inclusion in portfolios is jointly based on E/P quintiles
and MV quintiles. The reason two-way classifications are chosen at the
quintile rather than the decile level is that in some years no firms would be
included in a portfolio constructed with a two-way decile classification. For
example, m 1966 no firms were simultaneously in the highest MV decile and
the highest E/P decile. Clearly, with this classification scheme, the number of
securities within each E/P-MI/ portfolio will not be the same for all such
portfolios. The largest portfolios tend to be the highest E/P, lowest MV and
the lowest E/P, highest MV one. Regardless of the number of securities in an
E/P-MI/ portfolio, within each portfolio equal weights are applied to the
security returns.
Tables 10 and 11 contain the mean excess returns and estimated betas of
the twenty-five E/P-MV portfolios in years Y + 1 based on the equal-
weighted index, respectively. An excess return is defined as the daily return of
the E/P-MI/ portfolio less the equal-weighted NYSE-AMEX market daily
return. As is true of the other tables in this section, the statistical results in
these tables are calculated by stacking the fourteen years of daily return data
into one time-series vector. By reading across a given row in table 10, one
observes the effect of varying market values while roughly holding constant
the E/P ratios. Similarly, in each column one can assess the effect of
changing E/P ratios while holding market values roughly constant. Thus, the
purpose of this two-way classification is to hold one anomaly constant and
to investigate the impact of the other.
In table 10 one observes a value effect across all E/P levels. The smallest
firms in a given E/P quintile systematically outperform the high market value
firms in that quintile, and this result is true for each of the five E/P quintiles.
One can formally test whether the mean excess returns of the lowest MV
portfolios are equal to those of the highest MV portfolios within each E/P
class. The test can be formulated within Zellner’s seemingly unrelated
regression framework. One first estimates the contemporaneous covariance
matrix for all twenty-live portfolios and then simultaneously tests the five
appropriate linear constraints. If one ignores the approximation of the true
covariance matrix by the sample one, then the appropriate test statistic has
an F distribution under the null hypothesis [Theil (1971, p. 314)]. The five
and one percent limits for F(5,a) are 2.21 and 3.02, respectively. However,
since each portfolio contains 3505 observations, the one percent significance
level is probably a more appropriate criterion for testing the hypothesis than
 M.R. Remganum, Anomahes concerning size and P/E ratios 43

Table 10
Mean excess returns of twenty-five E/P-MI/ portfohos m years Y+ 1 (based on equal-
wetghted NYSE-AMEX index).”

MV qumttle
Ei’P
qumttle Lowest 2 3 4 Htghest

Lowest 0 540 0001 -0088 -0285 - 0.348

(3.89) (0.01) (-0.99) (-390) (-4 17)
2 0.331 -0 100 - 0.308 -0294 -0 318
(2.53) - 1.17) (-471) (-4.71) (-470)
3 0.303 -0.119 -0 220 -0 181 - 0.262
(2.81) - 1.66) (-3.72) (- 3.08) (-3.91)
4 0.240 - 0.028 -0.119 - 0.058 - 0.209
I,2 71) -045) (-1 89) (0.83) (-271)
Htghest 0.375 0.029 -- 0.005 - 0.082 - 0.229
(4.87) (0.48) (-007) (-0 96) (-1 89)

“The mean excess returns reported above are multtphed by 1000. T-values are m
parentheses. An excess return IS defined to be the dally portfoho return less the equal-
wetghted NYSE--AMEX market return. The stattsttcs are based on 3505 dally
observattons from 1963 to 1977. The placement of a firm m a gtven portfoho depends
on the firm’s E/P ratto and the market value of Its common stock.

