You are on page 1of 40

American Economic Association

Efficient Capital Markets and Martingales


Author(s): Stephen F. LeRoy
Source: Journal of Economic Literature, Vol. 27, No. 4 (Dec., 1989), pp. 1583-1621
Published by: American Economic Association
Stable URL: http://www.jstor.org/stable/2727024
Accessed: 06-11-2015 04:52 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.

American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to Journal of Economic
Literature.

http://www.jstor.org

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
Journal of Economic Literature
Vol. XXVII (December 1989), pp. 1583-1621

Efficient Capital Markets and


Martingales

By STEPHEN F. LEROY
Universityof California, Santa Barbara, and Federal Reserve Bank of San Francisco

I have received helpful commentsfrom Frank Diebold, Steven Dur-


lauf, Christian Gilles, Pete Kyle, Stephen Ross, Kevin Salyer, Robert
Shiller, and ChristopherSims. The paper benefitedfromexceptionally
diligent and capable refereeing. I am indebted to Aarne Dimanlig,
Judy Horowitz, and Barbara Bennett for research assistance and
editing. This paper was written while I was a visiting scholar at the
Federal Reserve Banks of Minneapolisand San Francisco. I am grate-
ful to both institutions. Views expressed here are those of the author
and are not necessarily those of the Federal Reserve System or its
staff.

I. Overview comparative advantage is conferred by


differences in information held by inves-
AT ITS MOST GENERAL LEVEL, the theory tors, rather than differences in productiv-
of efficient capital markets is just the ity among producers. The analogue in
theory of competitive equilibrium ap- financial markets of Ricardo's assertion
plied to asset markets. An important idea that absolute advantage is irrelevant is
in the theory of competitive equilibrium the proposition that information that is
is the Ricardian principle of comparative universally available cannot provide the
advantage: England exported cloth to basis for profitable trading rules. Thus
Portugal and imported wine from Portu- if it is generally known that a firm has
gal not because England necessarily had favorable earnings prospects, the theory
an absolute advantage over Portugal in of efficient capital markets says that the
producing cloth, but because England price of the firm's stock will be bid to
produced cloth comparatively more the point where no extranormal capital
cheaply than wine relative to Portugal.' gain on the stock will occur when the
The same idea applies in analyzing equi- high earnings actually materialize.
librium in financial markets. exceDt that Therefore knowledge that earnings will
1 Ricardogracefullyheaded off a criticism of chau- rise in the future does not imply that
vinism by specifying that Portugal had an absolute the stock should be bought now. It is
advantageover England in producingboth cloth and
wine ([1817], 1960, p. 82), rather than England over only differences in information-infor-
Portugal. mation that is not "fully reflected" in

1583

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
1584 Journal of Economic Literature, Vol. XXVII (December 1989)

prices-that confer comparative advan- efficient, they are signaling that at a mini-
tage, and that therefore can form the ba- mum they want to think of asset prices
sis for profitable trading rules. as being determined by the interaction
Most of the lessons of market efficiency of rational agents-that is, as being deter-
are direct consequences of thinking mined as an economic equilibrium-and
about financial asset prices as determined that they see the proposed statement as
by the conditions of equilibrium in com- following from this fact. Frequently,
petitive markets populated by rational however, the term efficient capital mar-
agents. Some of these lessons are obvi- kets carries in addition the presumption
ous. For example, the decision by a com- that the amount of information which is
pany to split its stock (i.e., issue two or publicly available, and which for this rea-
three new shares in exchange for each son cannot be used to construct profitable
old share) should have no effect on the trading rules, is large. In the limit, the
rate of return on this stock. This proposi- doctrine of capital market efficiency con-
tion is a direct corollary of the fact that tains the assertion that individuals do not
in any economic equilibrium the choice in fact have different comparative advan-
of numeraire is arbitrary. Other lessons tages in information acquisition. In such
of market efficiency, however, while ap- a world there are no profitable trading
parently equally direct consequences of rules. This extended meaning of capital
the nature of competitive equilibrium, market efficiency underlies statements
go deeply against the grain of finance such as the following: In an efficient capi-
practitioners and financial journalists. tal market, agents should have no invest-
For example, during the era of conglom- ment goals other than to diversify to the
erate formation in the 1960s and 1970s maximum extent possible so as to mini-
(which, of course, has since been suc- mize idiosyncratic risk, and to hold the
ceeded by the current wave of leveraged amount of risk appropriate to their risk
buyouts accompanied by conglomerate tolerance.
breakups), firms routinely justified ac- The importance of the topic of capital
quiring other firms in unrelated lines of market efficiency is evident. Investors
business on the grounds that the acquisi- have no choice but to base their invest-
tion served to diversify their activities, ment decisions on information. In evalu-
thereby reducing risks to stockholders. ating their information, investors must
In an efficient market, this justification consider not only whether it is accurate,
makes no sense at all. Firms have no but also whether it is generally known-
comparative advantage over individuals in practitioners' parlance, whether it has
in diversifying risk because individuals already been discounted in the market
can diversify risk simply by buying the price. Because the value of information
stock of several firms or the shares of a depends on the extent of its dispersion,
mutual fund that holds many firms' investors' decisions about what informa-
stocks. This example indicates that in fi- tion to acquire depend on whether they
nance, as everywhere else in economics, think capital markets are efficient; to the
economists risk offending entrenched extent that markets are informationally
opinion to the extent that they insist on efficient, acquisition of information is a
taking seriously even elementary conclu- waste of time.
sions drawn from equilibrium analysis. Suppose that capital markets are effi-
When economists defend some state- cient with respect to some information
ment as being a consequence of the fact set (D. Then by definition an individual
(or assumption) that capital markets are investor who acquires information (D

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
LeRoy: Efficient Capital Markets and Martingales 1585

does not gain comparative advantage rational (and who have rational expec-
over his rivals, this information being al- tations) interact when it is common
ready fully reflected in prices. The earli- knowledge that they have different in-
est empirical investigations of capital formation. There is no question that
market efficiency tested this postulated mastery of the asymmetric information
failure of information to confer compara- literature is indispensable to a deep un-
tive advantage by constructing hypothet- derstanding of capital market efficiency.
ical trading rules based on particular in- However, this area, besides having no
formation sets and testing their close connection with martingales, does
profitability under actual securities re- not bear directly on the empirical work
turns. Buying and selling stocks accord- on market efficiency.
ing to some prescribed formula based on Another important literature that bears
1Dshould not result in systematic success on capital market efficiency as defined
if capital markets are efficient with re- above, but which is not discussed in this
spect to FD,but might do so otherwise. paper, is that on portfolio separation.
Although it was insufficiently realized Contrary to the implication in the first
at first, these empirical tests of whether paragraph, it is not generally true that
asset prices fully reflect available infor- only differences in information give
mation also presume the validity of a par- agents reason to trade securities. If fu-
ticular equilibrium model specifying pre- tures markets are incomplete, changes
cisely how information is reflected in in wealth and conditional distributions
prices: the martingale model. Martin- of future returns will in general interact
gales will be defined and described be- with agents' risk aversion so as to induce
low. The fact that the empirical literature them to trade even when there is no dis-
on capital market efficiency is inextrica- agreement about the conditional distri-
bly linked to the martingale model justi- bution of returns. However, under cer-
fies our taking the martingale model as tain restrictions on preferences and
the unifying theme for this survey. It is return distributions it can be shown that
true, however, that there exist branches identically informed agents will hold
of the literature on efficient capital mar- identical, or virtually identical, portfo-
kets that are unrelated to martingales but lios. Under these restrictions it is true
that nonetheless are important to a full that all, or virtually all, differences in
understanding of market efficiency. comparative advantage in holding securi-
These are sketched in the following two ties can be traced to differential informa-
paragraphs. Coverage of these sublitera- tion. The theory of portfolio separation,
tures in addition to the martingale litera- which derives these restrictions on
ture would result in a survey that is dis- agents' optimal portfolios from assump-
jointed and superficial. Accordingly, tions about preferences and return distri-
topics are emphasized and deleted here butions, is discussed in introductory
primarily according to how closely they graduate texts in finance (e.g., Jonathan
are linked to the martingale model. Ingersoll, Jr. 1987; Chi-fu Huang and
The principal omission that is justified Robert Litzenberger 1988).
on grounds of unrelatedness to the mar- If the origin of the efficient capital mar-
tingale topic is the large and important kets literature is dated in the 1930s, as
literature on rational expectations equi- is reasonable, the martingale model ap-
libria under asymmetric information. In peared on the scene after the chronologi-
the asymmetric information literature cal midpoint of the literature (1965). Up
the focus is on how agents who are to the mid-1960s, market efficiency was

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
1586 Journal of Economic Literature, Vol. XXVII (December 1989)

associated with the random walk model. do not succeed well. Continuing, the
The literature on the random walk model large finance literature on anomalies in
is reviewed in Section II. After this back- asset pricing is reviewed in Section VII.
ground material is presented, the martin- Finally, conclusions are presented in
gale model is presented in Section III. Section VIII.
It is shown there that, as just indicated,
empirical tests of market efficiency are II. The Prehistory of Efficient Capital
in fact tests of a joint hypothesis which Markets
includes the martingale specification.
Further, it is shown that, despite being Early works that were directly related
a descendant of the random walk model, to securities analysis as it is now practiced
the martingale is closely related to the were J. B. Williams' The Theory of In-
fundamentalist model which had earlier vestment Value (1938) and Benjamin
been thought to be diametrically op- Graham and David Dodd's Security
posed to the random walk. In Section Analysis (1934), upon which a generation
IV Eugene Fama's influential analyses of of financial analysts was educated. These
capital market efficiency are discussed. put forth the idea that the "intrinsic" or
Section V begins the presentation of "fundamental" value of any security
empirical developments in the analysis equals the discounted cash flow which
of efficient capital markets over the past that security gives title to, and that-actual
two decades. It was realized around 1975 prices fluctuate around fundamental val-
that the martingale model implied that ues. Accordingly, analysts were in-
asset prices should be less volatile than structed to recommend buying (selling)
they apparently are. An extended de- securities that are priced below (above)
bate, not yet concluded, then began over fundamental value so as to realize trading
whether the observed volatility of asset profits when the disparity is eliminated.
prices in fact exceeds that which capital Because calculating present values is ana-
market efficiency implies, or whether in- lytically trivial-particularly so inasmuch
stead the apparent violations reflect as the theory gave little practical guid-
nothing more than statistical problems ance as to what discount rate to use-
in the (purported) demonstrations of ex- "fundamentalanalysis" consisted in prac-
cess volatility. A closely related litera- tice mostly of forming projections of fu-
ture, that on mean reversion in asset ture cash flow. This involved analyzing
prices, is then reviewed. The discussion demand for the product, possible future
of the latter topic is abbreviated because development of substitutes, the probabil-
the literature on mean reversion, being ity of recession, changes in the regulatory
still at a very early stage in its develop- environment; in short, all information
ment, has not yet arrived at any kind of relevant to future profitability.
consensus. Section VI turns to examina- The only problem with fundamental
tion of alternatives to the martingale analysis was that it appeared not to work.
model. It is easy to show that relaxation Alfred Cowles (1933) demonstrated that
of the strong restrictions on preferences the recommendations of major brokerage
and return distributions required for the houses, presumably based at least partly
martingale model could in principle rec- on fundamental analysis, did not outper-
oncile observed asset price volatility with form the market. The implication was
that implied by market efficiency. How- that investors who paid for these recom-
ever, it turns out that empirically these mendations were wasting their money.
generalizations of the martingale model Other clouds shortly began appearing on

