loumal of Financial Enomics 22 (1988) 27-59.

North-Holland

MEAN REVERSION IN STOCK PRlCES

Evidence and Implications*

James M. POTERBA
Msachuselts lnsll ule 0/ Technolo" Cambridge, MA.01l 39, USA
Nalional Bureau 0/ Economic Research, Cambridge, MA 02138, USA

Lawrence H. SUMMERS
Haroard UniversilY, Boslo MA 02138, USA
Nalional Bureau 0/ Economic Research, Cambridg MA 02138, USA

Received August 1987 , final version received March 1988

Thi paper investigates transitory components in stoc k. prices. After showing that statistical tests
have little power to detect persistent deviations between market prices and fundamental values, we
consider whether pris are mean-reverting, using data from the United States and 17 other
countries. Our point estimates imply positive autocorrelation in returns over short horizons and
negative autocorrelation over longer horizons, a1 though random-walk price behavior cannot be
r ected at nventional statisticallevels. Substantial movements in required retums are needed to
account for these correlation pattems. Persistent , but tran tory dispa ties between prices and
fundamental values could so explain our findings.

1. Introduction

The extent to which stock prices exhibit mean-reverting behavior is crucial

in assessing assertions such as Keynes' (1 936) that all sorts of considerations
enter into market valuation which are in 00 way relevant to the prospective
yield' (p. 152). If market and fundamental values diverge, but beyond some
range the d iffl
rencc
e s are eliminated by spec
e la
t
iv
ve
fo
rc

e s then stock prices
will revert to their mean. Retums must be negatively serially correlated at

.We are grateful to Changyong Rhee , lelf Zwcibel and , especially, David Cutler for excellent
rescarch assistance, to Ben Be rnanke , Fischer Black, Olier Blanchard , lohn Campbell , Robert
En gle, Eugene Fama, Terence Gorman, Pete Kyle , An drew Lo, Greg Mankiw, Robert Merton ,
lu \i o Rotemberg , Kenneth neton Mark Watson , an anonymous referee, and the editor,

Wi1l iam Schwert, for helpful comments , and to lames Darcel d Matthew Shapiro for data
assistance. Thi s research was supported by the National Science Foundation and a Batterymarch
Financial Fel10wship to the first author, and is part of the NBER Programs in Economic
Fluctuations and Fi nancial Markets. A data appendix is on file with the ICPSR in Ann Arbor ,
Michigan.

03 04- 405X/88/S3.50@1988 , Elsevier Science Publishers B. V. (North-Hol1 and)

 i
J.M. Polerba and LH. Summers, Mean rl't' ers;on;n slock prices 29
28. J.M. Polt! rba and LH. Summers, Mean ret' ers;oll ;n slock pr;ces

1~
j
a
conclusion discusses some implications of our results and directions for future
some frequency if erroneous market moves are eventually corrected. 1 Merton

ji1R 3 a*!***i
research.
(1987) notes that reasoning of this type has been used to draw concI usions
about market valuation from failure to r ect the absence of negative serial
correlation in returns. Conversely , the presence of negative autocorrelation
2. MethodologicaI issues involved in testing for transitory components

2--:
may signal departures from fundamental values , although it could also arise

from variation in risk factors over time. A vast literature dating at least to Kendall (1953) has tested the efficient-

t;1
Our investigation of mean reversion in stock prices is organized as follows. marketsjconstant-required-returns model by exarnining individual autocorre-

Section 2 evaluates alternative statistical tests for transitory price components. lations in security returns. The early literature, surveyed in Fama (1970) ,
We find that variance-ratio tests of the type used by Fama and French (1 986a) found little evidee of patterns in secu ty returns and is frequently adduced
and Lo and Mac Ki nlay (1988) are cI ose to the most powerful tests of the null in support of the efficient-markets hypothesis. Recent work by Shiller and
hypothesis of market efficiency with constant required returns against plausi- Perron (1 985) and Summers (1 986) has shown that such tests have relatively
ble alternative hypotheses such as the fads model suggested by Shiller (1984) little power against interesting alternatives to the null hypothesis of market
and Summers (1986). These tests nevertheless have little power, even with efficiency with constant required returns. Several recent studies using new tests
monthly data for a 60-year period. We concI ude that a sensible balancing of for serial dependence have nonetheless rjected the random-walk model. 3
Type 1 and Type 11 errors sgests using critical values above the conventional This section begins by describing several possible tests for the presence of
0.05 level. stationary stock-price components, incI uding those used in recent studies. We
Section 3 exarnines the extent of mean reversion in stock prices. For the then present Monte Carlo evidence on each test s power against plausible
U.S. we analyze monthly data on real and excess New York Stock Exchan2e alternatives to the null hypothesis of serially independent returns. Even the
(NYSE) returns since 1926, as well as annual returns data for the 1871-1985 most powerful tests have little power against these alternatives to the random
.period. We also analyze 17 other equity markets and study the mean-reverting walk when we specify the conventional size of 0.05. We concI ude with a
behavior of individual orate securities in the U.S. The results consistentlv discussion of test design when the data can only weakly differentiate alterna-
suggest the presence of transitory components in stock prices , with returns tive hypotheses , addressing in partic ar the degree of presumption that should
showing positive autocorrelation over short periods but negative autocorrela- be accorded to our null hypothesis of serially independent returns.
tion over longer periods.
Section 4 uses our variance-ratio estimates to gauge the significance of
transttory prce components. For the U.S. we find the standard deviation of 2.1. Test methods
the transitory price component varies between 15% and 25% of value, depend-
ng on our assumption about its persistence. Th e point estimates imply that Recent studies use different but related tests for mean reversion. Fama and
transltory components account for more than half of the monthly return French (1 986a) and Lo and MacKi nlay (1988) mpare the relative variability
variance, a finding confirmed by international evidence. of returns over different horizons using variance-ratio tests. Fama and French
Section 5 investigates whether observed patterns of mean reversion and the (1988b) use regression tests that also involve studying the serial correlation in
associated movements in ex ante returns are better explained by shifts in multiperiod returns. Campbell and Mankiw (1987) study the importance of
required returns due to changes in interest rates or market volatilitv or as transitory components in real output using parametric ARMA models. Each
byproducts of noise trading. 2 We argue that it is difficult to account for of these approaches involves using a particular function of the sample autocor-
observed transit relations to test the hypothesis that all autocorrelations equal zero.
The variance-ratio test exploits the fact that if the logarithm of the stock
price, including cumulated dividends, follows a random walk, the return
ISlochaslic speculalive bubbles, considered by Blanchard and Walson (1982) , could creale
devialions belween marke l. prices and fundamenlal values wilhoul negative 'serial correlalion in
relums. In Ihe. presence of any Iimils on valuation errors sel by spe alors or real inveslmenl 3 Fama (1976) acknowledges the difficty of distinguishing the random-walk model from some
opporlunilies , however, such bubbles could 001 exist allemative specifications. In addition to the recent work of F1a and French (1988b) and Lo d
2 Noise Iraders are inveslors whose demands for seclrilies are besl trealed as eXOl1.enous. ralher
MacKinlay (1988), 0 Brien (1 987) demonslrales the presence of negative serial correlation at ve
long (up to twenly-year) horizons. Huizinga (1987) prodes a speclr. inlerpretalion of the
Ihan Ihe. result of maximizing a conventional utilily function u sin
g
raaiIlon

n a1
e Xp
e
:Ia
Ilon

1
variance-ratio estimalor and reporls evidence that exchange rates also show long-horizon devia-
lu

m
l1m
d ist u
tion.
B lac
k (1986
) Campbe Jl and Kyle (1986) , De Long el . (1987) , and Sler
6
(1984) discuss a viety of possible models for noise Irader behavior tions from random-walk behaor.