Table 11
Esttmated betas of twenty-five E/P-MV portfohos m years Y + 1 (based on equal-
wetghted NYSE-AMEX Index) *

MV qumtile
EIP
qumttle Lowest 2 3 4 Htghest

Lowest 1 17 1 19 111 100 0.92

(68.1) (83 5) (1004) (104 6) (88 1)
2 1.10 1.08 0.93 0.85 0 81
(66 6) (100 6) (1139) (1139) (102.4)
3 0.98 096 0.85 0 82 0 75
(72 5) (106.2) (120 7) (121.4) (1019)
4 0 92 0 89 081 081 0.76
(82 9) (118 7) (111.8) (99.2) (86.1)
Htghest 0.93 0.94 0.89 0.95 0.86
(96 1) (1244) (102 3) (89 3) (56 5)

aBetas are market model esttmates usmg an equal-wetghted NYSE-AMEX market

Index. T-values are m parentheses. The stattsttcs are based on 3505 dally observattons
from 1963 to 1977.

the five percent level. The computed value of the test statistic is 8.28; one
would reject the null hypothesis. Thus, the evidence indicates the presence of
a substantial value effect n-respective of a security’s E/P ratio.
44 M.R. Reinganum, Anomalies concerning size und P/E ratios

The evidence does not weigh heavily in favor of an E/P effect after
controlling portfolio returns for market values in table 10. In fact, within the
low MI’ quintile, the lowest E/P securities possess a mean excess return
greater than that of the highest E/P securities. Within this group of securities,
the predicted E/P effect may be reversed. Again, one can simultaneously test
for an E/P effect in each ML’ quintile using the methodology described
above. The computed value of the test statistic is 1.12. Thus, at the one
percent significance level one could not reject the null hypothesis. Thus, after
controlling ‘abnormal’ returns for the value effect, an E/P effect is not
detected in the Y + 1 years.
Tests conducted with the Y +2 year data corroborated the finding that
E/P ratios do not discriminate between excess returns after one controls for
market values. The hypothesis that the highest and lowest E/P securities
within each MT/ quintile possessed identical mean excess returns could not
be rejected at the one percent significance level. Thus, the evidence from the
Y +2 years indicated that high E/P securities did not systematically
outperform low E/P securities after the market value effects were removed
from the ‘abnormal’ returns. However, the hypothesis that the highest and
lowest MI/ securities within each E/P quintile have identical mean excess
returns was rejected at the one percent level.

5.3. Concluding remarks

Undoubtedly one goal of scientific inquiry is to reduce the complexity of

the world by applying some sort of Occam’s Razor. The evidence in this
section suggests such a simplification. In particular, the evidence indicates
that the E/P anomaly and value anomaly proxy for the same set of factors
missing from the specification of the simple one-period CAPM. However, the
evidence also reveals that this set of factors is much more closely associated
with firm size than with E/P ratios. Thus, the tests demonstrate that the
value effect subsumes the E/P effect.

6. Summary and conclusions

The evidence in this study strongly suggests that the simple one-period
capital asset pricing model is misspecified. The set of factors omitted from
the equilibrium pricing mechanism seems to be more closely related to firm
size than E/P ratios. The misspecification, however, does not appear to be a
market inefficiency in the sense that ‘abnormal’ returns arise because of
transaction costs or informational lags. Rather, the source of the
misspecification seems to be risk factors that are omitted from the CAPM as
is evidenced by the persistence of ‘abnormal’ returns for at least two years.
 M.R. Reingcmum, Anomulies concerning size and P/E ratios 45

In sections 2 and 3, quarterly earnings data, primarily collected from 1976

and 1977 issues of the Wall Street Journal, were analyzed in two distinct
ways. Portfolios formed on the basis of standardized unexpected earnings
exhibited no ‘abnormal’ return behavior; capital markets rapidly
incorporated any new information contained in these ephemeral signals into
prices. On the other hand, the same earnings data were used to create high
E/P portfolios that systematically outperformed low E/P portfolios, even after
beta risk adjustment. In fact, the ‘abnormal’ returns of about six to seven
percent per quarter persisted for at least six months. Analysis of the annual
data over longer time periods in section 4 documented the persistence of the
anomalous E/P results. Even during the second year after its formation, the
high E/P decile portfolio experienced significant ‘abnormal’ returns; a lifty-
firm subset of this portfolio earned ‘abnormal’ returns of about seven percent
per year during the second year.
Section 5 explored the relationship between the E/P anomaly and the
market value anomaly. An analysis of the E/P sample firms revealed that
within this sample small firms systematically experienced average rates of
return significantly greater than those of large firms with equivalent beta risk
for at least two years. After controlling returns for any E/P effect, a strong
firm size effect still emerged. But, after controlling returns for any market
value effect, a separate E/P effect was not found. Hence, while an E/P
anomaly and value anomaly exist when each variable is considered
separately, the two anomalies seem to be related to the same set of missing
factors, and these factors appear to be more closely associated with firm size
than E/P ratios.
One must surely conclude that alternative models of capital market
equilibrium ought to be seriously considered and tested. For evidence in this
study clearly demonstrates that, at least for portfolios based on firm size or
E/P ratios, the simple one-period capital asset pricing model is an inadequate
empirical representation of capital market equilibrium.