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
LeRoy: Efficient Capital Markets and Martingales 1587

the horizon. In (1934) Holbrook Working paper. Bachelier's study is excerpted in


argued that random walks-cumulated Paul Cootner's (1964) collection of papers
series of probabilistically independent on the random walk model.
shocks-characteristically developed pat- At first the random walk model seemed
terns that look like those commonly as- flatly to contradict not only the received
cribed by market analysts to stock prices. orthodoxy of fundamental analysis, but
Was it possible that stock prices follow also the very idea of rational securities
a random walk? In his (1960) paper, pricing.2 If stock prices were patternless,
Working provided additional evidence in was there any point to fundamental anal-
favor of purely random stock prices by ysis? The random walk model seemed
showing that, if data generated by a ran- to imply that stock prices are exempt
dom walk were averaged over time, spu- from the laws of supply and demand that
rious correlation between successive determine other prices, and instead look
changes would result. Thus existence of more like the casino or musical chairs
such correlations did not necessarily con- game that John Maynard Keynes (1936)
stitute evidence against the random walk chose as metaphors for the stock market.
model. However, economists immediately real-
The "random walk hypothesis"-fore- ized that such a conclusion was prema-
runner of the efficient capital markets ture. Harry Roberts (1959) pointed out
model-was inaugurated in earnest with that in the economist's idealized market
a major statistical study by M. G. Kendall of rational individuals one would expect
(1953) which examined seriously the exactly the instantaneous adjustment of
proposition that stock prices follow a ran- prices to new information that the ran-
dom walk. Kendall found that they do, dom walk model implies. A pattern of
as Working's results had suggested. Clive systematic slow adjustment to new infor-
Granger and Oskar Morgenstern (1963) mation, on the other hand, would imply
followed up Kendall's result with an the existence of readily available and
econometric study using spectral analysis profitable trading opportunities that
that supported the same conclusion. As were not being exploited.
it turned out, however, the results of These considerations raised awkward
Kendall and Granger and Morgenstern questions for proponents of fundamental
had been anticipated in a remarkable analysis: If fundamental analysis worked,
PhD dissertation written in 1900 by why did not new entrants into the busi-
Louis Bachelier, a French mathemati- ness of fundamental analysis, realizing
cian. Bachelier conducted an empirical this fact and planning to participate in
study of French government bonds, find- the trading gains, compete these gains
ing that their prices were consistent with away? That is what happens in every
a random walk model. Besides anticipat- other competitive industry in which prof-
ing the empirical work that was to come its exceed costs-why not in financial
more than a half a century later, Bache- analysis? Alfred Cowles' (1933) results
lier also developed many of the mathe-
2 "Adam Smith" (1968) expressed the skepticism
matical properties of Brownian motion
(the continuous-time analogue of the ran- about the randomwalk model that was characteristic
of market professionals,and also the sense that the
dom walk) which had been thought to randomwalk model is diametricallyopposed to the
have been first derived later in the physi- fundamentalist model: "I suspect that even if the
cal sciences. In particular, Bachelier had random walkers announced a perfect mathematic
proof of randomness, I would go on believing that
anticipated many of the mathematical re- in the long run future earnings influence present
sults developed in Albert Einstein's 1905 value . . . (pp. 157-58).

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
1588 Journal of Economic Literature, Vol. XXVII (December 1989)

suggested that in fact this was exactly incomes based on generating investment
what did happen. Fundamentalists had advice is as much a thorn in the side of
no good answers to these questions. the random walkers as the failure of this
However, the random walk model left advice to generate extranormal trading
as many questions unanswered as it re- returns is a thorn in the side of funda-
solved, and its ablest proponents, such mentalists.
as Roberts, fully realized this. It was em-
barrassing for economists to have to III. Martingale Models
shelve the competitive theory of price-
Resolutions to the puzzles pointed out
surely the jewel in their professional
in the preceding section required situat-
crown-when it came to analyzing stock
ing the random walk model within the
market prices, instead making do with
framework of economic equilibrium.
informal and qualitative remarks such as
Such an account was not forthcoming
if stock prices did not follow a random
within the random walk literature. A
walk there must exist unexploited profit
quarter-century later, it is easy to see
opportunities. If stock prices had nothing
why: By requiring probabilistic indepen-
to do with preferences and technology,
dence between successive price incre-
what about the prices of the machines
ments, the random walk model is simply
that firms use? What about the wheat
too restrictive to be generated within a
the farmer produces and the baker uses,
reasonably broad class of optimizing
but which is also traded on organized ex-
models. However, a weaker restriction
changes just like stock? Where does Mar-
on asset prices that still captures the fla-
shall's Principles stop and the random
vor of the random walk arguments-the
walk start? Plainly there must be more
martingale3 model-turned out to be
to be said.
more tractable. Paul Samuelson's (1965)
There is another problem with the ran-
paper was the first to develop the link
dom walk model. Critics of the random
between capital market efficiency and
walk model can turn the random walkers'
martingales. The simplicity of Samuel-
own method of argument back on them:
son's argument led some (for example,
Huge sums of money are spent every
Mark Rubinstein 1975) to dismiss the re-
year on an activity-securities analysis-
sult as obvious. Perhaps it is, particularly
which, if the random walk model is cor-
with hindsight. However that may be,
rect, is entirely unproductive. Random
when the dust had cleared and the impli-
walkers, the critics observe, expect us
cations of Sameulson's argument were
to believe at once: (1) that unexploited
fully assimilated, the random walk model
patterns in securities prices cannot per-
had been jettisoned and replaced with
sist because for them to do so would im-
the martingale model. Most analysts now
ply that investors are irrationally passing
consider Samuelson's to be the most im-
up profit opportunities, but also (2) that
investors are nonetheless irrationally
3The word martingalerefers in French to a betting
wasting their money year after year em- system designed to make a sure franc. Ironically,
ploying useless securities analysts. If the this meaning is close to that for which the English
argument that no behavior inconsistent language appropriatedthe French word arbitrage.
The French word martingale refers to Martigues, a
with rationality and rational expectations city in Provence. Inhabitantsof Martigueswere re-
can persist in equilibrium is employed puted to favor a betting strategy consisting of dou-
it must be employed consistently, and bling the stakesafter each loss so as-to assurea favor-
able outcome with arbitrarilyhigh probability.
this the random walkers were not doing. I am indebted to Christian Gilles for supplying
Thus the continuing existence of large this background.

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
LeRoy: Efficient Capital Markets and Martingales 1589

portant paper in the efficient capital mar- rates of return are a fair game if and only
kets literature because of its role in ef- if a series closely related to prices-that
fecting this shift from the random walk is, prices plus cumulated dividends, dis-
to the martingale model. The martingale counted back to the present-is a martin-
model does not resolve all the puzzles gale. To prove this, let rt be the rate of
that accompany the random walk, but it return on stock (for example)6from t - 1
does resolve many of them. Unlike the to t, and suppose that rt, less a constant
random walk, the martingale model does p, is a fair game. Using the definition of
constitute a bona fide economic model the rate of return as the sum of dividend
of asset prices, in the sense that it can yield plus capital gain, less one, it follows
be linked with primitive assumptions on from the fair game assumption that stock
preferences and returns which, although price Pt is given by
restrictive, are not so restrictive as to
trivialize the claim to economic justifica- P= (1 + p)-'E(pt+? + dt+lj0t), (3.3)
tion. where d is dividends. Equation (3.3) says
A stochastic process xt is a martingale that the stock price today equals the sum
with respect to a sequence of information of the expected future price and divi-
sets <Dtif xt has the property dends, discounted back to the present
at rate p. When there is no ambiguity
E (xt+I I<Dt)= xt (3.1)
about the information set, as here, it is
and a stochastic process Ytis a fair game convenient to rewrite (3.3) more com-
if it has the property pactly as
E(yt+110t) = 0. (3.2) Pt = (1 + p)-'Et(pt+j + dt+1). (3.4)
Here (3.1) says that if xt is a martingale, None of the variables defined so far is a
the best forecast of xt+1 that could be martingale. The variable that is a martin-
constructed based on current information gale is the discounted value of a mutual
<etwould just equal xt (it is assumed that fund that holds stock the price of which
xt is in iDt).4 This is true for any possible follows (3.4). The mutual fund is assumed
value of the information Itl. Similarly, to reinvest received dividends in further
(3.2) says that if Ytis a fair game the corre- share purchases. To see that the dis-
sponding forecast would be zero for any counted value of this mutual fund follows
possible value of eIt. It is obvious that a martingale, let vt = (1 + p)-tptht be
xt is a martingale if and only if xt1 -xt the value of the mutual fund discounted
is a fair game.5 back to date zero, where ht is the number
The martingale and fair game models of shares of stock the mutual fund holds
are two names for the same characteriza- at t. The assumption that the mutual fund
tion of equilibrium in financial markets; plows back its dividend income implies
that ht+1 satisfies
4The exposition to follow comes with apologies
to Donald McCloskeywho, in instructingwriters of Pt+I ht+1 = (Pt+I + dt+1)ht. (3.5)
economics to avoid "prefabricatedand predictable"
prose- boilerplate--wrote: "Explaininga model of Now consider Et(vt+ ). We have
efficientcapitalmarketsby writingfor the thousandth
time 'Ptgiven It, where It is all the information'does
not advance understanding. If it didn't much help 6 Stock prices will be the principalsource of exam-
to make Eugene Fama's work clear when he first ples throughout this paper. Justificationfor martin-
uttered it, why suppose it will enlighten someone gale models for other sorts of financial prices-for
now?"(McCloskey1987, p. 24). example, futures prices-is sometimes different
S Fair games are for this reason sometimes called (Danthine 1977; LeRoy 1982; Gilles and LeRoy
martingaledifferences. 1986).

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
1590 Journal of Economic Literature, Vol. XXVII (December 1989)

Et(vt+1)= Et[(l + p)-(t+?)pt+1ht+1] fer from this ambiguity. For example, re-
- Et[(l + p)-(t+ )(pt+l + dt+?)ht] jection of the variance-bounds inequality
= (1 + p)-tptht = vt. (3.6) (discussed in Section V) implies rejection
of the martingale model for any specifica-
Here the second equality uses (3.5) and tion of agents' information sets.
the third uses (3.4). Hence vt is a martin- The specification that a stochastic pro-
gale. cess x, follows a random walk (coupled
It is worth emphasizing that (3.4) im- with the add,itional assumption that the
plies that the price itself, without divi- increments have zero mean) is more re-
dends added in, is not generally a martin- strictive than the requirement that xt fol-
gale in the class of models just set out: lows a martingale. The martingale rules
If the dividend-price ratio changes over out any dependence of the conditional
time because of fluctuations in current expectation of xt,1 - xt on the informa-
dividends relative to the variables that tion available at t, whereas the random
predict future dividends, as it generally walk rules out this and also dependence
will, the fair game model implies that involving the higher conditional mo-
the conditionally expected rate of capital ments of xt+ 1 The importance of the dis-
gain must vary in an offsetting manner tinction between the martingale and the
so as to maintain the nonrandomness of
random walk is evident: Securities prices
the conditionally expected rate of return.
are known to go through protracted quiet
Such variation in expected capital gain
periods and equally protracted turbulent
conflicts with the martingale definition
periods. Formally, one might represent
(3.1) (where Pt and Pt+? are substituted
this behavior using a model in which suc-
for xt and xt+1). Nevertheless, the prac-
cessive conditional variances of stock
tice in the efficient capital markets litera-
ture is to speak of stock prices as follow- prices (but not their successive levels)
ing a martingale; in such cases "price" are positively autocorrelated. Such a
should be understood to include rein- specification is consistent with a martin-
vested dividends. We will follow this im- gale, but not with the more restrictive
precise but convenient usage. random walk.
The most direct empirical tests of the Samuelson (1965) proved a result-
martingale model attempt to determine more precisely, pointed out the rele-
whether some variable in agents' infor- vance of a well-known result from proba-
mation set is a predictor of future re- bility theory, the rule of iterated expecta-
turns. If so, the martingale model is vio- tions-which put the theory of efficient
lated. For example, if agents know past capital markets on a firm footing for the
returns and are able to use these to pre- first time. Similar results were presented
dict future returns, returns cannot follow by Benoit Mandelbrot (1966) at about the
a fair game. Of course, this result points same time. Samuelson cast his original
to a fundamental ambiguity in the sim- statement in terms of futures prices,
plest tests of the martingale model: Find- However, continuity of exposition is best
ing some variable that predicts future re- maintained here if his result is restated
turns could mean either that the capital in terms of stock prices; in fact, Samuel-
market is inefficient-that is, does not son (1973) himself provided such a re-
satisfy the martingale property-or that statement. Samuelson's result was that
that variable is not in agents' information the fair game model (3.4) implies that
sets. However, some more sophisticated stock prices equal the expected present
tests of the martingale model do not suf- value of future dividends:

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
LeRoy: Efficient Capital Markets and Martingales 1591
00
the analogy implies, there is no contra-
Pt= E (1 + p)-iEt(dt+i). (3.7) diction even though the focus is different.
i=l
To be sure, in arguing for the similarity
To derive (3.7), replace t by t + 1 in between the fundamentalists' model and
(3.4) and use the resulting equation to the martingale model we have implicitly
substitute out Pt+ in (3.4) as written. redefined the fundamentalist theory of
There results asset valuation in a subtle but critically
Pt = (1 + p)-'Et[(l + p)-' important way. Instead of assuming that
Et+1(pt+2 + dt+2) + dt+?] (.) price fluctuates around fundamental
value (discounted expected cash flow),
If it is assumed that agents never forget Samuelson assumed (or proved, depend-
the past, so that 4t++ is more informative ing on which direction of implication is
than eIt, the rule of iterated expectations being considered) that price actually
guarantees that Et[Et+ I(Pt+2)] equals equals fundamental value. The impor-
Et(pt+2), and similarly for dividends. tance of this change is evident: If price
Therefore (3.8) becomes always equals fundamental value, then
Pt = (1 + p)-'Et(dt+1) no profit can be earned by trading on a
(3. discrepancy between the two, contrary
+ (1 + p)-2Et(pt+2 + dt+2)
to the fundamentalists' assertion. This
Proceeding similarly n times and assum- observation implies that it would be no
ing that (1 + p)-nEt(pt+n) converges to more correct to regard the fundamental-
zero so as to rule out speculative bub- ist model as originally formulated as iden-
bles,7 (3.7) results. Also, the reverse im- tical to the martingale model than it
plication obtains: The expected present- would be to view the two as diametrically
value model (3.7) implies that rates of opposed. Contrary to both of these, it
return are a fair game. is best to regard the martingale model
Samuelson's result implies that the ap- as an extreme version of the fundamen-
pearance noted in Section II of diametric talists' model: If we start with the funda-
opposition between the fundamentalist mentalists' model and modify it by as-
model and the efficient capital markets suming that a large majority of traders
model of asset prices-with the former are conducting fundamental analysis, are
(latter) apparently implying that asset arriving at the same estimates of funda-
prices are completely systematic (un- mental value, and are trading appropri-
systematic)-is entirely illusory. In fact, ately, then price will be bid to equality
Samuelson's result implies that if funda- with fundamental value and trading prof-
mentalists are correct in viewing stock its will disappear.
prices as equal to discounted expected Under what assumptions regarding
cash flows, then it follows that future re- preferences is the martingale model sat-
turns are unpredictable, just as the mar- isfied? Samuelson pointed out that it
tingale model postulates. The fundamen- would be satisfied if agents have common
talists, in focusing on the predictable part and constant time preference, have com-
of asset prices, are asserting that the glass mon probabilities, and are risk-neutral.
is half full, while the martingale model If these conditions are satisfied, investors
contends that the glass is half empty. As will always prefer to hold whichever asset
generates the highest expected return,
completely ignoring differences in risk.
7See Gilles (forthcoming)or Gilles and LeRoy
(1988b) for a statement of conditions under which If all assets are to be held willingly, as
this convergence is guaranteed. must be the case in equilibrium, all must

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
1592 Journal of Economic Literature, Vol. XXVII (December 1989)

therefore earn the same expected rate Fama's paper, like the literature it sur-
of return, equal to the real interest rate. veyed, was devoted almost exclusively
The interest rate, being equal to the con- to empirical work. However, some pre-
stant discount factor, is itself constant liminary theoretical discussion was also
over time. Therefore returns follow the included, and Fama's (1970) definition of
fair game model (3.2), or, equivalently, capital market efficiency became the in-
prices plus reinvested dividends follow dustry standard, reproduced in innumer-
a martingale. able subsequent papers, until it was sup-
Risk neutrality implies the martingale planted by his equally influential (1976a)
(3.1), but not the more restrictive ran- definition. In Fama's (1970) usage, a capi-
dom walk. If agents do not care what tal market is efficient if all the information
the higher moments of their return distri- in some information set 'L is "fully re-
butions are, as risk neutrality implies, flected" in securities prices. Fama, cred-
they will do nothing to bid away serial iting Harry Roberts with the original
dependence in the higher conditional statement, then distinguished three ver-
moments of returns. Therefore risk neu- sions of the efficient markets model de-
trality is consistent with nonzero serial pending on the specification of the
correlation in conditional variances: The information set (D. Capital markets are
fact that future conditional variances are "weak-form efficient" if (' comprises just
partly forecastable is irrelevant because historical prices. Weak-form efficiency
risk neutrality implies that no one cares implies that no trading rule based on his-
about these variances. Following Samu- torical prices alone can succeed on aver-
elson's paper, analysts realized that the age. Capital markets are "semistrong-
theoretical underpinnings for efficient- form efficient" if (D is broadened to in-
markets models in fact point toward the clude all information that is publicly
martingale rather than the random walk. available. Finally, capital markets are
Once aware of the distinction between "strong-form efficient" if (D is broadened
random walks and martingales, they also still further to include even insider infor-
realized that most (but not all; see the mation.
following section) of the empirical tests In light of the discussion in the preced-
for randomness were in fact tests of the ing section of the martingale model, it
weaker martingale model or, for exam- would seem natural to identify market
ple, the still weaker specification that efficiency with the specification that re-
rates of return are uncorrelated. turns follow a fair game, with (1) weak-
form, (2) semistrong-form, and (3) strong-
IV. Fama's Definitions and Evidence form efficiency obtaining depending on
whether the information set includes (1)
The dividing line between the "prehis- past prices and returns alone, (2) all pub-
tory" of efficient capital markets, associ- lic information, or (3) private as well as
ated with the random walk model, and public informnation. An attractive feature
the modern literature is Fama's (1970) of this specification is that, from a mathe-
survey. This influential paper brought matical property of conditional expecta-
the term efficient capital markets into tions, strong-form efficiency implies
general use and is widely interpreted as semistrong-form efficiency, which in turn
associating market efficiency with the implies weak-form efficiency, just as
martingale model, although it will be Fama's choice of terminology suggests.
seen that this interpretation reflects a However, Fama explicitly rejected this
misreading of the paper. specification. Instead, he identified mar-

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
LeRoy: Efficient Capital Markets and Martingales 1593

ket efficiency with the assumption that are based on the assumption that "the
Yt is a fair game: conditions of market equilibrium can
(somehow) be stated in terms of expected
E(yt+I| (Pt) = 0, (4.1) returns . . ., described notationally as
where Yt+I is defined to equal the price follows:
of some security at t + 1 less its condi- [1 + E(rt+It)]pt" (4.3)
tional expectation: II
E(pt+It) = II
(p. 384). Again we have a tautology: (4.3)
Yt+I = Pt+ - E(pt+I I Dt). (4.2) is obtained by applying a conditional ex-
Fama correctly observed that the fair pectations operator to the identity defin-
game model so defined ing the rate of return as equal to the price
relative Pt+ l'Pt (less one). The tautolo-
doesnotnecessarily implythatthe serialcovari- gous nature of Fama's characterization of
ancesof one-period returnsare zero. . . . In
the 'fairgame'efficientmarketsmodel[as de- capital market efficiency was pointed out
finedby (4.1) and,(4.2)], the deviationof the in LeRoy's (1976) comment; in his
returnfort + 1 fromits conditionalexpectation (1976b) reply, however, Fama rejected
is a 'fairgame'variable,but the conditionalex- the argument, explicitly denying the ex-
pectationitself can dependon the returnob- istence of any tautologous elements in
servedfort. (p. 392)
his definition.
Here Fama is explicitly rejecting the In a subsequent section continuing his
identification of capital market efficiency gloss on what it would mean for prices
with the requirement that rates of return to "fully reflect" available information,
themselves be a fair game variable- if Fama proposed the submartingale model
they were, the serial covariances of one-
E(pt+ I I t) ->-pt, (4.4)
period returns would in fact necessarily
equal zero (because past returns are as- so that (neglecting dividends) condition-
sumed to be in agents' information sets ally expected rates of return are nonnega-
under all three forms of market effi- tive. This submartingale characterization
ciency). In Fama's definition, however, of market efficiency is, of course, not tau-
it is only the deviation of price from its tologous. Fama asserted that if stock
conditional expectation that is a fair prices follow a submartingale, then no
game. trading rule based on 'L can outperform
The problem with Fama's characteriza- buy-and-hold. No support was given for
tion of market efficiency is that (4.1) fol- this claim, and it is easy to produce exam-
lows tautologously from the definition ples of economies in which the prices of
(4.2) of Yt+1-just take expectations con- all primitive securities follow submartin-
ditional on 1Dt,on both sides of (4.2). gales but in which there exist trading
Therefore the characterization of Yt+I as rules that outperform buy-and-hold in
defined in (4.2) as a fair game variable terms of expected return. In any case,
does not restrict the stochastic process the capital asset pricing model (CAPM)
for price in any way. On Fama's defini- implies that equilibrium asset returns
tion, any capital market is efficient, and will not necessarily follow a submartin-
no empirical evidence can possibly bear gale: A stock that covaries negatively and
on the question of market efficiency. The sufficiently strongly with the market
passage quoted in the preceding para- might well be priced to yield a negative
graph was not an isolated slip. In his the- expected return. Despite the negative
oretical discussion Fama observed that expected return, risk-averse investors
most empirical tests of market efficiency would be willing to include the stock in

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
1594 Journal of Economic Literature, Vol. XXVII (December 1989)

question in their portfolios because its Generally, the implication of Fama's dis-
negative correlation with the market im- cussion of Niederhoffer and Osborne
plies that it helps to insure the returns seems to be that markets are to be inter-
on the other stocks held, thereby reduc- preted as efficient either if price changes
ing overall risk. are serially independent or if they are
The ambiguity in Fama's theoretical serially dependent but a convincing eco-
discussion of capital market efficiency nomic explanation can be found for the
carried over to his interpretation of the dependence. This is very different from
empirical evidence. Fama generally in- the fair game interpretation of market ef-
terpreted the near-zero autocorrelations ficiency, according to which departures
of successive stock price changes as favor- from the fair game per se are identified
ing market efficiency, suggesting that he with inefficiency.
in fact identified efficiency with the char- Also arguing against Fama's identifica-
acterization of returns as a fair game, con- tion of market efficiency with the martin-
tray to his formal statement. Evidence gale model is the fact that several of the
that mechanical trading rules do not out- studies he interpreted as bearing on mar-
perform buy-and-hold (Sidney Alexander ket effciency use the CAPM to remove
1961, 1964; Fama and Blume 1966) was the risk-premium component of asset re-
similarly interpreted as favoring weak- turns (Michael Jensen 1968, 1969, for ex-
form efficiency, providing further sup- ample). In the CAPM (strictly, in the in-
port for this reading. However, Fama's tertemporal extension of the CAPM
interpretation of Victor Niederhoffer and discussed in Section VII), prices do not
M. F. M. Osborne's (1966) evidence on generally follow a martingale.
runs-successive price changes of the Fama acknowledged the existence of
same sign-is difficult to square with the some evidence against efficiency, partic-
fair game interpretation. Niederhoffer ularly against the implausibly restrictive
and Osborne found that reversals (pairs strong-form version, which requires that
of successive price changes of opposite the information set with respect to which
sign) occurred two to three times as fre- the market is efficient include even in-
quently as continuations. Such system- side information. For example, Nieder-
atic patterns are inconsistent with the fair hoffer and Osborne documented the fact
game model. Despite this, Fama con- that market makers on organized ex-
cluded and emphasized that such pat- changes have no difficulty converting
terns, even though statistically signifi- their monopolistic knowledge of supply
cant, do not imply market inefficiency and demand functions for stock, as em-
(p. 398).8 Fama apparently based this bodied in limit orders, into extranormal
conclusion on the fact that a plausible trading gains. This example is somewhat
explanation for the predominance of re- isolated, however; Fama reported that,
versals over continuations, reflecting the surprisingly, the evidence against even
way limit orders are executed on the or- strong-form efficiency is sparse. Mutual
ganized stock exchanges, can be con- fund managers, who presumably have ac-
structed (see Niederhoffer and Osborne). cess to expert securities analysis, are ap-
parently unable to acquire portfolios that
8 Strictly, Niederhoffer and Osborne's evidence systematically outperform the market
contradicts the more restrictive random walk, not (Jensen 1968, 1969). With regard to semi-
the martingale. However, this distinction does not strong-form efficiency, Fama et al. (1969)
appear to be what Fama had in mind in denying
that Niederhofferand Osborne'sevidence was incon- demonstrated that the information con-
sistent with marketefficiency. tained in stock splits is accurately re-