30 I.M. Poterba and LH. Summers, Mean reversion in stock prices
variance should be proportional to the return horizon. 4 We study the variabil-
ity of returns at different horizons, in relation to the variation over a one-year
period. 5 For monthly returns, the variance-ratio statistic is therefore
I
where
k-l
R~ = L R t-i,
;-0
,
R denoting the total return in month t. Thi s statistic converges to unity if
returns are uncorrelated through time. If some of the pce variation is due to
transitory factors , however, autocorrelations at some lags will be negative and
the variance ratio will fall below one. The statistics reported below are
corrected for small-sample bias by dividing by E[ VR(k)].6
The variance ratio is closely' related to earlier tests based on estimated
autocorrelations. Using Cochrane s (1 988) result that the ratio of the k-month
return variance to k times the one-month return variance is approx.i mately
equal to a linear combination of sample autocorrelations, (1) can be written
(k) 1+2 ()Pj- 2 () (2)
The variance ratio places increasing positive weight on autocorrelations up to
and including lag 11 , with declining positive weight thereafter. Our variance
ratios for k-period annual returns place declining weight on all autocorrela-
tions up to order k.
A second test for mean reversion, used by Fama and French (1988b),
regresses m
~!esting the~ relationship. between the variability of retums at dilferent horizons has a long
tradition: see Osbome (1959) and A1exander (1961).
SWe use twelve-month retums in the denominator of the variance ratio to permit comparability
with our results using annual retums data. With annu data. the variance.ratio denominator is
var( R ,).
6 Kendall d Stuart (1976) show that under weak restrictions, the expected value of the }th
sarnple autocorrelation is -lj(T - j). Using this r t we mpute VR(k)). When the horizon
the variance ratio is large in relation to the sample size, this can be substantially less than uty.
For exarnple, with T - 720 d k - 60, the bias is -0.069. It rises to -0.160 if k - 120. Detaild
Monte Carlo analysis of the varian-ratio statislic may be found in Lo and MacKinlay (1988).
I.M. Poterba and LH. Summers, Mean reversion in stock pces 31
de-meaned k-period return, the regression coefficient is
k= k(kfRtk)/k(_k)2
/
l
I
l
n
/
s statistic applies negative weight to autocorrelations up to order 2k/3 ,
followed by increasing positive weight up to lag k , followed by decaying
positive weights. 7 Fama and French (1988b) report regression tests because
they r ect the null hypothesis of serially independent returns more strongly
than the variance-ratio tes t. Thi s is the result of the actual properties of the
returns data, not a general rule about the relative power of the two tests. We
show below that returns display positive, then negative, seal correlation as
the horizon lengthens. In this case the regression test , by virtue of its negative,
then positive, weights on sample autocorrelations, will r ect the null hypothe-
sis of serial independence more often than the variance-ratio tes t.
A third method of detecting mean reversion involves estimating parametric
time-series models for returns, or computing likelihood-ratio tests of the null
hypothesis of serial independence against particular parametric alternatives.
Because returns are nearly white noise under both the null hypothesis and the
alternatives we consider, standard ARMA techniques often fai l. 8 When they
are feasible, however, the Neyman-Pearson lemma dictates that the like \i-
hood-ratio test is the most powerful test of the null of serial independence
against the particular alternative that generated the data, so its Type 11 error
rate is a lower bound on the error rates that other tests with the same size
could achieve. In practice, this bound is unlikely to be achieved , since we do
not know the precise data-generation process.
2.2. Power ca/cu/ations
We analyze the power. of tests for transitory components against the
alternative hypotheses that Summers (1986) suests where the logarithm of
stock prices (p,) embodies both a permanent (p ,*) and a transitory (u ,)
1 Further details on the relationship between regression tests and the sample autocorrelogram
are presented in an earlier draft, available on reques t.
8 We tried estimating ARMA models for the pseudo- retums generated in our Monte Carlo
study. Ahhough these data were generated by an ARMA(I , 1) model with first.order autoregres-
sive and moving-average c lllcients of roughly equal but opposite signs. standard ARMA
estimation packages (i.e. , RATS) had dilllculty recovering this process. For example. with
three-quarters of the variation in retums due to transitory factors. the estimation package
encountered noninvertibilities in the moving-average poJynomial and therefore brQke down in
more than a third of I Monte Carlo runs. Less than 10% of the cases led to welI-estimated
pararneters that were c1 0se to those from the data-generation process.
 32 J.M. Poterba and LH. Summers. Mean revers;on ;n stock pr;ces
J.M. Poterba and LH. Summers. Mean revers;on ;n stock pr;ces 33
componen t. We assume that p , = p ,. + ",. If the stationary component is an
AR(l) process Table 1
Simulated Type n error rates of a1 temative tests for transitory components in security returns
", = P1" ,-1 + (4) Each row describes the statistica1 properties of a partiular test for mean reversion. AII tabulations
are based on one set of 25)() Monte Carlo experiments using 720 monthly returns generated by
the process described at the column heading. Both underlying processes are ARMA(l , 1), with
and E/ = P/. - P/":-l denotes the innovation to the nonstationary component, parameters set by B. the share of return variation due to transitory components , and pj , the
then monthly serial correlation of the transitory comonen t. Each test we analyze has s 0.05.
Parameters of retum-generating process

.1 P/ = E ,+ (1 - L )(1 - P1 L) -1.
t
,
I

/ ,- 0.98 ,- 0.98

p B- 0.25 p ,,- 0.15
Test statistic
If p/ and E/ are independent, .1 p, follows an ARMA (1, 1) process. 9 Th is and return Type 11 Mean value Type 11 Mean value
measurement error of test error of test
description of returns allows us to capture in a simple way the possibility that interval rate stabsbc rate statJstJ c
stock prices contain transitory, but persistent, components- The parameIer Pl
First-order
determines the persistence of the transitory component, and the share of autocorrelation 0.941 -0.002 0.924 -0.001
rum vaation due to transitory factors is deterrnined by the relative size of Variance ratio
0.- ana O.~.
24 months 0.933 0.973 0.863 0.927
We perform Monte Carlo experiments by generating 25 ,OOO sequences of 36 months 0.931 0.952 0.844 0.867
720 returns, the number of monthly observations in the Center for Research in 48 months 0.929 0.935 0.839 0.815
60 months 0.927 0.920 0.820 0.771
Securities Prices (CRSP) data base. lO We set = 1 so that the variance of 72 months 0.925 0.906 0.814 0.733
retums (.1 p/) equals 1 + 2/(1 + P1) and ~t parameters for the return-gener- 84 months 0.927 0.894 0.814 0.700
ating process by chsing P1 and 8 = 2 /(1 + P1 + 2 ). Th e parameter 8 96 months 0.929 0.884 0.813 0.670
denotes the share of return variance accounted for by the stationary compo- Retum regression
12 months 0.933 -0.044 0.863 -0.0119
nent; and P1 deterrnine 0.2. We consider cases where 8 als 0.25 and 0.75. 24 months 0.929 -0.080 0.842 -0.158
We set P\ equal to 0.98 for both cases, implying that innovations in the 36 months 0.929 -0.112 0.841 -0.210
lransltory price component have a half-life of 2.9 years. 48 months 0.934 -0.141 0.856 -0.250
60 months 0.934 -0.167 0.868 -0.2112
In evaluating Type n
error rates , the probability of failing to reject the null 72 months 0.941 -0.194 0.8117 -0.3011
hypothesis when it is false, we use the empiricaI distribution of the test statistic 84 months 0.941 -0.221 0.903 -0332
generated with 8 = 0 to deterrnine the critical region for a one~sided 0.05 test 96 months 0.943 -0.250 0.914 -0.354
of the random-walk null against the mean-reverting alternative. Th e panels of LR test 0.924 1.4 0.7 4.491
table 1 report Type II error rates for each test when the data are generated by
the pr@ess indicated at the column head. The mean value of the test statistic
under the a Itemative hypothesis is also reported.
test has rninimal power against the a Iternative hypotheses we consider. Th e
Th e first row in table 1 analyzes a test based on the first-order autocorrela-
Type II error rate for a size 0.05 test is 0.941 (0.924) when one-quarter
tion coefficient. As Shiller and Perron (1985) and Summers (1986) observe, this
(three-quarters) of the variation in returns is from the stationary component
9le parameters of the ARMA(l ,l) model (1 - L)dp - (l + 9L) are (i. e. , 8 = 0.25 and 8 = 0.75).
-p" The next panel in table 1 considers variance-ratio tests comparing return
variances for several different horizons , indexed by k , with one-period return
9-{-(1+ )-2+ (1- p )[4 + (1 + '/2} /(2 +2p) variarlees. The variance-ratio tests are more powerful than tests based on
a;--(p+ )/9. first-order autocorrelation coefficients , but they still have little power to detect
IO~n p.ra~tice we draw 720 pairs of random variables, associate them with (e" ) and then persistent, but transitory, return components. When one-quarter of the return
truct dp,. variation is due to transitory factors (8 = 0.25) , the Type II error rate never
falls below 0.8 1. It is useful in considering the empirical rests below to note
34 I.M. Poterba and LH. Summers, Mean reversion in stock prices
1. M. Poterba and LH. Summers, Mean reversion in stock prices 35
that when the transitory component in prices has a half-life of less than three
years and accounts for three-quarters of the variation in returns (l = 0.75) , the 14
variance ratio at 96 months is 0.67.
--Homoskedas ic Relurns
The next panel in table 1 shows Type 11 error rates for the long-horizon
12
----He eroskedas ic Relurns
regression tests. Th e results are similar to those for variance ratios, aIthough ir. 10
>
the regression tests appear to be somewhat less powerful against our alterna-
8
tive hypotheses. For example, the best variance-ratio test against the l = 0.25
case has a Type 11 error rate of 0.925 , compared with 0.929 for the most 6
powerful regression tes t. @
4
The final panel of the table presents results on likelihood-ratio tests. ll
Although these are more power than the'variance-ratio tests, with Type 11 2
error rates of 0.922 in the l = 0.25 case and 0.760 in the l = 0.75 case, the
o
error rates are still high. Even the best possible tests therefore have Ii ttle power
to distinguish the random-walk model of stock prices from alternatives that Variance Ralia
imply highly persistent, yet transitory, price components. Fig. 1. Empiric distribution of 96-month variance-ratio statistic with homoskedastic d heter
One potential shortcoming of our Monte Carlo analysis is our assumption skedastic retums.
of homoskedasticity in the return-generating process. To investigate its impor- The solid curve shows the empirical distribution of the 96-month variance~ratio .statisti~ cal~~
tance , we fit a first-order autoregressive model to montbly data on the lated from 25 )() replications of 72O-observation time series under the null bypothe~is of se?Iy
logarithm of volatilityP We expand our Monte Carlo experiments to aIlow independent draws irom an identical distribution: lbe bro~en. cu~e pre~nts .~ si~la~ e~pirical
distribution calculated from the same number of Monte Carlo draws, but allowing for hetero-
to vary through time according to this process. Th e Type 11 error calculations skedasticitv in the simulated retums. The logarithm of the simulated retum variance evolves
from the resulting simulations are similar to those in table 1. Fig. 1 illustrates through time as noted in footnote 12
this, showing the empirical distribution function for the 96-month variance
ratio in both the homoskedastic and heteroskedastic cases.