Ball, R., 1978, Anomalies in relationships between securities’ ytelds and yteld-surrogates, Journal
of Financial Economics, June-Sept.
Banz, R.W., 1978, Limited diversttication and market equtlibrium: An empirical analvsis. Ph.D.
dissertatton (University of Chicago, Chicago, IL).
Basu, S., 1977, Investment performance of common stocks in relatton to their price-earnmgs
ratios: A test of the efficient market hypothesis, Journal of Finance, June.
Black, F., 1972, Capital market equthbrium wtth restrtcted borrowing, Journal of Busmess, July.
Brown, P. and J.W. Kennelly, 1972, The informational content of quarterly earnings: An
extension and some further evidence, Journal of Business, July.
Brown, S.L., 1978, Earnings changes, stock prtces, and market efficiency, Journal of Finance,
March.
Foster, G., 1977, Quarterly accounting data. Ttme-serves properties and predictive-abihty results,
Accounting Review, Jan.
46 M.R. Reinganum, Anomalies concerning size und P/E ruttos

Jones, C.P. and R.H. Lttzenberger, 1970, Quarterly earnrngs reports and mtermedtate stock price
trends, Journal of Finance, March.
Joy, O.M., R.H. Lttzenberger and R.W. McEnally, 1977, The adjustment of stock prtces to
announcements of unanttcipated changes in quarterly earnmgs, Journal of Accountmg
Research, Autumn,
Latane, H.A. and C.P. Jones, 1977, Measurmg and usmg standardized unexpected earnmgs,
American Finance Assoctation, Annual meeting m New York.
Latane, H.A. and C.P. Jones, 1977, Standardtzed unexpected earnmgs --~ A progress report,
Journal of Finance, Dec.
Latane, H.A., C.P. Jones and R.D. Rteke, 1974, Quarterly earnings reports and subsequent
holdmg period returns, Journal of Busmess Research, April.
Lintner, J., 1965, The valuatton of risk assets and the selectton of rtsky investments m stock
portfohos and capttal budgets, Review of Economtcs and Stattsttcs, Feb
Nicholson, S.F., 1960, Priceearnmgs rattos, Financtal Analysts Journal, July-Aug
Ntcholson, S.F., 1968, Prtce-earnmgs m relatton to Investment results, Fmancial Analysts
Journal, Jan.-Feb.
Remganum, M.R., 1979, Misspecdicatton of capital asset prtcmg: Empirical anomalies based on
earning yields’ and market values, Ph.D. dtssertation (University of Chtcago, Chicago, IL).
Roll, R., 1977, A crtttque of the asset pricing theory’s tests, On past and potential testabthty of
the theory, Journal of Financial Economtcs, March.
Scholes, M. and J. Wilhams, 1977, Estimating betas from non-synchronous data, Journal of
Fmanctal Economrcs, Dee
Sharpe, W.F., 1964, Capital asset prtces: A theory of market equthbrtum under condnions of
risk, Journal of Finance, Sept.
Stambaugh, R.F., 1979, Missing assets, measuring the market, and testmg the capttal asset
pricing model, Mimeo. (Umverstty of Chtcago, Chtcago, IL).
Theil, H., 1971, Principles of econometrics (Wiley, New York).
Watts, R.L., 1978, Systematic ‘abnormal’ returns after quarterly earnings announcements.
Journal of Financial Economtcs, June-Sept