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
LeRoy: Efficient Capital Markets and Martingales 1595

flected in stock prices at the time of the agents have the same information, in
split, implying that stock splits cannot which case informational efficiency is sat-
be used to construct profitable trading isfied trivially.
rules (unless, of course, one can find out The term efficient capital markets is
about forthcoming splits before they be- seen to have several possible meanings,
come public knowledge). even if one ignores definitions proposed
In sum, Fama (1970) concluded that in the asymmetric information literature
the evidence strongly but not unani- (Sanford Grossman 1978; Grossman and
mously supported market efficiency. Joseph Stiglitz 1976, 1980; James Jordan
Fama proposed a different definition 1983), as we do here. Nonetheless, the
of capital market efficiency in his (1976a) practice in the empirical finance litera-
finance text. A capital market is efficient ture is to speak of tests of market effi-
if (1) it does not neglect any information ciency as if this phrase had unambiguous
relevant to the determination of securi- meaning. For the most part, in the em-
ties prices, and (2) it (acts as if it) has pirical literature market efficiency is in
rational expectations. The assumption of practice equated with rational expecta-
rational expectations means that inves- tions plus the martingale model, and we
tors use their information to make those will follow this convention.
inferences about future events that are
justified by objective correlations be- V. Empirical Evidence: Variance
tween the information variables and the Bounds and Mean Reversion
future events, and only those inferences.
In other words, rational expectations Fama's (1970) survey marked a high
models treat the agents being modeled point for capital market efficiency; most
as knowing the structure of the model of the evidence accumulated in the
and the values of its parameters. Putting nearly 20 years since then has been con-
these ideas together, Fama defined capi- tradictory rather than supportive. In this
tal markets as efficient if the market uses section the discussion will concentrate on
all relevant information to determine se- the variance-bounds violations and the
curities prices, and uses the information literature on mean reversion that grew
correctly. Fama emphasized that effi- out of it. These topics are chosen because
ciency can be tested only jointly with they are directly related to martingales.
some particular model of market equilib- Also, other types of evidence, such as
rium, the nature of which depends on the calendar-based "anomalies"explored
endowments and preferences, but which in the finance literature, have recently
is not implied by market efficiency. Al- been surveyed elsewhere. This other evi-
though his (1976a) definition has a major dence will be acknowledged very briefly
drawback, it is a great improvement over in Section VII.
the (1970) definition. Most important, by Beginning in the mid-1970s, analysts
clearly and unambiguously defining capi- came to realize that the same models
tal market efficiency in a way that is logi- which imply that returns should be un-
cally independent of particular market forecastable also imply that asset prices
models, Fama resolved many of the am- should have volatility which is, in a pre-
biguities in his (1970) treatment of mar- cise sense, low relative to the volatility
ket efficiency. The drawback lies in his of dividends. Results of tests of these vol-
anthropomorphic characterizationof "the atility implications of market efficiency
market": One can speak unambiguously were circulated in 1975 in the paper by
of "the market's" information only if all LeRoy and Richard Porter (published

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
1596 Journal of Economic Literature, Vol. XXVII (December 1989)

1981). Robert Shiller, working indepen- Shiller, I call p* the "ex post rational"
dently, reported the results of tests of stock price. The difference between ex
similar volatility relations in his (1979) post rational and actual price, xt, is seen
and (1981b) papers. In both cases the out- to equal the discounted sum of the unex-
come was the same: Asset prices appear pected component of future returns.
to be more volatile than is consistent with Taking conditional expectations, (5.4)
the efficient-markets model. In setting yields
out the "variance-bounds" theorems (as
LeRoy and Porter called them) here, I Pt = E (pt I4?'), (5.6)
present the version that we developed so that Pt is a forecast of 'p* given agents'
rather than Shiller's version because, as information St. Given (5.6), (5.4) says
will be seen shortly, much of the subse- that p* can be expressed as the sum of
quent original (as opposed to critical) a forecast (Pt) and a forecast error (xt).
work on the variance-bounds theorems Optimal forecasting implies that forecasts
turned out to be more closely related to and forecast errors are uncorrelated. Un-
our paper than to Shiller's. correlatedness in turn implies that
To begin, note that the fair game as-
sumption (3.4) plus the definition of the V(pt*)= V(pt) + V(xt). (5.7)
rate of return imply that Pt can be written
as Because variances-specifically, V(xt)-
are always nonnegative, V(p*) is an up-
Pt = (1 + p)-'(dt+1 + Pt+i) (5.1) per bound for V(pt).
- (1 + p)-'et+, (.1 The implied variance inequality,
where et+1 is the unexpected component V(pt)-- V(Pt*)' (5.8)
of the one-period return on stock:
is attractive because the upper bound de-
+~
p+iP+id (5.2) pends only on the dividends model and
- Et+i-(pt+i + dt+i) the discount factor, but not on agents'
(it is assumed throughout that all varia- information sets. Thus, econometric
bles have finite means and variances). problems aside, rejection of (5.8) unam-
Now replace t by t + i in (5.1) and multi- biguously implies rejection of the martin-
ply both sides by (1 + p)-i: gale model for any specification of agents'
information sets. It will be recalled that,
(1 + p)-ip+i = (1 + p)-(i+l)(dt+i+l in contrast, under conventional tests re-
+ Pt+i+,) - (1 + p)-(i+l)et+i+.1 (5.3) jection could mean either that markets
Summing (5.3) over i from zero to infinity are inefficient or that whatever variable
and assuming convergence, there results allows prediction of future returns is not
in agents' information sets.
Pt = Pt + Xt, (5.4) The variance of the (unobservable)
where forecast error xt turns out to be propor-
tional to the variance of the (observable)
00
= unexpected component of returns, where-
*= (1 + p) idt+ (55)
i=1
the factor of proportionality depends on
the discount factor alone. To prove this,
and;xt ( + p) -et+ j. begin with the second equation of (5.5),
which says that the forecast error xt is a
Here p* is the price of stock that would discounted sum of the unexpected com-
obtain if future realizations of dividends ponents of future returns. Taking vari-
were perfectly forecastable. Following ances and evaluating an infinite sum,

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
LeRoy: Efficient Capital Markets and Martingales 1597

V(x,) is seen to be related to V(e,) accord- jt= E[E(p*lF(t)IHt] (5.11)


ing to
or
V(x,) =V(et) (59) Pt= E(ptHt), (5.12)
2p + p2' (.9
assuming that V(et) is constant.9 This using (5.6). Repeating the reasoning pre-
equation will prove useful later. sented in the derivation of (5.8) from
As LeRoy and Porter observed, equa- (5.6), but substituting At for Pt, Ht for
tion (5.7) may be shown to imply that Dt, Pt for p* , and zt for xt, where zt =
the more information agents have, the Pt- Pt, it follows from the fact that
greater will be the variance of price and V(zt) - 0 that
the lower will be the variance of dis- (5.13)
V(At) _' V(Pt).
counted returns. To see the first implica-
tion, consider some information set Ht Proving the second implication-that
which is less informative than agents' ac- the more information agents have, the
tual information set FD.Define Pt to be lower will be the variance of discounted
the price of stock that would obtain under returns-amounts to showing that re-
the information set Ht: turns are more volatile under information
Ht than under St. This demonstration is
Pt=E(Pt IHt). (5.10) direct. Defining xt as P* -tP and observ-
Here At, like p*, is a fictional stock price ing that (5.7) continues to hold with Pt
series that would obtain if investors had and xt replacing Pt and xt, it follows from
different information than they actually (5.13) that
do. In a sense, t and P* are on opposite
sides of Pt: The former is the price that x V(xt) (5.14)
would prevail if agents had less informa- which, in light of (5.9), implies
tion than they do, while the latter would
prevail if they had perfect information. e ' V(et), (5.15)
Because of this, one would expect that
the relation of Pt to Pt would be qualita- where e is the forecast error for returns
tively similar (in some sense) to the rela- under the information set Ht.
tion of Pt to p*. Specifically, one might This completes the statement of LeRoy
guess that, just as V(p*) is an upper and Porter's theoretical results. It is use-
bound for V(pt), it might also be true ful to summarize what has been proven.
that V(pt) is an upper bound for V(pt), Two basic facts about the martingale
or, equivalently, that V(At) is a lower model are that the variance of stock price
bound for V(pt). This guess turns out to and the variance of returns (multiplied
be correct. The rule of iterated by a constant) add up to the variance of
expectations'" implies that ex post rational price (5.7 and 5.9), and
that the variance of the ex post rational
9 Derivation of (5.9) makes use of the fact that be- price does not depend on how much in-
cause the et are serially uncorrelated, the variance
of the sum of the (1 + p)-iet+i terms equals the formation agents have. These facts imply
sum of the variances(the covarianceterms drop out). that hypothetical variations in agents' in-
Also used is the fact that the variance of a constant formation induce a negative relation be-
times a randomvariableequals the constant squared
times the varianceof the randomvariable. tween the variance of price and the vati-
10Formally, the rule of iterated expectations is ance of returns: That is, the more
used in exactly the same way here as in passing from information agents have, the higher is
(3.8) to (3.9) in the derivation of the present-value
relation. Its use may be easier to understand intu- the variance of price and the lower is
itively there than here. the variance of returns. Thus if agents

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
1598 Journal of Economic Literature, Vol. XXVII (December 1989)

have very little information, stock prices is thus an inequality restriction on param-
are usually not much different from the eters, whereas the null hypothesis of an
discounted sum of unconditional ex- orthogonality test is an equality restric-
pected dividends, a constant. Therefore tion. We constructed both tests using the
stock prices have low volatility. In this estimated parameters of a bivariate auto-
case realizations of actual dividends come regression model for prices and divi-
as near-complete surprises, l inducing dends (i.e., two regressions in which
high volatility in actual returns. How- price and dividends, respectively, were
ever, if agents have a great deal of infor- regressed on their own and the other's
mation about future dividends, stock lagged values). This model, together with
prices have almost as much volatility as the estimated discount factor, implies an
discounted actual dividends, the two be- estimate of the upper bound V(p*). The
ing highly correlated. In this case signifi- bounds test compared the estimate of
cant surprises occur very seldom, imply- V(pt) implied by the bivariate model with
ing that returns will usually be nearly the estimated upper bound. The inequal-
equal to their unconditional expectation. ity (5.8) was reversed empirically, contra-
Given that the volatilities of price and dicting the martingale model. Shiller
returns depend monotonically on how (1981a) reported rejection of a similar in-
much information agents have, it follows equality.
that if we can place bounds on agents' LeRoy and Porter's orthogonality test
information, these will induce bounds on was conducted by constructing an esti-
the variances of price and returns. The mate of each term of (5.7) from the es-
obvious choice for the upper bound on timated bivariate model for price and
agents' information is perfect informa- dividends: Instead of using only the infor-
tion, implying that V(p*) is an upper mation that V(xt) is nonnegative, as in
bound for V(pt) and, unhelpfully, that the bounds test, the orthogonality test
zero is a lower bound for V(et). Given used the fact that V(xt) is related to the
Fama's definition of weak-form effi- variance of one-period returns according
ciency, the obvious choice of a lower to (5.9). The test then consists of evaluat-
bound on agents' information is that ing the null hypothesis
agents know past returns, but nothing
else. It follows that V(At) is a lower bound Ho:V(pt) = V(pt*) ( 2 (5.16)
+t)
for V(pt), and V(et) is an upper bound
for V(et). Of the four variance bounds,
against the alternative
two are interesting empirically: V(p*) as
an upper bound for V(pt), and V(et) as
an upper bound for V(et). H1: V(pt) > V(p*) - 2 +t) 2 (5.17)
LeRoy and Porter reported the results
of two types of tests: bounds tests and Again the martingale model was re-
orthogonality tests. The null hypothesis jected, although the confidence interval
in a bounds test is satisfied if the variance for the null hypothesis turned out to be
of price (or returns) is less than its theo- extremely large.
retical upper bound. An orthogonality A major difference between Shiller's
test, on the other hand, is a test of the and LeRoy and Porter's interpretations
implications for variances of the equality of the variance-bounds violations was
restrictions on parameters implied by the that Shiller saw them as constituting evi-
orthogonality of forecasts and forecast er- dence against efficiency and in favor of
rors. The null hypothesis of a bounds test the existence of elements of irrationality