- - Li kelihood R ia
----- Variance Rali

2.3. Eua/uating statistica/ significance .... Regression Bela
@
For most of the tests described above, the Type 11 error rate would be

between 0.85 and 0.95 if the Type 1 eor rate were set at the conventional 0.05 @a M
level. Leamer (1978) echoes a point made in most statistics courses when he >
LF

writes that the [popular] rule of thumb, setting the significance level arbi-
trarily at 0.05 , is ... deficient in the sense that from every reasonable viewpoint
the significance level should be a decreasing function of sample size' (p. 92).
For the case where three-quarters of the return variation is due to transitory
11
le likelihd value under each hypothesis is evaluated using Harveys (1981) exact maximum
Ii kelihood method. Because estimating the mean induces a small-sample bias toward negative
autocorrelations, even under the null hypothesis of serial independence the mean Iikelihood ratios
for each altemative hypothesis are above one
12
The estimated volatility process that we use for our simulations is
0.2
o
Type 1 Errar
Fig. 2. Type 11 versus Type 1error rates for three altemative tests of mean reversion.
Each curve displays the tradeoff between Type 1 and Type 11 er~or r~tes .for ~ pa.rtic~lar te.s~ o!
mean reversion-in-stock retums. Critical reglons for each test are found using simulated empirical
dis ributions for the variance-ratio. regression-beta. and likelihd-ratio tests under the null
hypothesis of serially independent, homoskedastic retums. The ~yp~ 1I erro: rate fo~ ~ac~ tes!
u~der the altemativ hypo hesis of 8 - 0.75 , P\ - 0.98 is calculated using another set ofsiIllulated
o

log( 0 ,2) - - 2.243 + 0.7689 .Iog( 1) + ..
where has a nonnal distribution with mean zero and standard de ation 0.691. The monthly
volatility data are described in French, Schwert. and Stambaugh (1987).
empirical distributions: Under both the nu\l and the altemative hypothesi~ t~e empi?ca~ distribu-
lions are cCat using 25 )() replicalions of 72O-observation time series for synthetic re ms.
For variance-ratio, reJl.ression-beta, Oand likelihood-ratio tests with glven Type 1 error rates shown
ong the horizon . the figure shows the associated Type 11 error rate against the altemative
hypothesis.
 J.M. Poterba and LH. Summers. Mean reversion in stock prices 37
36 J.M. Poterba and LH. Summers. Mean retJersion in stock prices
Tablc 2
factors , fig. 2 depicts the attainable tradeoff between Type 1 and Type 11 errors Variance ratios for U.S. monthly dat a, 1926-1985
for the most powerfl variance-ratio and regression tests, as well as for the CalculatioDS are bascd OD the monthly returns for the value- weighted and ual-weighted NYSE
likelihood-ratio tes t. le Type 11 error curve for the variance-ratio test lies portfolios. as rcported in the CRS!, monthly retums file. The variance-ratio statistic is defined as
VR(k) - (12/k).v R k )/var( RI2 ) where RJ denotes returns over a j-period measurement
between the frontiers attainable using regression and likelihood-ratio tests. For interval. Values in parenthescs are Montc Carlo estimates of the standard errpr of the variance
the variance-ratio test, a 0.40 significance level is appropriate if the goal is to ratio, bascd on 25;)() replications under the null hypothesis of serially independent retums. Each
varian ratio is corrected for small-samplc bias by diding by the mean value from Monte Carlo
minimize the sum of Type 1 and Type 11 errors. To justify using the conven-
experiments under the null hypothesis of no scrial correlation.
tion aI 0.05 test , one would have to assign three times as great a cost to Type I
as to Type 11 errors. Retum measurcmcnt intcrval
An nual retum
Since there is little theoretical basis for strong attachment to the nu l1 standard 24 36 48 60 72 84 96
hypothesis that stock prics fo lIow a random walk, significance levels in excess Data scrics dcviation month months months months months mIths months months
of 0.05 seem appropriate in evaluating the importance of transitory compo-
Valuc-weighted 20.6% 0.797 0.973 0.873 0.747 0.667 0.610 0.565 0.575
nents in stock prices. Many asset-pricing models, involving rational and real returns (0 150) (0.108) (0.1 7 (0.232) (0.278) (0.320) (0.358) (0.394)
irrational behavior, suggest the presence of transitory components and time- Valuc-weiIted 20.7% 0.764 1.036 0.989 0.917 0.855 0.781 0.689 0.677
varying returns. Furthermore, the same problems of statistical power that exss returns (0.1 5 (0 .1 08) (0.1 77) (0.232) (0.278) (0.320) (0.358) (0.394)
plague our search for transitory components complicate investors' lives, so it Eq u-wei tcd 29.6% 0.809 0.963 0.835 0.745 0.642 0.522 0.400 0.353
may be difficult for speculative behavior to eliminate these components. Th e real returns (0.1 5 (0.108) (0.1 77) (0.232) (0.278) (0.320) (0.358) (0.394)
onJy solution to the problem of low power is the collection of more data. In Equal-weighted 29.6% 0.785 1.010 0.925 0.878 0.786 0.649 0.487 0.425
the next section , we bring to bear as much data as possible in evaI uating the excess relurns (0.150) (0.108) (0.1 7 (0.232) (0.278) (0.320) (0.358) (0.394)
importance of transitory components.

3. Statistical evidence on mean reversion

This section uses variance-ratio tests to analyze the importance of stationary
components in stock prices. We analyze excess and real returns using four
major data sets: monthly retums on the NYSE for the period since 1926,
annual returns on the Standard and Poor s-Cowles stock price indices for the
ptod since 1871 , post-World War 11 monthly stock retums for 17 stock
markets outside the U.S. , and retums on individual firms in the U.S. for the
post-1926 period.
3.1. Month NYSE returns, 1926-1985
We begin by analyzing monthly retums on both the vaI ue-weighted and
equal-weighted NYSE indices from the CRSP data base for the 1926-1985
peod. We consider excess retums with the risk-free rate measured as the
Treasury bill yield, as well as real retums measured using the Consumer Price
Index (CPI) inflation rate. le variance-ratio statistics for these series are
shown in table 2. We confirm the Fama and French (1988b) finding that both
real and excess retums at long horizons show negative serial correlation.
Eight-year returns are about four rather than eight times as variable as
one-year retums. Despite the low power of our tests, the null hypothesis of
serial independence is rected at the 0.08 level for value-weighted excess
retums and at the 0.005 level for equal-weighted excess retums Mean .u
reversion is more pronounced for the equal-weighted than for the value-
weighted retums, but the variance ratios at long horiwns are well below unity
for both.
The vaance ratios also suggest positive retum autocorrelation at horizons
shorter than one year. The variance of the one-month retum on the equal-
weighted index is only 0.79 times as large as the variability of twelve-month
returns implies it should be. A similar conclusion applies to the value-weighted
index. This finding of first positive then negative serial correlation parallels Lo
and MacKi nlay s (1988) result that variance ratios exceed unity in their weekly
data , whereas vaan ratios fa l1 below one in other studies concemed with
longer horizons. 14
One potential difficulty in interpreting our finding of positive serial correla-
tion at short horizons concems nontrading effects. If some of the securities in
\3lese p-vues are calculated from Ihe empiric distribulion of our tesl statistic. based on
Monle Carlo resu1ts. They permit rejeclion at lower levels than would be possible using the
normal appromation to the distribution of the variance ratio, along with the Monte Carlo
estimales of the standard devialion of the variance ratio. Further details are available on reques t.
14Prench and Roll (1986) apply vari-ratio tests to daily retums for a sample of NYSE and
AMEX stocks for thc peri 1963-1982. They find evidence of negative serial rrelation.
especiIy among smaller securities. The divergence between their fndings and thosc of Lo and
MacKi nlay (1988) is presumably due to dilferens in the Iwo data scts
 38 J.M. POlerba and LH. Summers, Mean reuersion in slockpces J.M. Polerba and LH. Summers, Mean reuersion in sl kprices 39

the market index trade infrequently, returns will show positive seri aJ correla- Table 3
tion. We doubt this explanation of our results since we are analyzing monthly Varian ratios for U.S. data, 1871-1985
returns. Nontrading at this frequency is likely to affect only a small fraction of Each entry is a bias-adjusted variance ratio with a mean of l!nity und~! the null hyPothesis. The
securities, whereas accounting for the degree of positive correlation we observe variance-r"atio statisticis defined as VR(k) - (12jk). var( R k )jvar( R 12 ) , where RJ denotes the
retum measured over a j~month interval. Values in parentheses are Monte Carlo standard
would require that one security in ten typically did not trade in a given month. deviations of the variance ratio, based on 25 1)() replications under the null hypothesis of seri
We also investigated the incidence of nontrading in a portfolio similar to the independence le un4erlying data are annu retums on the Standard and Poor's. composite
value-weighted index by analyzing daily returns on the Standard and Poor s stock index, backdated to 1871 using the Cowles data as rorted in Wil50n and Jones (1987).
Index [see Poterba and Summers (1986)] for the period 1928-1986. le Retum measurement interval
first-order autocorrelation coefficient for daily returns is only 0.064, and Annual retum
standard 24 36 48 60 72 84 %
groupng returns into nonoverlapping five-day periods yields a first-order Data series deviation months months mIthS months months months months
autocorrelation coefficient of - 0.009. ls sgests that autocorrelation pat-
terns in monthly returns are not likely to be due to infrequent trading. Excess retums 16.2% 0.915 0.612 0.591 0.601 0.464 0.425 0.441
1871-1925 (0.1 40) (0.210) (0.265) (0.313) (0.358) (0.398) (0.436)
A second issue that arises in analyzing the post-1926 data is the sensitivity
Real retums 17.2% 0.996 0.767 0.806 0.847 0.737 0.737 0.807
of the findings to inclusion or exclusion of the Depression years. A number of 1871-1925 (0.1 40) (0.210) (0.265) (0.313) (0.358) (0.398) (0.436)
previous studies, such as Officer (1973), have documented the unusual behav- Excess retums 1.047 0.922
18.9% 0.929 0.913 0.856 0.821 0.833
ior of stock price volatility during the early 1930s. One could argue for 1871-1985 (0.095) (0.1 43) (0.179) (0.211) (0.24 (0.266) (0.29
excluding these years from analyses designed to shed light on current condi- Real re!ums 19.0% 1.035 0.880 0.876 0.855 0.797 0.769 0.781
tlons, although the sham increas in market voIatility in the last quarter of 1871-19115 (0.095) (0.143) (0.1 79) (0.211) (0.240) (0.266) (0.290)
l987 undercuts this view. The counterargument suggesting inclusion of this
period is that the 1930s, by virtue of the large movements in prices, contain a
great deal of information about the persistence of price shocks. We explored therefore consider real and excess returns based on the Standard and
the robustness of our nndings by tmncating the sample penod at both the Poor's-Cowles Commission stock price indices , revised by Wilson and Jones
beginning and the end. Excludng the first ten yrs weakens the evidence for (1987), which are avalable beginning in 1871. These data have rarely been
mean reversion at long horizons. Th e results for both equal-weighted real and used in studies of the serial correlation properties of stock returns, although
excess returns are robust to the sample choice, with variance ratios of 0.587 they have been used in some studies of stock market volatility, such as Shiller
a 0.736 at the 96-month horizon, but the long-horizon variance ratios on the (1 981).
value-weighted index rise to 0.97 and 1.10, respectively. The one-month The results are presented in table 3. For the pre-1925 period , excess returns
variance ratios are not substantially changed by treatment of the early years. display negative serial correlation at long horizons. For real returns, however,
For the post-1936 period , the one-month variance ratios are 0.782 and 0.825 the pattern is weaker. Although the explanation for this phenomenon is
for value- and equal-weighted reaJ returns and 0.833 and 0.851 for vaJ ue- and unclear, it appears to result from the volatility of the CPI inflation rate in the
equal-weighted excess returns. 1S Truncating the sample to exclude the last ten years before 1900. This may make the ex post inflation rate an unreliable
years of data strengthens the evidence for m C( an revers measure of expected inflation during thisperiod. The two lower rows in table 3
present results for the full 1871-1985 sample period. Both series show negative
3. 2. Historical data for the United States serial correlation at long lags , but real and excess returns provide less evidence
of mean reversion than the monthly post-1925 CRSP data. 16
1e CRSP data are the best available for analyzing recent U.s. expehen@,
but the low power of available statistical tests and data-mining risks stressed 3.3. Equity markets outside the United States
by Merton (1987) suggest the value of examining other data as well. We
Additional evidence on mean reversion can be obtained by analyzing the
15
We so experimented with cmde Iechniques for accounling for tIme-varying stock market behavior of equity markets outside the U.S. We analyze returns in Canada for
~olatility i~ es.tima~ing ~rian~e ~a.tios. Es timating sample autoco;;'elati~~~-wiihJ~-het~~~~ie::i;;ti~~
ity corre~tio.? b~ on ~rench Schwert, and Stmbaugh s (1987) estimate of the previous months 16
The variance ratio for the full sample (1871-1985) period is not a simple wei ted averae of
retum volatility elfectly ruces the weight of the early Depression yes yielding variance-ratio the variance ratios for the two subperiods, pre- and post-1926. The 96-month variance ratios for
estimates closer to unity. the post-1926 period excess and re S&P data, for example, are 0.463 and 0.731 , respectively
 g

Table4
Vari ratios for intemational data on re monthly retums
Each ent reports the variance-ratio statistic (or in the bottom panel, average of variance-ratio statistics) for a particular nation and retum horizon.

--1

The variance rati defined by VR(k) - (12/k)*v ar( R )/v R12 ) , where- RJ denotes the real retum over a j-month measurement interval. Data
under1 ying the variance ratios are real dividend-exc1 usive retums calculated from share price indices in the 1ntemational Monetary Funds'

~SRF %
Jnternational Financial Statistics. For most countries the monthly 1MF data span the period 1957:1-1986:12; other data ranges are noted. Values in
parentheses are Monte Carlo standard deviations of the variance-ratio statistics (and standard deviations for the averages in the last three rows). For
the averages. these are computed lowing for a constant correlation across countries. lf this correlation was estimated to be negative. we assume it is
zer. 1n I cases expt those marked with an asteri (*) the data are monthly averages of daily or weekly values. Th e U.K. data are point-sampled
but on1y at the end of each year. The varian ratios are corrected for the time a;regation induced by averaging closing values of the index within
each month

Return measurement interval

saSaeSss
Retum series Annual retum 1 24 36 48 60 72 84 %
Country/sample standard deviation month months months months months months months months

Cada/1919-1986 20.1% 0.711* 1.055 0.998 0.912 0.799 0.692 0.617 0.575
(capital gains o n1y) (0.141) (0.102) (0.166) (0.218) (0.261) (0.301) (0.336) (0.370)
U.K./1939-1986 20.9% 0.832 0.987 0.868 0.740 0.752 0.807 0.806 0.794
(0.16 (0.125) (0 .1 98) (0.259) (0.317) (0.358) (0.400) (0.440)

:n
Austria/1957-1986 21.4% 0.663 1. 205 1.206 1.1 32 0.864 0.666 0.582 0.502

a:
(capital gains only) (0.214) (0.156) (0 254) (0.334) (0.403) (0.464) (0 .5 18) (0.566)

Belum/1957-1986 17.0% 0.718 1.054 1.1 37 1.1 21 1.060 0.876 0.807 0.776
(capital gains 0y) (0.214) (0.156) (0.254) (0.334) (0.403) (0.464) (0.5 18) (0 .566)
Colombia/1959-1983 21.4% 1.223 0.822 0.743 0.724 0 .5 83 0.477 0.386 0.180
(capital gains on1y) (0.214) (0 .1 56) (0.254) (0.334) (0.403) (0 4) (0.518) (0 .566)
Germany/1957-1986 23.8% 0.610 1.309 1.251 0.987 0.747 0.687 0.581 0.462
(capital gains o n1y) (0.214) (0.156) (0.254) (0 .3 34) (0.403) (0.464) (0.518) (0.566)
Finland/1957-1986 22.1 % 0.504 1.141 1.262 1.3 96 1.463 1.381 1. 215 1.014
(capital gains 0y) (0.214) (0.156) (0.254) (0 .3 34) (0.403) (0.464) (0.518) (0 .566)

%
43

o OMm5 nm
k4
a%
(

O

r
m

1 Km
m! ,

( OJ

mF mF m

, (
tF
t

o63
( y xm

m
o

om O o o M2 On

n %

V&
5
n* ,

ml
o

mv m ( o j R
n

mM
*

tr

1m MXn
(n mmm

ao omm4 %

m% om L B r*ij

9 %40

5 ,

/,a
%

m m

( (
mm

te
W

ii
83

a
L n

J
m L l Lm4 %

m% om Lm
%

R @m 7

m
(m

m ,

I
(o

m
n

a
y
vS
om Lm o mi
oo o& o a8j T
Om%
M

N W R V&

m
//
m E o4 ., m* ., | mw j

----

g
8% om
W

om OOnn o m
v

om
O

om o
m

j
mn R W

h
,

r 3

4 m4 %l
mm

(o

m
.i

g
l

11 -m nm

MrSS
o 3 Oq l %
m

omm jj
%m

1mw
n R lm
) /.

j
(o m* % m3

04

m
&. i

o mm 1 l z % 233 2m u
36

l mX
) ,

R 2@

%6

9
%

(o m m% m ro

m m m m 04
N :

g I

o9 oj %wq o65
o %mM
) , 84

J % % om o 0

/ m3 m
( o3 m %

m m 0
1/

t m

l

t
g

o 1 o4 mm1 o33
3

1om l
X)
-m

% % o mm l 0
S

Em

a mm
m4 % m3 mm
m
M -

m ( 4M
/

m

M

ti

o8 mw

o6 % o8 % oI mm
.

om--m o o

o )
u % % o %
:

1l
-

6 3 n)-)
m m-o
m

4%
-sh

-o

m--m
-o -)
m-o (-
-
---

./1
-A

l1

g - -7-v
- --m -M --y -%
8l occ;
M

-
o

gm

m3 n
;

mu m2 17 m

j g
-

2
u7

l m
./

-gh

---J om
Mm
W

o ; om 8% om

omm l 1m
mu )

9
mm 2m
1

m

m m J 1l 2 )

Average value
7

y inI
m

(excluding U.S.) oX 8n
om j
om o

o lm o 1E

W%

m3 n,, 3mX,, m4

Average value m m ,
u )
(excluding U.S.. Spain)

-s
 42 J.M. Poterba and LH. Summers, Mean revers;on ;n stock pr;ces
the period since 1919, in Britain since 1939, and in 15 other nations for a
shorter postwar period.
Th e Canadian data consist of monthly capital gains on the Toronto Stock
Exchange. The British data are monthly returns, inclusive of dividends , on the
Financial mes-Actuaries Share Price Index. The first two rows of table 4
show that both markets display mean reversion at long horizons. Th e 96-month
variance ratio for the Canadian data is 0.585 , while for the British data it is
0.794. Both markets also display statistically signi6cant positive seal correla-
tion at lags of less than 12 months. For Canada, the one-month variance is
0.718 times the value that would be predicted on the basis of the 12-month
variance. For Britain, the comparable value is 0.832.
The variance ratios for the 15 other stock markets are calculated from
monthly returns based on stock price indices in the International Mo
onetar
v
Fun
I
dsll
nternat/On
1 the relative prices of individual stocks than in the market as a whole, some
yle
l
ds so the reported returns correspond to capital gains alone. To assess the
importanc of this omission , we reestimated the variance ratios for dividend-
exclusive CRSP and British stock market returns. The results, available from
the authors on request, show only minor differences as a result of dividend
omission. For example, the 96-month variance ratio for real value-weighted
CRSP retums inclusive of dividends is 0.575 and that for dividend-exclusive
returns is 0.545. We suspect that yield-inclusive data , although superior to the
returns we use, would affect our results in only minor ways.11
Table 4 presents the vaance ratios for individual countries, based typically
on data starting in 1957. Most of the countries display negative serial correla-
tion at long horizons. In Germany , for example, the 96-month variance ratio is
0 .462; in France it is 0.438. Only three of the fifteen countries have 96-month
variance ratios that exceed unity, and many are substantially below one.
Evidence of positive seriaI correlation at short horizons is also peasive.Onlv
one untq Colombia, has a one-month variance ratio greater ihan unity. The
short data samplg, and associated large standard errors, make it dimcult to
r ect the null hypothesis of serial independence for any individual country.
The similarity of the results across nations nevertheless supports our earlier
finding of substantial transitory price components.
Aver
17.
ln %me cases, me monihly stock lndex data fIom the IFS are time averages of daiIy or weekly
index v~~~~. Wo~I?ng (1.~0) showed tha t. the first difference of a time-a:eraged r~dom ~;;Jk
would exhibit positive seri aJ correlation, with a first-order autocorrelation >efficient of 0.25 as the
number of observaIlOns in the average becomes large. This Wlll bias our estimated vahance ratios-
For the countries with time aggregated data we therefore m our sm aJ l-sample bias correction.
Ins!ead. of ~~_:g..th~ expected. ~~ue of ~e first:()rder autocrrelation to b~c-1/il'=':l')-~h~~
ev aJ uating VR(k)) we use 0.25 -lj(T-1) le reported variance ratios have been bias-
adjusted by dividing by the relting expected vaJue
J.M. Poterba and LH. Summers, Mean revers;on in stock pr;ces 43
aggregated and 0.653 when we exclude Spain, an outlier because of the
unusual pattern of hyperinflation followed by deflation that it expeenced
during our sample period. By averaging across many countries , we also obtain
a more precise estimate of the long-horizon variance ratio, although the
efficiency gain is attenuated because the results for different countries are not
independen t. 18
3.4. /ndividual firm data
Arbitrageurs should be better at trading in individual seuities to correct
mispricing than at taking positions in the entire market to offset persistent
misvaluations. Although we expect transitory components to be less likely in
previous work has suggested that individual stock returns may show negative
seal correlation over some horizons [Lehmann (1987), DeBondt and Th aler
(1985)]. We examine the 82 firms in the CRSP monthly master file that have
no missing return information between 1926 and 1985. This is a biased sample,
weighted toward large firrns that have been traded actively over the entire
period. Firms that went bankrupt or began trading during the sample period
are necessarily exclud.
We mpute va ance ratios using both real and excess returns for these 82
firms. Beca use the retums for different firms are not independent, we also
examine the retums on portfolios formed by buying one dollar of each firm
and short-selling $82 of the aggregate marke t. That is, we examine properties
of the time series R i / - R mt where R mt is the value-weighted NYSE retum.
Table 5 reports the mean values of the individual-firm variance ratios, along
with standard errors that take account of cross-firm correlation. The results
suggest some long-horizon mean reversion for individual stock prices in
relation to the overall market or a risk-free asse t. The point estimates sgest
that 12% of the eight-year variance in excess retums is due to stationary
factors , and the increased precision gained by studying retums for many
independent firms enables us to r ect the null hypothesis that all of the price
variation arises from nonstationary factors. The last row, which reports
variance-ratio calculations using the residuals from market-model equations
estimated for each firm (assuming a constant for the entire period) , shows
18 The standard errors for the cross-country averages aJlow for correlation between the varian
ratios for different tries. If I nations have a constant pairwise correlation 'T between their
variance ratios and these variance ios have constant variance then the expected vaJ ue of the
sample variance of the varian-ratio statist is )- (1 - 'T). Replacing the expected
sample variance with the actu Vue we estimate 'T as 1 - / The variance of the sample
mean for N observations, each with the same variance ;x iut constant cross-correlation' 'T,
is [1 + (N -1) 'TJ/ N. We use our estimate of 'T to evaJuate this expression, generalized to allow
for different sampling variances for different variance ratios on the basis of our Monte Carlo
standard errors from table 4.
 45
J.M. Poterba and LH. Summers. Mean reuersion in stock prices