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
LeRoy: Efficient Capital Markets and Martingales 1599

in securities pricing, whereas LeRoy and Flavin and Kleidon's papers gave pro-
Porter characterized the violations ponents of market efficiency reason to
merely as an anomaly requiring explana- hope that the apparent evidence of excess
tion (LeRoy 1984). At first it appeared volatility was entirely a consequence of
that LeRoy and Porter's reluctance to flawed econometric procedures. How-
draw any but the weakest conclusions ever, the next round of variance-bounds
from the variance-bounds violations was papers made it evident that the variance-
better justified than Shiller's willingness bounds violations were here to stay, and
to base strong conclusions on the finding that Shiller's willingness to draw far-
of excess volatility: Shortly after publica- reaching conclusions based on these vio-
tion of the original studies it became clear lations (and other evidence) may in fact
that at least some of the variance-bounds have been justified. Shiller (1988), re-
tests were subject to severe econometric sponding to Kleidon (1986), contended
problems. Focusing on Shiller's tests, that under realistic parameter values the
Marjorie Flavin (1983) demonstrated that bias which Kleidon had pointed out was
small-sample problems led to bias against insufficient to explain the magnitude of
acceptance of efficiency. She did this by the violations (in turn, Kleidon, 1988a,
showing that the estimated variances of took issue with Shiller's criticism). 13
both p* and p were biased downward, More important, a new round of "second-
with the bias in the former estimate ex- generation" variance-bounds tests, alleg-
ceeding that in the latter. The reason for edly free of the biases that had been
the downward bias in estimating the vari- pointed out in Shiller's original tests, led
ances of p* and p is that the sample to the same conclusion of excess volatil-
means of both p* and p must be esti- ity. These are surveyed by Gilles and
mated, and the usual fixup (reduce de- LeRoy (1988a) and West (1988b). N.
grees of freedom by one) gives an inade- Gregory Mankiw, David Romer, and
quate correction for the induced Matthew Shapiro (1985), following Por-
downward bias in the sample variance ter's suggestion to Flavin (see Footnote
to the extent that the underlying series 11), tested the variance-bounds inequal-
is autocorrelated. Because p* is more ity using second moments around zero
highly autocorrelated than p, the down-
ward bias is greater in estimating the and Marshand Merton (1986). For brief summaries
variance of p* than of p, which is why of these papers see LeRoy (1984) or Kenneth West
the net effect is to bias the test toward (1988b);for a fairlydetailed expositionand evaluation
rejection. " Allan Kleidon (1986a) focused of these papers see Gilles and LeRoy (1988a).
13 Also, Gilles and LeRoy (1988a)showed that the
on the econometric consequences of vio- criticisms leveled at Shiller by Kleidon and Flavin
lation of a stationarity assumption. He do not extend to LeRoy and Porter. Because LeRoy
showed that, if dividends have unit roots, and Porterused a differenttrend correctionthan Shil-
ler did, Kleidon's demonstrationthat Shiller's tests
problems similar to Flavin's could persist are invalid if the underlying data are nonstationary
even in arbitrarily large samples.'2 does not apply to LeRoy and Porter (however, see
LeRoy and William Parke 1988). With regard to
Flavin's criticism, Gilles and LeRoy showed that in
11
Incidentally, Flavin noted a potential remedy addition to Flavin's bias toward rejection of effi-
(suggested to her by Porter; see Flavin 1983, p. 950) ciency, there exists another bias which skews the
for this problem: estimate variances around zero test toward acceptance. It is not known which bias
rather than around the sample mean. It is easy to is stronger. In establishing that LeRoy and Porter's
verify that, under the null hypothesis, the noncentral test has a bias of indeterminate sign, however, we
variances of p* and p obey the same inequality as have said no more than would in any case follow
the variances around their common mean. immediately from the fact that the variance of ex
12 Other papers making similar points as Flavin post rationalprice is a nonlinear function of the un-
and Kleidon in different ways were Kleidon (1986b) derlying parameters.

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
1600 Journal of Economic Literature. Vol. XXVII (December 1989)

rather than around the sample means so Finally, John Campbell and Shiller
as to avoid the bias Flavin had pointed (1988a) reported a variety of tests, includ-
out. The particular form of their test was ing what are effectively variance-bounds
ingenious. Suppose that po?is any "naive" tests, of a class of models that include
forecast of p*-that is, any function of the martingale. Campbell and Shiller's
investors' information, however inaccu- paper shows much greater similarity to
rate as a forecast of p*. Subtracting and LeRoy and Porter's (1981) paper than to
adding Pt, we have the identity Shiller's earlier papers. 15Unlike Shiller's
earlier papers, Campbell and Shiller's is
Pt -Pt(t -Pt) + (Pt-Pt ). (5.18) an orthogonality test rather than a
If Pt is an optimal estimator of p*, the bounds test. Also like LeRoy and Porter,
difference between the two will be un- Campbell and Shiller tested the martin-
correlated with investors' information gale model by constructing a bivariate
variables, and therefore also with Pt - time-series model for stock prices and
Pt?. Accordingly, we have dividends and determining whether the
= restrictions on the coefficients of the
E(t -t) PO)' = E (p*t
E (p* - -pt)2
0(t) ?2 (5.19) model implied by the martingale model
+ E(pt -
pt)2 (.9 are satisfied. Specifically, Campbell and
implying in turn Shiller noted that if current stock price
is used to construct forecasts of divi-
E (p* -pO)2 ? E (pt -pt)2 (5.20) dends, and if these forecast dividends are
and discounted back to the present, the result
should equal current price. This equality
E(p* -PtO)2 ' E(pt -pO)2. (5.21) between constructed and actual price im-
Mankiw, Romer, and Shapiro con- plies testable restrictions on the parame-
structed the sample counterparts of the ters of the bivariate process for dividends
population parameters in (5.20) and and stock price. Campbell and Shiller
(5.21) and checked the associated ine- found that these restrictions are not satis-
qualities empirically. They found that
both were reversed, implying excess vol- mation sets Ht and It obeying It D Ht (for example,
atility of pi. They characterized this exer- West noted in a footnote that if Ht contains no infor-
mation at all, the innovations in p* will be zero identi-
cise as an unbiased test of the variance- cally, implying that the purported upper bound will
bounds inequality, although they proved be zero). For practical purposes this limitation is in-
only that the expectation of the sample consequential, however, because if Ht contains at
least past returns, as is always assumed in efficiency
statistic has the same positive sign as the tests, then innovations in p* do in fact coincide with
corresponding population parameter, not returns, implying that West's upper-bound test on
that it necessarily has the same magni- innovations coincides with LeRoy and Porter's upper-
bound test on returns.
tude. Further discussion of Mankiw, West, like LeRoy and Porter, Shiller, and Mankiw,
Romer, and Shapiro is found in Gilles Romer, and Shapiro, found empirically that the vola-
and LeRoy (1988a).14 tility of returns exceeds its theoretical upper bound,
indicating rejection of the martingale model.
15 However, Campbell and Shiller's tests were su-
14
West's (1988a) variance-bounds test is essentially perior to LeRoy and Porter's for several reasons.
the same as the upper-bound test on return variances Most important, by postulating an underlying log-
(5.15), which LeRoy and Porter derived but did not linear process and then linearizing, they eliminated
conduct. However, there are minor differences. West the need for trend correction, therefore avoiding any
defined the inequality on the variances of innovations error introduced by faulty trend removal (see Gilles
in Pt and 't rather than on the forecast errors as in and LeRoy 1988a for exposition of LeRoy and Porter's
(5.15), which turned out greatly to complicate the trend removal algorithm; LeRoy and Parke 1988
derivation of the bound. West's innovations version, showed that this algorithm induces a downward trend
unlike LeRoy and Porter's, does not hold for all infor- in the supposedly trend-adjusted data).

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
LeRoy: Efficient Capital Markets and Martingales 1601

fied, actual prices having about twice the short intervals (for example, successive
standard deviation of the constructed daily or weekly returns), whereas the
price series. variance-bounds theorems test the or-
The second-generation variance- thogonality of a smooth average of past
bounds tests, like the first-generation returns over a period of years and a simi-
tests, found excess volatility"6 This out- lar smooth average of future returns.
come conflicted with the early work re- The obvious way to evaluate this expla-
viewed by Fama (1970): How can it be nation for the differing results of vari-
that, if stock price volatility is excessive, ance-bounds tests and the conventional
successive daily or weekly stock returns return autocorrelation tests is to estimate
are uncorrelated? This discrepancy posed directly the correlation between average
a major analytical problem. Several returns over the interval from t - T to
explanations resting on sophisticated t-call this rt - T,t-with rt,t + Tfor various
econometric arguments were proposed values of T. Fama and Kenneth French
before it was recognized that there is a (1988a) conducted exactly this exercise.
simple answer. The central point, inade- They found a U-shaped pattern: For T
quately recognized at first, is that the of one year the correlation was essentially
variance-bounds inequalities are implica- zero. For T on the order of three to five
tions of return orthogonality conditions years about 35 percent of the variation
just as conventional efficiency tests are. of rt, t+ T is explained by rt- T, t, with the
To see this, write (5.4) and (5.5) as correlation being negative as expected.
For T of ten years the correlation reverts
Pt = Pt + Xt to approximately zero. Fama and
= Pt+ E (1 + p)->et+i (5.22) French's finding that five-year returns
i=l have a large forecastable component is
so thatthe restrictionon whichthe vari- exactly what the variance-bounds viola-
ance-boundstheoremsare based-or- tions would lead one to expect. The sim-
thogonalityof Ptandxt-says thata par- plicity of Fama and French's test and its
ticular weighted average of past returns outcome provide independent corrobora-
(which is all that Pt is) must be uncorre- tion of the econometric soundness of the
lated with a different weighted average variance-bounds tests.
of future returns. Excess volatility means The question becomes: What sort of
that, empirically, these weighted aver- model would generate the U-shaped pat-
ages of returns are negatively correla- tern in the return autocorrelations that
ted-otherwise V(pt) could not exceed Fama and French reported? Shiller
V(p*). The crucial difference between (1981a, 1984) and Lawrence Summers
conventional efficiency tests and vari- (1986) proposed that instead of modeling
ance-bounds tests is this: The former stock price (with dividends added in) as
tests the orthogonality of returns over a martingale, analysts should consider as-
suming that price comprises a random
16 LeRoy and Parke's(1988)paper is an exception. walk plus a fad variable, where the latter
Our purpose was to construct a bounds test of the is modeled as a slowly mean-reverting
inequality(5.8) that is valid if dividends follow a geo- stationary series. This specification, sim-
metric randomwalk. We found that the varianceof
stock prices is lower than the theoretical upper ple as it is, generates exactly the forecast-
bound, conformingto the variance-boundsinequal- ability pattern required. That returns
ity. However, LeRoy and Parke concluded that this over short intervals are approximately
evidence in favor of the martingale model is ex-
tremely weak. This is so because bounds tests inher- uncorrelated is a basic, but not ade-
ently have lower power than orthogonalitytests. quately known, fact about (a wide class

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
1602 Journal of Economic Literature, Vol. XXVII (December 1989)

of) stochastic processes (Christopher and correlated with returns. It is not yet
Sims 1984). Intuitively, the reason the known whether this qualification to the
return from t - T to t is for small T assertion above that variance-bounds and
approximately uncorrelated with the re- return autocorrelation tests are essen-
turn from t to t + T is that the contribu- tially similar is important empirically.
tion of the drift term to the variation of A third type of test, which can be inter-
price is proportional to T2, whereas the preted as a hybrid of variance-bounds
contribution of the dispersion term is and return autocorrelation tests, deter-
proportional to T (recall from economet- mines directly whether price, or some
rics the analogous fact that mean square variable closely related to price such as
error is the sum of variance plus bias the dividends-price ratio, predicts future
squared). For small T the dispersion returns. These tests usually lead to strong
dominates the drift, implying that the re- rejection of the martingale model (Fama
turn autocorrelations for (almost) any sta- and French 1988b; Campbell and Shiller
tionary stochastic process look like those 1988a, 1988b).
of a fair game (zero). Similarly, return The variance-bounds, return autocor-
autocorrelations over long horizons ap- relation, and price-return orthogonality
proach zero because the random walk tests constitute three ways to test the
term dominates the mean-reverting com- martingale model. A fourth way to test
ponent of price. In between, however, for mean reversion is to use variance ra-
a negative correlation is to be expected. tios (John Cochrane 1988). "7The variance
This occurs because for intermediate ratio is defined as the variance of k-period
values of T high returns from t - T to t returns divided by the variance of one-
imply a positive value (on average) for period returns, and also by k. Under a
the fad variable at t. Mean reversion im- random walk the variance ratio should
plies that the fad will probably have di- equal unity for any value of k. However,
minished by t + T, implying an abnor- James Poterba and Summers (1988)
mally low return from t to t + T. The showed that the variance ratios declined
extent of the induced negative correla- with k, indicating the presence of a
tion between rt_T, and rtt+T depends mean-reverting component.
on how quickly fads die out and on the The presence of a mean-reverting com-
respective error variances. ponent in stock prices implies substan-
The preceding discussion exaggerated tial forecastability of intermediate-term
the similarity between the variance- returns, and therefore also (by the
bounds and return autocorrelation tests. variance-bounds theorem) substantial
The problem lies with the assertion fol- differences between price and "funda-
lowing equation (5.22) that current price mentals," meaning by the latter the (ra-
is a weighted average of past returns, so tional) expectation of ex post rational
that variance-bounds tests (which are price. Thus there is no inconsistency be-
based on the orthogonality of price and tween essentially unforecastable short-
future returns) and return autocorrela- term returns and wide discrepancies
tion tests (which are based on the orthog-
onality of past and future returns) are es- 7The material under discussion was anticipated
by HolbrookWorking. In his (1949)paper, Working
sentially equivalent. In fact, price is a proposed that statistical series be modeled as the
nonlinear function of past returns; even sum of a random walk and a stationaryseries, and
if the function relating current price to explicitlyproposedthe use of varianceratiosto deter-
mine the relative importanceof each component.
past returns is linearized, the weights de- I am indebted to FrankDiebold for this reference.
pend on dividends, which are random See also Diebold (1988).