44 J.M. Poterba and LH. Summers. Mean reuersion in stock prices

less evidence of serial correlation than the results that subtract the market

P
,

(*.
gga&BER

g aa3
(

return. These results suggest that transitory factors account for a smaller share

g m

i
mN

a
[

.
F @
of the variance in relative returns for individual stocks than for the market as a

.)

*
@

.
B .S a

Ng

)
guEua whole.

sg (a

m g( m
g
3.5. Summary
m m @ @

aa -ggQ U-t
S@g
).
F R @
.
)
j

--)

Sau Our point estimates generally suggesl that over long horizons return van-

ance increases less than proportionally with time, and in many cases ey
imply more mean reversion than our examples in the last sectlon, Where
-g US
C ( (

(
@ m m

m
transitorv factors accounted for three-fouIlhs of the variatIOn in returns. Many
). . ).
R @ N [ @

R @
F
. <

)

ZSE8 il
gS

of the results rgiect tile null hypothesis of se independence at the 0 .1 5 level,
.
B .
i
m lNgi

i@
a level that m be appropriate giveu our previous discussion of sze versus
EEE aiSB power tradeoRs. Furthermore, each of the difrerenl types of data we analyze
g ($ {( *m [(

@

nrOvides evidence of departure from serial independence in stock returns.

--) {.

EaZ

{
aken t02ether. the results are stronger than any individual fin 1

.)
)
@ (] .

PSIg

i
4u

<

not by a;Immnmlulchas they would be if the various dFta sets were independent.
T11ere is some tendency for more mean reversIon ln less bro-bas and
{
{-egsi N

E
U

agaauu
-m

( sophisticated equity markets. The U.S. data before 1925 show greater eVIdence

m R
(
g(
gis @ F

N)
R
m

g m
) . a). {. .
of mean reversion than the post-1926 data. The ual-weited portfolio
* {
Ra () . .


{gABu

-@

<
gE%--

NYSE stocks shows more mean reversion than the value-weighted portfio.

<
i

m g In recent years, mean reversion is more pronounced in smaller foreign equty

U a a
uis

g markets than in the U .S.

gi

(-

g(( ( m

@ a
{. {. ). {. ).
@
{. { N @ R

g m
S

s 4. The substantive importance of transitory components in stock prices

Sa
ga

8

PC

This section assesses the substantive importance of mean reversion in stock

a

Ua3u

(N m
g * S m
m m (N

* . g {. ). {. )

prices. One possible approach would involve calibrating models of the class
m

{

--

]
pugg>

Considered in the 6rst section. We do not follow thlS strategy because our
ua

gSgg

<

Q

E3E

6nding of positive autocorrelation over short intervals implies that the AR(1)

SDeci6cation of the transitory component is inappmpdate and because of our

U

a aEiiu

ug g R
-

( (
diHiculties in estimating the ARMA(1 , 1) models implied by this approach

m
8gg> N
*a g S a S F F

{. . ).

{
. Instead, we use an approach that does not require us to specify a pross for
i

)(<
<
)

. S

.
g
.Eg

<

the traIisitory mmponent, but nevertheless allows us to focus on its standard

P@ip@

deviation and the fraction of the one-period return variance that can be

attributed to it.

jSP

m Z --S
M
C%

ug
iRP%{

I9We conjectured that the grealer mean reverslOIlln the equal-weighted than the value-weighted

gi&i

s ;?nm:ibt *;ttszrSne?i$5ieESSlXJxeatr

q;gi Q

a8

ggEag}
a -

~~;;;''-~~i~t~d portfolio. Assuming similar-s:.:e~ ~ove~en!s in th~ permanenomponent of the

ip-a@
Q

two indices, this conjecture can be sted by analyzing the degree of me.an reve~~i~n in the relati~e
ilii
g aka

g

retums on the wo indlCes. nlese Ieturns show positive serial correlatlOn a all lags , contrary to

Raug
%uQt

gU

our conJecture.
g g

 46 J.M. I'o(erhu alld 1.. 11. Summer.\', Meall rellt'rsion in s(o("/< pris J.M. Po(erba and LH. Summers, Mean revt' rsion in s(ock prices 47

We treat thc logarithm of the stock price as the sum of a permanent and a Table 6
trans Jtory cponen t. The permanent component evolves as a random walk Permanent and transitory retum componen U.S. monthly data
nd the transry component follows a stationary process. This decomposi-
Each en ry reports he standard deviation of the transitory component of prices, measured al
ton may bc gl
annu rates (G.) , as well as the share of retum variation due to transitory factors , ca\c u13ted from
to
ry com
1(ponent may reflect fads - speculation-induced de
lations of prices
v eqs. (7 a) and (7b) to match the observed pattem of variances in long- and short-horizon retums
from furmental ves - or it may be a consequence of changes in relired
Th e variance-ratio estimates that underlie this table are drawn from he entires for 96-month
variance ratios for e"cess retums in table 2. The diffent cases of Pl2 (11%) correspond to different
r urns. In cither case, describing stochastic properties of the stationarv assumptions about the 12-month (96- month) autocorrelation in the transitory price componen l.
pnce component is a way of characterizing the part of stock price movements
that IIIOl he explained by changing expectaIions about future cash Rows. Pl2 - 0.0 Pl2 - 0.35 PI2 - 0.70
Ulven our assumptions, the variance of T-period returns is
,
_2 _2
1 - /GI 1 -
_2 ,_2
G,-/l .v 1'- /

Value-weighted excess retums

(J';' = T(J,.2 + 2(1 - PT )(J;, (6) 11% -0.00 9.7% 0.369 12.5 '1. 0.4l 21.6'.(, 0.554
fJW, ,,:,, 0.15 12.3% 0.386 20.5% 0.500
fJW, - 0.30 12 .1% 0.373 19.6'.(, 0.456
where (J,2 is the variance innovations to the permanent price mponent
is the variance of Siamnaq componem , and PT is lhe 7:period auiocone- Equ-weighted
exces re ums
lation the statioy coonen t. Given data on the vaIce of returns over p% O ) 16.11% 0.657 21.7% 0.712 37.7% 0.9116
1'% = 0.15 21.4% 0.687 35.8% 0. !l9<
lwo horlZOIIS T and Tf and assumplions about PT and pr, a pair of equalions 21.0'.(, 0.664 0.1112
11% -0.30 34.2~
h lhF form (6) can k solved to yield estimates of q2 and . Using (J~ for
the vancc of one-period relurns, and VR(T) for lhe 7:period variance :atio
in rckill(1Il lo one-period returns, estimates of o,2 and oj are given hy

n3 = o .i [ VR ( T )( 1 - Pl" ) T - VR ( T')( 1 - Pr ) T l for between 43% and 99% of the variance in equal-weighted monthly returns
(1 - Pr ) T - (1 - PT ) T (7a) depending on our serial correI ation assumption , and it has a slandard devia-
tion of between 14% and 37%. Results for value-weighted returns also suggesl
-

-
-

rji---- --- a substantial , though smaller, transitory componen l. Since other nations and

n--R v T ---------
T T
o--u=
,
l

-I - -o -T -,.
t
historical periods show patterns of variance-ratio decline similar to those in
-