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
LeRoy: Efficient Capital Markets and Martingales 1603

between price and fundamental value. tocorrelation. This finding, together with
This result is best seen as document- the fact that the most recent studies con-
ing a pronounced bias in our psycho- tinue to conclude that the variance-
logical metric-even though there is no bounds inequalities are violated empiri-
question that complete unpredictability cally, raises further questions about
of short-term returns implies exact whether the variance-bounds violations
equality between price and fundamental are empirically the same thing as mean
value (speculative bubbles aside), the re- reversion. At this writing these questions
sult here is that a "surprisingly small" remain unresolved.
degree of forecastability of short-term re-
turns is consistent with a "surprisingly VI. Nonmartingale Models
large" discrepancy between price and
fundamental value (Shiller 1984; Sum- Documenting the existence of system-
mers 1986). atic empirical departures from the mar-
Shiller's suggestion that asset prices be tingale model may seem to be entirely
modeled as the sum of a random walk beside the point. After all, Samuelson's
and a mean-reverting process is seen to derivation of the martingale model as-
give a parsimonious model that predicts sumed risk neutrality, whereas in fact
(1) near-zero autocorrelations for daily people are risk-averse. So why should
and weekly returns as reported in the one be surprised when the martingale
early efficient markets literature, (2) neg- model does not work empirically? Aware
ative autocorrelations for returns over of this point, analysts were led to look
holding periods of several years, and (3) for an analogue to the martingale model
variance-bounds violations. Unfortu- that would remain valid if agents were
nately for this tidy story, however, sev- risk-averse. It has not proved difficult to
eral recent studies have raised questions formulate such extensions theoretically,
about the validity of the purported facts but, as will be reported in this section,
for which the mean-reversion model it has turned out to be very difficult to
gives a unified explanation. Andrew Lo correlate the departures from the martin-
and A. Craig MacKinlay(1988) found that gale that these theories lead one to ex-
weekly and monthly stock returns had pect with the departures that one sees
positive autocorrelation coefficients on in the data. Therefore allowing for risk
the order of 30 percent, contradicting aversion does not in practice go far to-
both the finding of approximately zero ward resolving the empirical puzzles that
autocorrelation reported in the early effi- attend the martingale model. Conse-
cient markets literature and the predic- quently, not much is lost empirically by
tion of approximately zero autocorrela- ignoring risk aversion, which is why that
tion from the mean-reversion model. was done in the preceding sections.
Moreover, several studies have ques- Samuelson was not aware that his der-
tioned Fama and French's conclusion ivation of the martingale model de-
that returns are significantly negatively pended critically on the assumption of
autocorrelated over three- to five-year risk neutrality: He conjectured that risk
holding periods. Myung Jig Kim, Charles aversion could be handled simply by in-
Nelson, and Richard Startz (1988), for ex- cluding a risk premium in the discount
ample, found evidence of mean reversion factor used to calculate present values.
only in data sets that include the 1930s- However, it is easy to see why asset re-
for the post-World War II period they turns will not generally be a fair game
found no evidence of negative return au- if agents are risk-averse. Suppose that

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
1604 Journal of Economic Literature, Vol. XXVII (December 1989)

risk covaries positively over time, so that general equilibrium determination of re-
big price changes (positive or negative) turns on multiperiod assets such as stock.
are likely to be followed by big changes In multiperiod models it makes little
and small changes by small changes.18 sense to determine current risk premia
If agents are risk-averse, they will hold endogenously while taking future risk
risky assets only if expected returns vary premia, as embodied in expected next-
so as to compensate them for these period price, as exogenous.
changes in risk. One would expect that What was needed was a model that
returns therefore will in general be partly would generate price from the probabil-
forecastable: If the current realization of ity distribution of next-period returns,
(t implies high risk over the near future, and that would simultaneously character-
should it not also imply high expected ize agents' probability distribution of
return? next-period returns in a manner that is
To formalize this reasoning, one would consistent with agents' expectations that
like to have in hand a model that allows price will be determined in a similar fash-
for risk-averse agents and that can gener- ion when the next period arrives. This
ate an intertemporal sequence of equilib- required a new concept of equilibrium.
rium prices and returns. The problem In my (1971) dissertation and (1973) pa-
in incorporating risk aversion into effi- per, equilibrium was defined to consist
cient-markets theory was that as of about of a single function simultaneously map-
20 years ago the only equilibrium asset ping current dividends into current price
pricing model extant in which risk and and next-period dividends into next-pe-
risk aversion were adequately handled riod price such that if agents have rational
was the equilibrium version of the CAPM expectations about future dividends and
of William Sharpe (1964), John Lintner optimize, then markets clear for any level
(1965), and Jan Mossin (1966). (General of dividends. 19The solution method then
analytical frameworks like the Arrow- was to specify a general class of price
Debreu contingent claims setup are, of functions and derive the appropriate
course, capable in principle of dealing equilibrium condition under the assump-
with risk aversion, but unless suitably re- tion that both current and next-period
stricted, are too general to be of much price conform to this function. The equi-
use in applied work.) The CAPM takes librium price function was that for which
the mean and variance of next-period this equilibrium condition is satisfied as
price as exogenous and determines cur- an identity in dividends (it would con-
rent asset prices as those prices that just tradict the exogeneity of dividends if
induce agents to bear existing risk will- markets failed to clear for some values
ingly. Price, in other words, equals dis-
counted expected return less a correction 19 Merton (1973) reported an intertemporal exten-
that reflects risk and risk aversion. Now, sion of the CAPM at the same time. Neither LeRoy
the fact that next-period expected price (1973) nor Merton (1973) provided a full general equi-
librium analysis. For example, in both models the
and the variance of next-period return risk-free interest rate is taken as exogenous rather
are given exogenously in the CAPM than determined from preferences and technology.
means that even though the CAPM de- However, the former paper pointed more clearly in
the direction of general equilibrium by contributing
termines the current risk premium en- the idea that equilibrium can be characterized as a
dogenously, it does not give a complete stable function linking exogenous state variables to
asset prices, an idea not found in the latter paper.
General equilibrium versions of LeRoy's and Mer-
18
Empirical evidence supports this specification ton's models were provided by Lucas (1978) and Cox,
(for example, Poterba and Summers 1986). Ingersoll, and Stephen Ross (1985), respectively.

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
LeRoy: Efficient Capital Markets and Martingales 1605

of dividends, hence the need for the per, James Ohlson showed that if divi-
equilibrium condition to hold as an dend growth rates are serially indepen-
identity). dent and agents have constant (but not
As it turned out, the identical concept, necessarily zero) relative risk aversion,
which came to be called rational expecta- then the conditional expected rate of re-
tions equilibrium, was being developed turn on stock will be constant and returns
at the same time in the macroeconomics will be unforecastable. In a sense Ohl-
literature. Further, exactly the same son's case was very specialized because
"undetermined coefficients" solution if agents are risk-averse the martingale
method-seeking coefficients such that requires restrictions both on return dis-
the equilibrium condition holds as an tributions and risk aversion, rather than
identity-came into use in linear rational just the latter as in the risk-neutrality
expectations macroeconomic models case. However, neither of Ohlson's as-
(Robert Lucas 1973). sumptions is as wildly at odds with reality
In the intertemporal version of CAPM as the assumption of risk neutrality. The
just described, the conditional expected practical implication of Ohlson's result is
return per dollar fluctuates over time as that even though the conditions under
dividends change. Because dividends are which he derived an exact martingale are
autocorrelated, conditional expected re- restrictive, the assumption that these
turns are autocorrelated as well, imply- conditions are satisfied to a tolerable ap-
ing that actual returns are partly forecast- proximation may not be so implausible.
able. This forecastability goes contrary The foregoing discussion has con-
to the martingale model. It is, however, cerned asset prices that are or are not
consistent with equilibrium because martingales with respect to the probabili-
equilibrium stock prices are such that the ties that agents actually have- mDre pre-
fluctuations in risk per dollar invested cisely, with respect to the probabilities
induced by dividends fluctuations corre- that, under the axioms of choice under
late with the fluctuations in expected re- uncertainty, are implicit in agents' order-
turns so as to leave agents just willing ings over portfolios. Suppose, however,
to hold existing assets. In other words, that we start from the other end by as-
even though the existence of serial de- suming that asset prices always follow
pendence in conditional expected re- martingales with respect to some proba-
turns implies that different formulas for bilities. It is easy to show that there al-
trading bonds and stock will generate dif- ways exist such probabilities: They are
ferent expected returns, because of risk, readily derived by repackaging the Ar-
these alternative trading rules are utility- row-Debreu prices that underlie any
decreasing relative to the optimal buy- equilibrium (Stephen Ross, 1977, was
and-hold strategy. Of course, if as a spe- the first clearly to appreciate this point;
cial case it is assumed that agents are see also Ross 1978; J. M. Harrison and
risk-neutral, these effects disappear and D. M. Kreps 1979; Harrison and S. R.
the martingale model obtains. Pliska 1981).20 These probabilities are
These considerations made clear that,
20 As a sidelight, it is interesting to note that in
in general, risk aversion will lead to de-
the finance literature Ross' (1977) paper is almost
partures from the martingale model. It universally incorrectly referred to as having been
does not follow from this that risk neu- published in 1976. This practicewas startedby Ross,
trality is the only case in which condition- who deliberately misdated references to this paper
in order to encourage readers interested in the arbi-
ally expected returns will be constant. trage pricing theory to read it before taking on his
In his (1977) comment on my (1973) pa- more difficult,less intuitive, and more rigorous(1976)

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
1606 Journal of Economic Literature, Vol. XXVII (December 1989)

called risk-neutral probabilities in the fi- respect to some fictional probability mea-
nance literature because asset prices can sure that has no directly observable
always be expressed as discounted ex- counterpart.
pected returns-as would be appropriate Both my (1973) paper and Ohlson's
if agents were risk-neutral-if the expec- (1977) comment were essentially coun-
tation is taken with respect to these prob- terexamples, the former to the proposi-
abilities rather than the probabilities im- tion that capital market efficiency neces-
plicit in agents' orderings of portfolios. sarily implies martingales, and the latter
In other words, asset prices can be ana- to the proposition that risk neutrality is
lyzed as if agents are risk-neutral, but required for martingales. As such, there
take expectations with respect to the risk- is nothing wrong with the fact that they
neutral probabilities rather than their ac- are highly specialized. For general analy-
tual probabilities. The risk-neutral prob- sis, however, more powerful methods are
abilities coincide with actual subjective needed so as to derive equilibria in more
probabilities if agents are in fact risk-neu- general settings. These were supplied in
tral-otherwise they contain in addition Lucas' (1978) paper. (Related material,
adjustments for risk aversion. developed independently, was presented
The fact that there always exist martin- in Douglas Breeden, 1979, and in John
gale representations of asset prices is Cox, Jonathan Ingersoll, and Stephen
very convenient for theoretical work. It Ross' 1985 paper, which was circulating
is also useful in such applied work as the as a working paper in the mid-1970s.)
pricing of redundant assets, the central Lucas assumed that identical infinitely
problem of applied finance.2' For the lived agents maximize the utility function
study of capital market efficiency, how- J(1 + p)-TU(ct+T), which allows for risk
ever, this line of research is not directly aversion (U strictly concave) as well as
relevant. Given that market efficiency risk neutrality (U linear). Using dynamic
includes rational expectations (Fama programming, Lucas demonstrated the
1976a), the subjective probabilities im- existence and uniqueness of a pricing
plied by agents' orderings over portfolios function similar to that of my (1973) pa-
must be identifiable with the objective per. Even though the equilibrium pric-
probabilities specified to obtain in the ing function is nonlinear in Lucas' model
model under discussion. In the present and is usually not amenable to closed-
setting it is therefore of no help to know form representation, many of its proper-
that returns are always fair games with ties can be derived analytically.
In Lucas' model equilibrium prices sat-
treatmentof the arbitragepricing theory. The (1977) isfy the stochastic Euler equation
paper was actuallywritten in 1971. Most subsequent
writers on the arbitragepricing theory followed Ross' ptUt = (1 + p)-'Et(pt+j + dt+1)U+1 (6.1)
lead in dating the (1977)paper as 1976. An exception
is Ingersoll (1987), who, going Ross one better, re- Here the marginal utilities Ut and Ut+1
ferred to the publication date of Ross' (1977) paper
as 1975. are evaluated at the endowment, reflect-
21 For example, Cox, Ross, and Rubinstein (1979) ing the equilibrium condition that con-
presented an intuitive derivationof the Black-Scholes sumption must equal the endowment in
model of option pricing using martingalerepresenta-
tions. Specifically,they derived risk-neutralprobabil- an exchange economy. To understand
ities from the assumed price of stock and the interest the Euler equation (6.1), suppose that
rate, and then calculated the price of an option on an investor is considering selling one
the stock by discountingits expected return, where
the expectationwas calculatedusing the risk-neutral share of stock and consuming the pro-
probabilities. ceeds. The utility gain is ptU'. Assuming

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
LeRoy: Efficient Capital Markets and Martingales 1607

that consumption at dates other than t model there is a general presumption


and t+ 1 remains unchanged, the budget that the more risk-averse agents are, the
constraint implies a drop in consumption more volatile asset prices will be. The
at t+1 of Pt+i + d,+1. The right-hand argument is very simple. In an economy
side of (6.1) gives the expected utility with no production, agents must con-
cost of the decline in consumption, dis- sume their randomly fluctuating endow-
counted back to t. If the investor is at ment (taking account of capital and inter-
an optimum, the utility gain at t must temporal production would complicate
just equal the expected utility loss at t+ 1. the story, but would not alter its funda-
Equation (6.1) agrees with the martin- mentals). The price system must induce
gale model them to do so willingly. Highly risk-
averse agents, however, will want very
Pt = (1+p)-'Et(pt+1+dt+1) (3.4) much to smooth their consumption
except that in (6.1) price at t is weighted streams over time. This they cannot do
by current marginal utility and next-pe- in the aggregate. To induce them not to
riod price by next-period marginal util- save (by buying stock) in periods of pros-
ity. Under risk neutrality U' and Ut+1 perity, and not to dissave (by selling
are equal to a common constant, so (6.1) stock) in periods of shortage, stock prices
and (3.4) agree. Lucas therefore again must be very high in periods of prosper-
pointed out that martingales generally ity and very low in periods of shortage.
would obtain only under risk neutrality. Thus the more risk-averse agents are, the
Also, Lucas' work made clear that the more volatile equilibrium stock prices
connection between risk neutrality and will be. However, this argument is not
martingales obtains without qualification completely general. As Ohlson (1977)
only in exchange economies. In produc- showed, if dividend growth rates are in-
tion economies in which corner solutions dependently distributed, then prices will
are possible, prices will reflect the tech- follow a martingale for any degree of
nology as well as preferences whenever (constant relative) risk aversion.22 In such
corner solutions occur, so risk neutrality settings risk aversion cannot be the expla-
by itself is insufficient to generate the nation for asset price volatility in excess
martingale. This qualification was not of that implied by the martingale model.
stated in Samuelson's paper or mine. In See Kevin Salyer (1988) for a general dis-
production economies like that of Wil- cussion of price volatility in models like
liam A. Brock (1982) in which the tech- Ohlson's.
nology excludes corner solutions, on the These theoretical developments raised
other hand, risk neutrality is sufficient the possibility that the variance-bounds
for the martingale model without qualifi- violations (or, equivalently, the partial
cation. forecastability of intermediate-term re-
An immediate payoff of Lucas' model turns) reflected departures from the mar-
was that it provided an analytical frame-
work in which to determine whether the 22
Ohlson's model is not a special case of Lucas'
violations of the variance-bounds theo- because dividend levels are nonstationaryin the for-
mer. However, Mehra and Prescott (1985) formu-
rems reflect the unrealism of the under- lated a general frameworkanalogousto Lucas'except
lying risk-neutrality assumption. It was that dividend growth rates rather than levels are sta-
shown by LeRoy and C. J. LaCivita tionary. Ohlson's model is a special case of Mehra
and Prescott's.In Mehra and Prescott'ssetting there
(1981), Grossman and Shiller (1981), and is no simple connection between risk aversion and
Ronald Michener (1982) that in Lucas' asset price volatility (Salyer 1988).

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
1608 Journal of Economic Literature, Vol. XXVII (December 1989)

tingale model induced by risk-aversion. plain the equity premium cannot be eas-
Grossman and Shiller (1981) and Lars ily dismissed.23
Hansen and Kenneth Singleton (1982, There remains the fads model pro-
1983), among others, attempted to deter- posed by Shiller (1981a, 1984). Here, of
mine whether asset price fluctuations course, we are dealing with an alternative
could be interpreted as reflecting risk- to the efficient capital markets model,
averse agents' attempts to smooth con- not with a modification of it. Most econo-
sumption over time. Results to date have mists are extremely reluctant to resort
been disappointing (see Singleton 1987 to fads models because doing so would
for a survey of this literature). The prob- involve relaxing the stable-preferences
lem is that consumption-based models of assumption that many economists regard
asset pricing, at least in their simplest as an indispensable part of their outlook
form, imply that stock returns will be (George Stigler and Gary Becker 1977).
positively and strongly correlated with In any case, pending a theory of what
consumption growth, and this turns out causes fads to come and go or a specifica-
not to be true empirically. Therefore, in- tion of potential phenomena that would
troducing risk aversion does not gener- be inconsistent with a fads model, it is
ally improve the performance of the pre- not clear that anything is gained by char-
dicted price series much in tracking acterizing an unexplained variation in as-
actual prices relative to the martingale set prices as a fad. One is reminded of
model. However, this pessimistic evalua- Robert Solow's (1957) labeling as techno-
tion is not universally shared: Kleidon logical change the unexplained residual
(1988b), for example, expressed doubt in output growth after allowing for in-
that the variance-bounds violations re- crease in inputs: Precisely because the
flect anything deeper than an unjustified residual is unobserved, one is free to ac-
assumption of a constant rate of time dis- cept or reject the interpretation; nothing
count (and perhaps, given the economet- is at stake either way. Advocacy of a fads
ric problems, not even that). Also, more model is perhaps best interpreted as a
sophisticated representations of risk statement of belief that the most fruitful
aversion (for example, George Constan- avenues of future research will involve
tinides' 1988 non-time separable utilities) social or cognitive psychology, rather
may improve the results. than as referring to any well-formed
In fact, rather than resolving the diffi- model that is now available.
culties attending the martingale model,
passing to the consumption-based asset VII. Other Evidence
pricing model has given rise to new prob-
The discussion of empirical evidence
lems. Rajnish Mehra and Edward Pres-
in the preceding two sections was nar-
cott (1985), studying a representative-
rowly concentrated on the time structure
agent model, showed that no reasonable
of asset returns and such closely related
specification of agents' rates of time pref-
topics as variance bounds. This restricted
erence and risk aversion was able to gen-
focus was adopted to avoid spreading the
erate real returns on bonds as low as
discussion too thin. But there is no point
those measured, while at the same time
in basing conclusions on only a small sub-
generating real returns on stock as high
as those measured. It is true that Mehra
and Prescott's model is highly simplified, 23 Proposed resolutions to Mehra and Prescott's
but the dramatic failure of the consum- equity premiumpuzzle have been suggested by Rietz
tion-based model of asset prices to ex- (1988), Constantinides(1988), and Nason (1988).

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
LeRoy: Efficient Capital Markets and Martingales 1609

set of the available evidence. This section advisory services. If n is large, one would
briefly acknowledges the existence of expect the best of n services to perform
other types of evidence that bear on the extremely well purely by chance. And
question of capital market efficiency. For surely the population of investment ser-
more extensive reviews, see the papers vices is large, especially if, as is appropri-
in Elroy Dimson (1988) (especially Don- ate, one counts the services that drop
ald Keim 1988), G. William Schwert out because of a poor track record.24
(1983), Richard Thaler (1987a, 1987b, The advent of cheap computing and
1988), Josef Lakonishok and Seymour large financial data bases brought new
Smidt (1988), and Ross Clarkand William anomalies. The consensus now is that the
Ziemba (1987). anomalies pose a serious problem which
There always existed a subculture cannot be shrugged off, as had been pre-
within the finance profession that re- sumed earlier. The best known of these
jected the majority conclusion in favor is the "Januaryeffect" (see Thaler 1987a
of efficiency. These heretics pointed to or Clark and Ziemba 1987 for surveys).
the "P-E anomaly":stock with low price- Michael Rozeff and William Kinney
earnings ratios appeared systematically (1976) found that stock returns averaged
to outperform those with high price- 3.5 percent in- January, while other
earnings ratios (Francis Nicholson 1968; months averaged 0.5 percent, a pattern
Sanjoy Basu 1977, 1983; Marc Reinga- which, being nonstationary, is inconsis-
num 1981; David Dreman 1982). Re- tent with a martingale. Subsequent stud-
cently Werner DeBondt and Thaler ies (for example, Reinganum 1981, 1982,
(1985, 1987) documented the related 1983, and Richard Roll 1983) replicated
proposition that "losers"-stocks that had and refined the Januaryeffect. Rolf Banz
recently undergone large drops-appear (1981) found that small firms have higher
systematically to generate higher returns returns than is consistent with their riski-
than winners. Another similar result is ness. Keim (1983) showed that the small-
that the ratio of price to book value is a firm effect and the January effect may
predictor of returns (Barr Rosenberg, be the same thing: The January effect
Kenneth Reid, and Ronald Lanstein appears only in samples that include and
1985). This evidence of systematic over- give equal weight to small and large firms
reaction to current information may be (see also Lakonishok and Smidt 1988 and
related to the excess volatility docu-
mented in the variance-bounds litera- 24 One is reminded of the story about an entrepre-
ture. Also, the apparent success of some neur who wanted to sell recommendationsto football
investors-Warren Buffett-and some bettors. He divided a list of 16,000 potential custom-
investment services-Value Line-in ers into two sublists of 8,000 names each. He in-
formed the first sublist of his prediction that the
outperforming the market is difficult to Redskinswould beat the 49ers on Sunday, while the
reconcile with capital market efficiency. second sublist was given the reverse prediction.
Proponents of market efficiency have al- When the Redskins did beat the 49ers, he threw
out the second list. The next week he divided the
ways minimized such evidence. It is true first list into two new sublists of 4,000 names each.
that the correspondence of the Value He reminded both that he had correctly predicted
Line stock rankingswith subsequent per- the outcome of last week's game. For the first sublist
he picked the Giants over the Eagles; the second
formance appears too strong to have oc- sublist received the reverse prediction. After four
curred by chance if Value Line is thought weeks he was left with 1,000 names. He then wrote
of as a single prespecified observation. to these reminding them that he had correctlycalled
the past four games, and expressed a willingness to
But suppose that Value Line is thought tell them the outcome of the next game in exchange
of as the best performing of n investment for $10,000.