7ll
-4

--
. (7b)
U.S. data, we do not present parallel calculations for them. As one would
expect, nations with 96-month variance ratios lower than those for the U.S.
have Jarger transitory components.
Many pairs of variance ratios and assumptions about the serial C\lrrelation
Table 6 indicates that increasing the assumed persistence of the transilory
properlcs of u, uld he analyzed hy using (7a)-(7b). We begin by postulat-
component raises both its standard deviation and its contribution 10 the relurn
llIg ilull UI lS Serially uncorrelated al lhe horizon of 96 monlhs. For various
degrees of serial C?rrelalion ai other hong ns. we can then eSImaIe the , variance. More persistent transitory components are less able to accounl for
declining variance ratios at long horizons. To rationalize a given long-horizon
vanance of thc tsJtory component. (J;, and the share of the return varialion variance ratio, increasing the transilory component s persistence requires in-
due 10 trranstl

l0

or
1ry ccon
)mlponenls
creasing the weight on the transitory component in relation to Ihe permanenl
vJluesofO u 35 and 0.70 fo IP I2' thc twe
eIve-mon
o
0nt

n1i
h
a
ultoco componen t. Sufficien tI y persistent transitory components will be unable to
fmldmgs are mseive to our choice of P%: we Orl values of 0 , 0.li and account for low long-horizon variance ratios, even if they account for all of the
0.30.
return variation. A transitory component that is almost as persistent as a
Table 6 presents estimales of the standard deviation of the transitorv random walk , for example, will be unable to explain very much long-horizon
omponl in slock pes for the value-weighted and equal-weighted NY mean reverS lO n.
porIfolk1s over the permd l926-I985 for various values of Pl1 assuming Which cases in table 6 are most relevant? As an a priori matter, it is difficult
P% = 0. For lhe equal-welglued POIifollo. the transitory componenl accounts to argue for assuming that transitory components should die out rapidly.
48 J.M. Poterba and LH. Summers, Mean revers;on ;n stock pr;ces J.M. Poterba a'ld LH. Summers , Mean reverson n stk prces 49

gx;gsSRZsce 34: k3Sa$$sb;3;;;:dgta;:

Previous claims that there are fads in stock prices have typically suggested t
half-lives of several years , implying that the elements in the table correspond-
ing to P12 = 0.70 are most relevan t. With geometric decay, this sgests a
half-life of two years. One other consideration supports large values for P12'
For given values of and eq. (6) permits us to calculate PT over any
th
il3z g;l:Z$C:obse fa
acctOl
expected returns that r.
in stock
hav%e transitory
tmd generate to account for the

components
orices. e assume for simplicity that the
horizon. A reasonable restriction, that PT not be very negative over periods of transitory component follows a?
up to 96 months, is satisfied only for cases where P12 is large. For example,
a R(l
ess
1) Pproc a s post
tu lated in Summers (1986). Thi s has the rtue 01

with P96 = 0 , imposing P12 = 0 .3 5 yields an implied autocorrelation for the tractabiliiv, although it is inconsistent with the
observation that actual returns
If required returns snow
stationary component of -0.744 at 36 months, -1. 27 at 60 months , and show poShive, th negative serial correlation.
positive autocorrelation, the1 innovation that raises
- 0.274 at 84 months , In contrast, when P12 = 0.70 and P96 = 0 , the implied required returns will
values of P36 and P60 are 0 .1 68 and - 0 .1 73 , respectively. Similar resu1ts obtain period loss, followed by er
reduce share prices This will induce a holding
for other large values of P12' Thi s is because actual variance ratios decline returns follow an AR(1)
returns. The appendix shows that when required
between long and longer horizons, and as eq. (6) demonstrates , rationalizing
process ,23 ex post returns (R ,) are given by
this requires declining values of PT' If PT starts small, it must become negative
to account for the obseed pattem. Larger autocorrelations at short horizons
do not necessitate such pattems. l+g
Insofar as the evidence in the last section and in Fama and French (1 988b) R.-R 1+;:-P1(1+ ) (- r)
is persuasive in suesting the presence of transitory components in stock
prices, this section s results confirm Sllers (1 981) conclusion that models
assuming constant ex ante retums cannot account for all of the variance in
(1 +;:)-1(1 + )2
stock market retums. Since our analysis does not rely on the present-value - (+1 - r) + rl ' (8)
relation between stock prices and expected future dividends, it does not suffer 1 + ;:~ P1 (I + )
from some of the problems that have been highlighted in the volatility-test
debate. 20
orthogonal to innovatio
where L a serially uncorrelated innovation that is
s revisions 10 expecte<!,
about the future bath of requi returns (~) reft
future dividends. The average dividend d and
S. The source of the transitory component in stock prices dividend growth rate are d
and respectively; in steady state, = d +
;:
Transitory components in stock prices imply variation in ex ante returns. 21
If changes in r uired returns and profits are positively
elated then the
Any stochastic process for the transitory price component can be mapped into
Wlll understate the varlance m
a stochastic process for ex ante returns , and any pattem for ex ante returns asslimntion that F and tt are orthogonal
can be represented by describing the associated transitory price componen t. an& returns I1ded to rationalize
mean reversion in stock prices. lt is
interest rates
examples in which pro6ts and
The central issue is whether variations in ex ante retums are better explained DoSsible to construct theoretical
ZIe negatively related, as in Campbell (1986), but the eI

20.
Shiller s con cI usion that market retums are too volatile to be reconciled with valuation models wiih lime-varylIlg
22Lucas (1978) and Cox, IngersolI, and Ross (l985) Silldy lhe pricing of assets and
assuming constant required retums is controversial; see West (1988) for a survey of recent work. papers , including Black (l986) , Campbell Kyle (l986),
reQUired returns. Several recem ,f noise
DeLong et . (1987) , dS
21 Several recent studies have c)nsidered the extent to which equity retums can be predicted possible lilt1ueIm ers
(1984) , have dlscussed the oF
ow that the ne
security prices d r uired retums. Fama French (l 986b)fac or, and inlerpret
v
using various information sets. Keim d Stambaugh (1986) find that between 8% d 13% of the th1S
lin <l lf
variation in retums for a portfolio of stocks in the bottom quintile of the NYSE can be predicted cormlatlOIl in dlRRent Stocks may be attributable lo a COInFIOn
using lagged information. A much smaller share of the variation in retums to larger companies as support for the time-varying retums view of mean reversion.

mo ;$C;tt & i
:cg positive
nCfelllrn iJt2: The?
S
can be accounted for in this way. Campbell (1987) finds that appro mately 11 % of the variation
ln ex ss retums can be explained on the basis of lagged information derived from the term Lt:;igi3 f
structure. Fama and French (1988a) find that lagged dividend yields can predict a much higher ret1likes the exnected excess to be , exacI parallel belween lhe ilme-vary-
whICh however
fraction of retums over longer horizons. ing ret s model ala the fads model would not hold in this case-
 ~(l

J ..\1. P >ter l>a I L lI. S"''' r - I IIr'('t".fl -k pr;'t".(
1/I
J.M. PI ,'rl> d 1.11. S"1/I'"r ..\f" n'I''r.~1 J. ('r l
positive correlation bet\\ een bond and stock returns suggests either posit \'e or T:lhl,' 7
k negative correlation between shocks to cash tlo\\'s and re ired returns.~"
Anwunt nf .iation in r'q rcd rcturns nl'l'dl'd t" l:lunl f~\r nh.'an f""\'l'r 4.111 in sl 'l prc
Uur assumption that required returns are given hy -r= (1 -p L)' l~ , Eadl cntry answcr th l' qU l'stion: 'If hoth r,'quirl r,'turns mJ pri" fIJ f"ll ir-rJ'r
enables us to re-rite (8) , defining autorcgrl'si'n~ 'ith halr. 1i cs ind at l'J in thl' ro lII argin. anJ th ,' 1nh",unl r 1tl. ""an . f 'n.''fu,ln 11\
L.

ohSl'f\'l'J rcturns is onS1 tcnt ith a pric faJ ith a stanJarJ k llhll\ ) g,'n in th ,' .1\llII n
-~'+I(1 + ;:)-1(1 + f/[l +r-PI(l + J. hcading, what ,luld thc stanJarJ dc\'a ti,'n "f r,'quircJ r,'tulll Ill'l'J hl h\.' hl g.l'n.;fII ~ Ih,' l1 l'
tint.:-s.:n.: procc s for (', '.<I rcturn')' Our 'alulatins "lIIpl th ,' f t Ih:1I ilh AR(I) <<'q \l ir,'l
as rcturns. th,' l.sl rcturn prncl'Ss is givcn h cq. (K). Silllilarl thc pri" f:lJ i a,S\l IIICJ t" f'\l
an AR( 1) that yidds a pr.li,.'CS Iilc (5) for . F'.\'1 fl'lurn c th l'n :1 \\hIt :lhl" f ". r
illlpli tl\1 is nccdcJ t" gl' ncratc a giwn 1Zl' trlIl Slt ,'r\' pnc palt ,'rn illlpli,',1 h n Thc
(l-p ,L)(R , -R) + t , - (1 + J){,_, - P,t'-l' (9) ,'aku l:llions arc cililihr:a
i

The first-order autocovace of the expression 011 the lI -hand side of

tlhc 1n6-19 !1.5 pcri and arc hascd "n cq ( H0)) in th ,' t" XI. nlcC aCfiag" "X I.'I.' flt?tur
pcn.ld i K. 9% pcr ycar. wilh :1 di lknJ icl1 of 4.5'f




rn fllfr tlhli
i
( s nonzero. but a l1 higher-order autocovariances equal ~ero. ~~ Pro
0
1
o{> O. retu
rns fo l1 ow
a n ARMA( l. l) proc
lCess: if C1 = O. then re'turns are i Slaniard Jc\'iali"ll "f trall tl'~ 'llIUpl)n 'nl ,, ,
white noise , Half-lirc lS.O'i- 20.(I~ 25.0} 1 i
The siple model of stationary and no ationary price compol1 ents sum- 1. 4 \lcars 7.9't 1O.6'{. 1.l .."!'I- l \.N'l
marized in eq , (5) also yields an ARMA(1 1) representation for returns , This 1. 9 vears 6 .l't !I.."!'{. 1O.2~ 12.1r
allows us io calculale the varialion in required reIurns that is needed lo 2.9 vears 4.4% 5.!1F 7..1~ 11 ,7r,
generate the same time-series process for ohseed returns as fads of Varil)US
sues- In Ihe appeFdix we show, thaI Ihe required reIurn variaI1 corresponding
10 a given fad variance is
to explai l1 the same size price Cads when required rcturn shod.s arc Icss
C1_2 = [1 + ;: - PI(1 + gW(I - P,)2(1 +;:)2 persisten t. These estimates of the standard deviation of re4uircd relurns are
( 10) large in relation to the mean of (' post excess relurns and imply that if t1I l(
{(1 + J)(l + )- 1 + (1 + ) (1 + g)~ relurns are never negative they must frequently exceed 20%.
It is diftkult to think of risk factors that could accounl for SUdl varialion in
Table 7 reports calculations based on (10) , It shows the standard deviation reqliired returns , Campbel\ and Shiller (1987). using data 011 rcal intcrcsl rates
of required excess reIurns, measured on an annual basis, implied by a variely and market volatilities tnd no evidence that stock pri l'cs h e\ p to f 'rel'asl ,
of fad models. We calibrate Ihe calculations using tl1e average excess RIllrn future movements in discou l1 t rales as they should if stol'k price movcmcnls
(89% per year) on the NYSE equal-weighted share price index over the are caused by tluctuations in these factors. 1f> Although thcy show that slod;
1926-1985 period. The dividend yield on lhese shares averages 45%. implying prices do forecast consumplion tluctuations. the sign is :ounter 10 Ihc hcory 's
an average dividend growlh rate of 44%. We use estimaies of the variance prediction , On the other hand if the transitory componcnts are vicwed as a
rato at 96 months to calibrate the degree of mean reversion. reflection of mispricing. they are also large in r e\ ation to traditional views of
Substantial variability in required returns is needed to explan mean rever- market efficiency ,
SlOn in prices. For example- if we poSullate that the slandard deviain1Il Of Ihe The second diftculty in explaining the ohservcd correlation pallerns with
transtory price component is 20% then even when required return shocks models of time-varying returns arises from our tinding of positive fo l\owed hy
have a half-life of 2.9 years. lhe standard deviation of ex ante returns mllSI be negative serial correlation , Models with first-order aUlorcgrcssive transilory
5.8% per annum. Even larger amounts of required return varialion are needed components (;an rationalize the second but not Ihe tirst of Ihcse ohservations

It is instru ;tive to consider what type of cxpecled retu fIl s hehavior is necessilry
!4Campbell (l 9R7) eslimales Ihallhe correlalion belw~ncxcess relurns on long-tcrni honds anJ to account for both observations.

corporale equ es was 0.22 ror Ihe 1959-1979 pcriod and 0.36 for Ihe morc~:~~';1979-::' 9i
period.
1
---AInky- Spivey. aIId Wrobleski (1977) prove lhal an alllocorrelogram with zero tntries heWIld
order k implies an MA(k) proces~ !bConlrary cvidcncc suggesting Ihal lock rClurns J" prcJicl flllur ,' vnlIl ilil~ pallcrn
providcd hy Frcnch. Sc hwerl. d Slamhallgh (1 7).
 52 J.M. POlerba a,.d LH. Summers, Mea,. reversio,. in slock prices J.M. POlerba and LH. Summers, Mean reversion i,. slock prices 53

There are two potential explanations for the positive autocorrelation in
observed retums at short lags. First, contrary to our maintained specit.cation,
shocks to required retums and to prospective dividends may be positively
correlated. This could lead to positive autocorrelation at short horizons
because increases in expected dividends, which would raise share prices, would
be fo l1owed by higher ex ante retums. We explored this possibility by forming
monthly dividend innovations' (IDI ) for the 1926-1985 peod as the
residuals from a regression of real dividends (on the value-weighted NYSE
portfolio) on twelve lagged values of real dividends , a time trend , and a set of
monthly dummy variables. We then regressed real retums on the value-
weighted index on lagged values of IDI . A representative equation, includ-
ing six lagged values, is shown below. R , is measured in percentage points and
standard errors are given in parentheses:
R ,= 1. 568 + 0.844 IDI V._. - 0 .1 09 IDI V. ~
. (0.04 (1.380) (1. 380) ,- ..
able proxies for ruired retums display such stochastic properties. Studies of
volatility such as French, Schwert, and Stambaugh (1987) or Poterba and
Summers (1986) suggest that shocks are persistent but that their moving-aver-
age representations show declining coefficients. An aitemative possibility is
that movements in required retums are due to changes in the equity demands
of noise traders. For example, assume that the required retum of sophisticated
,
traders is equal to a + where S is the fraction of the outstanding common
stock that these investors must hold. Equity demands of noise traders (which
in equilibrium must equal 1 - ) that follow a moving-average process similar
to one of those for required retums that generate positive, then negative,
autocorrelation in ex post retums will also generate this pattem in ex post
retums. Th e notion that noise trading impulses intensify and then decline
comports with qualitative discussions of fads , but further work is clearly
necessary to evaluate this conjecture.
6. Conclusions

M - m v
mWI
i
l
l
t
--
J

i
v
l
u
--
/
n *
1

Our results suggest that stock retums show positive serial correlation over

- 1.061 IDI- - 1.769 .IDI V. -
(1. 377) . - (1. 374)
R 2 = 0.037 , 1927:7-1985:12.
The coefficients on lagged values of IDIV should be positive if required
retums and prospective dividends are positively correlated , but the results
provide no support for this view. If anything, they suest a negative but
statistically insignit.cant relationship between dividend innovations and subse-
quent returns. This would suggest that positive dividend news is followed bv
lower required retums, a pattern that should be reftected in negative autocorre-
lation of ex post retums over short horizons.
The send potential explanation for positive seal correlation is that the
autocorrelogram of ex post retums reftects the dynamics of required retums.
Some required-retum processes could generate positive, followed by negative,
return autocorrelation. The required-retum processes with this feature that we
have identit.ed all show increasing coefficients in some part of their moving
average representationY We are unaware of evidence suggesting that obser
27
Two exarnples of required retum pr:esses are twelfth-order moving-average processes wi th
the following coefficients: 1, -1.5. -0.75 , -0.5 , -0.5 , 0.75 , 0.75 , 0.75 , 0~75 0.75 , .75 , 0.75. 0.75
an 1, 1. 5, 2, 2.5 , 3, 3.5 , 4, 4.5 , 5, 4, 3, 2. 1. e autocorrelograrn of the former process displays
positive, then negative, rrelation in required rcturns, while the second process exhibits positive
autocorrelation at 1 lags. 80th processes generate positive, then negative, aUlocorrel~tion in
ex pOSI relums.
short periods and negative correlation over longer intervals. This conclusion
emerges from data on equal-weighted and value-weighted NYSE retums over
the 1926-1985 period , and is corroborated by data from other nations and
time periods. Although individual data sets do not consistently permit ec
tion of the random-walk hypothesis at high signit.cance levels, the various data
sets together strengthen the case against its validity. Our point estimates
suggest that transitory price components account for a substantial part of the
vanance m retums.
Our t.nding of signit.cant transitory pce components has potentially im-
portant implications for t.nancial practice. If stock price movements contain
large transitory components, then for long-horiwn investors the stock market
may be less risky than it appears to be when the variance of single-period
retums is extrapolated using the random-walk model. Samuelson (1 988) dem-
onstrates that in the presence of mean reversion, an investor s horizon will
influence his portfolio decisions. If the investor s relative risk aversion is
greater (less) than unity, as his horizon lengthens he will invest more (less) in
equities than he would with serially independent retums. Th e presence of
transitory price components also sgests the desirability of investment
strategies, such as those considered by DeBondt and Thaler (1985), involving
the purchase of securities that have recently declined in value. It may also
justify some institutions' practice of spending on the basis of a weighted
average of their past endowment vnlues, rather than current market value.
Although the temptation to apply more sophisticated statistical techniques
to stock retum data in an efforl to extract more information about the
magtude and structure of transit components is ever present , we doubt
- J.M. POlerba and LH. Summers. Mean reversion
-
that a great deal can be learned in this way. Even the broad characteristics of
in slock prices
-.

and g,+I:
J.M. Polerba and LH. Summers. Mean reversion in slock prices 55

the data exarnined in this paper cannot be estimated precisely. As the debate
ap.
R E((r- D,+ [+j- rl
1

over volatility tests has illustrated , sophisticated statistical results are often
very sensitive to maintained assumptions that are ilifficult to evaluate. We
have validated the statistical procedures in this paper by applying them to
pseudo data conforrning to the random-walk model. Our suspicion, supported

;Igt+l- 1)
by K.leidon s (1986) results , is that such Monte Carlo analysis of much of the
more elaborate work on stock-price volatility would reveal poor statistical +
properhes. (A.2)
We suggest in the paper s final section that noise trading, trading by
investors whose demand for shares is deterrnined by factors other than their D,(1 + r)
U,\J. T'J ,
D (1 + ) ~ f
.~'~"\I'"'=\E{ Ej(r,+j-r]}
l
expected return , provides a plausible explanation for the transitory compo- ;- (1+;)(;:- ) -, \ /-:0'- l 'TJ J}
nents in stock prices. 28 Pursuing this wi\l involve constructing and testing
theories of noise trading, as well as theories of changing sk factors , that could
account for the characteristic stock return autocorrelogram documented here. + E{j[g+j-gl}
Evaluating such theories is lik e\ y to require information other than stock
returns, such as data on fundamental values, proxies for noise trading such as
the net purchases by odd-lot traders , turnover, or the level of participation in
investment clubs, and indicators of risk factors such as ex ante volatilities where = (1 + )/(1 + r). We denote D,(1 + ;:)/(i - g) as P,. In the special
implied by stock options. Only by comparing models based on the presence of case of
noise traders with models based on changing sk factors can we judge whether
financial markets are efficient in the sense of rationally valuing assets, as well ( - r) = Pl(-1 - r) + ~I' (A.3)
as precluding the generation of excess profits.
we can simplify the second term in (A.2) to obtain
Appendix
,
-D (l + )
;:]
Derivation 0/ ex post return process when required returns are AR(l) P.-P. E I -
(1+;:)(;:- ) j~O
The price of a common stock, PI' equals

n + l E{ i(g'+i - l} (A .4)
--

Z ,..,,
t

PI
l
l
E ,I

-- +a D,

1
r

.,

0
,., 7 ,
l
+ (A .l)
l
l
l , jl l

(1 + )(-;:) +;E ( j lO'T
(r- )(I+;:-Pl(l+g))' r-g-'\j~o'-
, vJ}l}
[g'+i-
where +i denotes the required real return in period t + i , D, is the dividend
paid in period t , g ,+i is the real dividend growth rate between pe ods t + i
and t + i + 1, and E ,{ .} designates expectations formed using information Now recall that the holding period return , R" is given by
available as of period t. We lineaTze inside the expectation operator in ,

28
P.d+D. (A.S)
R =

Cutler, Poterba, and Summers (1 988) dolment the difficulty of explaining a significant -1.
fraction of relurn varialion on Ihe basis of observable news aboul Culure cash Hows or discounl
rales.
 56 J.M. Poterba and L H. Summers, Mean reversion in stock prices r

J.M. Poterba and LH. Summers, Mean reversion in stock prices 57

It can be linearized around and P +l as follows: , This ~ields an ARMA(l , 1) representation of retums. Since (1 + ;)/(1 + )'"
n , P'+l-P'+1 ( Pt+ l + D,L
x (1 + d) , this is eq. (8) in the tex t.
R. R+ -- Rz (R-R (A.6)
We now explore the parallel between the time-varying retums model and
the fad model , which postulates that retums evolve according to

where R+ I = (1 + a)R and R = F(1 + a)/(1 + F) Substituting (A.4) into (A.6) (l-PI L )(R ,-R) E,- Pl l!, -1 + - "'-1' (A.1
yields
For this ARMA(l , 1) process to be the same as (A.9), two restrictions must be
+ ) Dt+ l(l satisfied. We find them by equating the variances and first-order autocovari-
R , -R p'(r- g)[l +;- Pl (1 + rr(+1 -;) ance of the right-hand sides of (A.9) and (A .1 0):

[1 + (1 +d)2] + (1 + ) =2 + (1 + )
IJ 0
(A.ll)
+ JEt+ l {g ,+l+J - g}

m + f ----,
+ (1 + d) + Pl=+ Pl. (A.12)

-

-n-+
i
l
--

-
+ ---
-
--
--! 7 --T rv
l
l
l
I - Using (A.12) to eliminate from (A.ll) we find

(1 +;) (1- Pl)2

- I JE {g'+J - g} (A.13)
(A.7) (1 + d)(l + )- 1 + (1 + d)2) 0.-

This can be rewritten as 0;

Recall that thvariance of the fad , , equals /(1- ). Using this and the
definition of ~" we find from (A .1 3) that the variance of required returns
-;:; ~ D,(1 + )2 corresponding to a given fad variance is
R.-R -
- P(r- )(1 +;- Pl(l + ))
? l+;-Pl(l+ )]2(1 - Pl)2(1 + r)1 7
(A .l 4)
x [(+l- r) _-l( - r)] + t , {(1 +d)(l+ )- + (1 + d)2]} (1 + g)2
(A.8)
(1 +)
S leads immediately to (10) in the tex t.
1+; - Pl (I + g)
References
X[ (I -PIL)-I~'+l- (l-PI L )-l +
Alexander, Sidney S. , 1961 , Price movements in speculative markets: Trends or random walks ,
Industri a1 Management Review 2, 7-26
where L reRects changes in expected future dividend growth rates between t Ansley, Craig, W. Al1en Spivey, and Wi1liam Wrobleski , 1977, On the structure of moving averagc
and t + 1, and the last expression exploits the fact thatj1 - PlL)(1; - F) = &I processes, 10uma1 of Econometrics 6, 121-134.
Now defining {(l + )/[1 + ; - Pl (1 + )]H +l = we can multiply Black, Fischer, 1986, Noise , 10uma\ of Finance 41 , 529-543.
through by (1 - P1L) so that Blanchard, Olivier 1. and Mark Watson, 1982, Bubbles, crashes, and rational expectations, in:
P. Wachte1, ed. , Crises in the economic and fincial structure (Lexington Bks Lexington ,
MA) 295-315.
Campbell , 10hn Y., 1986, Bond and stock retums in a simple exchange model , Quarterly 10umal
(1-P 1 L)(- -1\1
;-gj'
-1 + t - p ,t_ -
(A.9) of Economics 102, 785-803
Campbell , 10hn Y. , 1987, Stock retums and the term structure , Jouma1 of Financia\ Economics
18, 373-400.
I 58
- - -
J.M. Poterba al/ d LH. Summers. Mean rer'ersion in stock pr;ces
Campbell, Iohn Y ald Albert s. Kyle, l986, Smart money? I1Olse lradlng, and stock pnce
behavlOr, Unpubllshed manuscripl (Princelon University, PnnCeton, NI)
Campbell, John Y. and N. Gregory 'Mankiw, 1987 , Are ouP fluctuations transtory?, Quarlerly
Joumal of Economics 102 , 857-880
Campben- Iohn Y. and Robert I. shiller, 1987. The dividend-price ralio and expectations of
~uture d.iv~de~~s ~~ discounl factors , Mi meo. (Yale Universiiy, New Hav~;;' c:n:
C@hrane, Jqhn H l988, How big is the random walk in GNP?, lournal oI Pitic Economy.
forthcoming
Cox , John c., Jonathan Ingcrsoll. d S~cp~e.~ ~~s~ 1985 , An intertemporal general equilibrium
model of asset prices, Econometrica 53, 363-384.
Cutler, David C , James M. poieIba. aIId LaWIfIl@ H. Summers, 1988. whal moves slock prices?
Joumalof Portfo\io Management, forthcoming
DeBondt. Werner and Richard Tbaler, l985, D@S the stock market overreact, Journal of Finance
40, 793-805.
DeLong. J. Bradford. Andrei Shleifer, Lawrence Summers. and Robert Waldman, 1987. The
economlc conquences of noise traders. Nation Bureau of Eco nomic Resech working
pap~r no. 2395 (NBER, Cambridge , MA).
Fama. Eugene F. , 1970, Efficient capt markets: A reew of thcory and empirical work , Journal
of Finance 25 , 383-417
Fama , Eugene F. , 1976, Foundations of tnancc (Bic Books , New York. NY)
Fama- Eueene F. and Kennelh R. FIencl1, I986a. Permanent and lemporxy componenls of sl@k
pnces, CUller for Research in Security Prices working paper no. I78 (University of Chicago,
Chicago, IL)
Fama, Eugene F and Kennelh R French, I986b, Common faclors m lhe senal correlation of
sl@k RIllI1ls , Centr for Research in Secllr1Iy PriCs working paper no 200 (UIliverslty or
Chicago. Chicago, IL)
Fa~a Eu~en_e _F. and _~nneth R. Frcnch, 1988a, Di vidend yids and expectcd stock retums ,
Journal of Financial Economics, this issuc.
Fama: Eug~ne F. ,an? _K~.n~e~h_R. Frenc~.t ~88b' Permancnt and tcmporary components of stock
prices. Joumal of Po\i tical Economy 96 , 246-273.
French Kennelh R. and Richard RolI, 1986, Stock retum variances: Th e arrival of information
and Ihe reaclion of Iraders. Joumal of Financial Eonomics 17, 5-26.
French. Kennelh R. , William Schwerl , dRert Stambaugh. 1987, Expected stock retums and
stok market volatility, Joumal of Financial Economics 19. 3-30
Harvey. Andrew. 1981 , Time series mode\s (Philip AII Oxford).
Hu~nga. John , 1987, An empirical investigation'of he lon
J.M. Poterba and LH. Summers. Mea l/ re!'ersio in sfock prices 59
Lucas, Roberl , 1978. Asset prices in exchange economy, Economelri :a 66. 1429-1445.
Merton. Robcrt c., 1980, On estimating Ihe expecled rctum on Ihe markel, Journal of Financial
Economics 8, 323-361.
Merlon , Robert C. , 1987 , On Ihe slale of the efficienl market hypolhcsis n tnancial economcs.
n: R. Dombusch, S. Fischer , and J. Bossons. eds .. Macroeconomcs and tnance: Essays in
honor of Franco Modigliani (MIT Press, Cambridge. MA) 93-124
O'Brien , Janles M. , 1987 , Testing for Iransitory elements n stock prices. Mme. (Board of
Govemors of the Federal Rcserve , Washington. DC).
Officer, Robert R. , 1973 , The variabi \i ty of lhe market factor of Ihc Ncw York Stock Ex change.
Joumal of Business 46. 434-453.
Osbomc, M.F. , 1959, Brownian motion in thc slock markel. Operations Research 7, 145-173
Poterba, James M. and Lawrence H. Summers, 1986, The persislence of volalilily and slock
market retums , American Economic Revew 76. 1142-115 1.
Samuelson, Paul A. , 1988 , Longrun risk tolerance when equily relurns are mean regressing:
Pseudoparadoxes and vindication of business man s risk, Mimeo. (Massachusells lnslilule of
Technology. Cambridge, MA).
Shiller, Robert J. , 1981 , Do stock prices move too much to be juslted by subsequent changes in
dividends? , American Economic Review 71.421-436
Shiller, Robert J. , 1984, Stock pris and social dynamics , Brookings Papcrs on Economic Activily
2, 457-498.
Shiller, Robert 1. and P. Perron, 1985 , Testing the random walk hypolhesis: Power versus
frequency of observation. Economics Lellers 18 , 381-386.
Summers, Lawrence H. , 1986, Does t11e slock markcl raliona \l y reflecl fundamenlal values?
Joumal of Financc 41 , 591-601.
West , Kenncth , 1988, Bubbles, fads. and stock pce volatility: A parlial evalualon. Joumal of
Financc 43 , 639-655
Wilson , Jack W. and Charles Joncs , 1987, A comparison of annual common slock relurns
1871-1925 wlh 1926-1985 , Journal of Business 60. 239-258.
Working, Holbrook , 1960, Note on the corre1a tion of trst dierences of averages in a random
chain, Econometrica 28, 916-918.