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
1610 Journal of Economic Literature, Vol. XXVII (December 1989)

Mustafa Gultekin and N. Bulent Gulte- much news about fundamentals is gener-
kin 1987), as opposed to samples that ated on Wednesdays as other weekdays,
weight firms by value. this "Wednesday effect" suggests that it
Not only is the Januaryeffect an anom- is the trading process itself rather than
aly in its own right, but it contaminates news about fundamentals that generates
the one regularity that finance theory price changes.25 The Wednesday effect,
(specifically the CAPM) predicts should like the Januaryeffect and assorted other
be found in the data: the 'relation be- calendar effects, appears difficult to rec-
tween risk and expected return. Fama oncile with the martingale model. Fi-
and James MacBeth (1973) and others nally, Robert Ariel (1987) showed that
had earlier confirmed the CAPM (or returns are positive on average only in
commonsense) prediction that riskier the first half of the calendar month.
stocks should earn higher average re- It is difficult to know how seriously to
turns. Seha Tinic and Richard West take these asset pricing anomalies. As
(1984) were motivated by the findings Robert Merton (1987) and many others
just summarized to analyze the monthly have noted, there is a problem of selec-
patterns in the risk-return relation. In- tion bias in these results. An analyst who
credibly, they found that the risk-return conducts an empirical study investigating
trade-off occurs entirely in January:They a purported correlation between stock
could not reject the hypothesis that dur- returns and the stage of the moon, for
ing the other eleven months investors example, and finds no correlation is un-
are not compensated at all for bearing likely to succeed in reporting this result
risk (however, see also Tinic and West in the journals. Therefore the published
1986). literature is skewed toward interesting,
The Januaryeffect is only one of several that is, anomalous, results, and away
calendar-based anomalies that have been from boring confirmations of the absence
unearthed in recent years. Another is the of anomaly. A related problem is that
"weekend effect" (Frank Cross 1973; anomalies are typically tested on the
French 1980; Keim and Robert Stam- same data on which they are discovered,
baugh 1984; Lakonishok and Maurice and analysts frequently construct their
Levi 1982; R. Rogalski 1984; Jeffrey Jaffe classifications so as to maximize the
and Randolph Westerfield 1985; Law- anomalous nature of the finding. For ex-
rence Harris 1986), which finds that stock ample, Ariel (1987) included the last day
returns are on average negative from the of the preceding month along with the
close of trading on Fridays to the opening first half of the current month because
of trading on Mondays. A similar effect returns on the last day of the month are
exists for bonds (Michael Gibbons and very high, implying an increased re-
Patrick Hess 1981). Further, we have the ported disparity between returns in the
"Wednesday effect": In 1968 the New first half of the month and returns in the
YorkStock Exchange was closed on Wed- second half (see Lakonishok and Smidt
nesdays in order to ease the paperwork 1988 for discussion).
backlog at brokerage houses. French and Different types of evidence bear more
Roll (1986) found that the volatility of directly on the assumptions of rationality
prices from Tuesday to Thursday was and rational expectations that underlie
lower than over other two-day intervals,
25 However, see Slezak (1988) for an alternative
suggesting that prices fluctuate more
explanationfor the Wednesdayeffect which is consis-
when markets are open than when they tent with (a sophisticated version of) the efficient
are closed. Because, presumably, as marketsmodel.

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
LeRou: Efficient Cavital Markets and Martingales 1611

market efficiency (and, consequently, are one of the major anomalies of efficient
less closely related to martingales). For capital markets. Paul Milgrom and Nancy
example, there is some evidence that as- Stokey's (1982) paper and Jean Tirole's
set prices are subject to "winner's curse" (1982) paper (see also Harrison and Kreps
(Edward Miller 1977; Stuart Theil 1988; 1978) showed that rational agents with
Kenneth Hendricks and Robert Porter asymmetric information will not offer to
1988; S. Michael Giliberto and Nikhil Va- trade securities based on a naive inter-
raiya 1989). If agents have different opin- pretation of their private information.
ions about the value of some asset to be Rather, they will take account of the fact
sold at auction, and if their bids are that if they are- able to consummate a
naively based on these opinions, the win- trade, that will occur because some other
ner will be the bidder with the most in- agent with different but perhaps equally
flated estimate of the asset's value. On accurate information is willing to take the
average, winners will overpay. Winner's other side of the trade. Such transactions,
curse is inconsistent with full rationality: being a zero-sum (or negative-sum, if
Each bidder's strategy should make al- brokerage charges and costs of informa-
lowance for the possibly biased nature tion acquisition are included) game, are
of his own appraisal of value (R. Preston pure risk uncompensated by positive ex-
McAfee and John McMillan 1987). Rich- pected gain. Risk-averse agents will re-
ard Thaler (1988) interpreted the finding ject such trades. Contrary to the predic-
of Walter Mead, Asbjorn Moseidjord, tion of Milgrom, Stokey, and Tirole's
and Philip Sorensen (1983, 1984) that model, large numbers of investors for-
winning bidders on wildcat offshore oil sake the buy-and-hold strategy that effi-
leases overpay on average as evidence cient-markets theory dictates in favor of
of winner's curse. 26 actively betting their information against
A very striking piece of evidence con- other investors' information. Of course,
flicting with market efficiency is the high it is not the fact that the volume of trade
volume of trade on organized securities is positive that causes the problem: Mil-
exchanges. For some reason this is sel- grom, Stokey, and Tirole's theorem de-
dom listed in the finance literature as pends on assumptions that are not even
approximately satisfied empirically-for
26
However, Thalerdid not note that these authors example, that agents have common pri-
suggested an explanation different from winner's ors (see Hal Varian 1985, 1989 for analy-
curse for the low returns to successful bidders on ses of models in which agents have heter-
wildcat leases. The successful bidder on a wildcat
lease-a lease for which there exists no drilling data ogeneous priors) and that the pretrade
that would indicate potential productivity-acquires allocation of securities is Pareto-optimal.
valuable proprietaryinformationabout oil reserves Given market incompleteness, rational
in neighboringtracts. When drainageleases-leases
on tracts adjoining tracts from which oil is already investors will want to buy or sell securi-
being extracted-on these neighboring tracts come ties to provide for or finance large expen-
up for auction, the holder of the wildcat lease can ditures or adjust risk exposure. How-
modifyhis bid in light of this privileged information.
Mead, Moseidjord, and Sorensen (1983, 1984) ever, it is clear that only a small
showed that, as this argument leads one to expect, percentage of stock market trades can be
returnsto successfulbidders on drainageleases were rationalized in this way. The majority of
exceptionally high when these bidders were those
who had already leased neighboring tracts. Mead, trades appear to reflect belief on the part
Moseidjord, and Sorensen suggested that the low of each investor that he can outwit other
returns to successful bidders on wildcat tracts are investors, which is inconsistent with
consistent with market efficiency when allowance is
made for the value of the informationgained by the common knowledge of rationality.
successful bidder about neighboringtracts. The Milgrom, Stokey, and Tirole re-

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
1612 Journal of Economic Literature, Vol. XXVII (December 1989)

sult poses a problem: Either analysts of explain ex post more than a small fraction
financial markets must ignore the exis- of the variation in individual stock prices,
tence of high volumes of securities trad- even using data like industry average
ing or they must incorporate irrationality prices and market price indices as ex-
into their models, at least when analyzing planatory variables.
complete-market environments. Given It would seem almost self-evident that
the traditional hostility toward irrational- the recent wave of leveraged buyouts
ity as manifested, for example, in Shil- provides strong evidence against market
ler's fad variables, neither alternative is efficiency: The astronomical fees to in-
attractive. Fortunately, Fischer Black vestment bankers that these mergers
(1986) came to the rescue. By renaming generate are difficult to reconcile with
irrational trading "noise trading" Black any nontautologous version of market ef-
avoided the I-word, thereby sanitizing ficiency, as are the stock price gyrations
irrationality and rendering it palatable to that accompany leveraged buyouts.
many analysts who in other settings Mergers themselves, of course, are con-
would not be receptive to such a specifi- sistent with efficiency; indeed, they are
cation. The economic effects of noise implied by efficiency if they result in syn-
traders is now an active research area ergies in operations or serve to remove
(Campbell and Albert Kyle 1986). bad management. However, most stu-
Inasmuch as efficient-markets theory dents of corporate takeovers believe that
attributes asset price changes exclusively such effects are of secondary importance.
to information about fundamentals, it im- On Roll's (1986) account, takeovers may
plies that returns should be explainable be consistent with market efficiency even
ex post by fundamentals. Curiously, fi- if motivated solely by the "hubris"of the
nancial economists have until recently acquiring group. Roll interpreted the
displayed a marked lack of interest in stock price declines that typically follow
testing this implication of market effi- takeovers as validating the pretakeover
ciency, strongly preferring instead to valuation of the firm on the part of the
concentrate their attention on testing the large majority of investors, and as invali-
martingale implication that returns dating the runup that occurs upon take-
should not be explainable by fundamen- over. The majority of traders, then, value
tals ex ante (see Summers 1985 for dis- the firm correctly; only the acquirer is
cussion). However, two recent studies by led by "hubris" to overpay. Roll argued
Roll are distinguished exceptions. After from this that "the market," which he
persuasively arguing that information on identified with the majority of traders,
weather in Florida-specifically, infor- is efficient. This argument will not do
mation bearing on the probability of a at all. The simplest efficient-markets rea-
freeze, which would adversely affect the soning implies that no systematic pattern
orange crop-should be the dominant in- of price decline should occur in the wake
fluence on orange juice futures prices, of a publicly known event like a success-
Roll (1984) showed that weather informa- ful takeover. Further, as proponents of
tion could explain empirically only a market efficiency themselves insist in
small fraction of the variation in these other contexts, the market price of a com-
prices. He could not identify any variable pany is the price that the firm trades at,
that explained the remainder of the varia- no more and no less. Even (in fact, espe-
tion. In his presidential address to the cially) within the logic of efficient capital
American Finance Association, Roll market theory, which rejects any distinc-
(1988) continued along the same lines, tion between market price and "true
showing empirically that it is difficult to value," no case whatever can be made

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
LeRoy: Efficient Capital Markets and Martingales 1613

for discounting the price runup on the The failure of many financial econo-
grounds that only a minority of traders mists to appreciate the extent of the gulf
are involved. separating market efficiency interpreted
Finally, we have the October 19, 1987, as economic equilibrium and market effi-
stock market selloff. As readers are well ciency interpreted as the martingale
aware, stock values dropped half a trillion model has led them to vacillate between
dollars on that single day in the complete viewing market efficiency, on one hand,
absence of news that can plausibly be as hard-wired into their intellectual capi-
related to market fundamentals. The un- tal and unfalsifiable and, on the other
deniable and spectacular presence of hand, as consisting of a specific class of
nonfundamental factors affecting stock falsifiable models of asset prices. In ab-
prices on Black Monday renders more stract discussions, financial economists
credible the presence, and perhaps dom- almost always characterize market effi-
inance, of similar factors when the stock ciency as a specific theory which in prin-
market is functioning normally. ciple is falsifiable, but which in practice
turns out not to be falsified empirically.
VIII. Conclusion At an applied level, however, they fre-
quently find it difficult to specify con-
The central idea of efficient capital
cretely what evidence would in principle
market theory is that securities prices are
contradict the theory. This is most evi-
determined by the interaction of self-in-
dent in Fama's (1970) discussion, where
terested rational agents. At this most ba-
market efficiency was described as a sub-
sic level, the assertion that capital mar-
stantive theory generating falsifiable pre-
kets are efficient therefore reduces to the
dictions, but where at the same time the
assertion that it is economic theory rather
mathematical formulation of the market
than any other discipline that provides
efficiency was tautologous. Further, it
the analytical tools appropriate for under-
was noted in Section IV that several
standing securities pricing. The intuitive
pieces of evidence that seemed to contra-
presentation of efficient capital market
dict market efficiency were dismissed by
theory in the introduction was intended
Fama for reasons that were not made
to convey its essential identity with eco-
clear.
nomic theory. Empirical tests of capital
There is no shortage of other examples
market efficiency, however, are in prac-
of lack of clarity and consistency in dis-
tice usually tests of the martingale model.
cussions of capital market efficiency.
This survey should by now have made
amply clear that the transition between
the intuitive idea of market efficiency and
tion of the no-arbitragecondition implies the exis-
the martingale model is far from direct. tence of a consistent equilibriumprice system, Ross'
Few financial economists, surprisingly, identificationof market efficiency with the absence
have taken direct issue with the prevail- of arbitrage opportunities is essentially equivalent
ing practice in the finance literature of to our identification in the introduction of market
efficiency with economic equilibrium. Now, most
identifying market efficiency with the va- economists regardthe propositionthat the data they
lidity of a particular specialized model observe were generated by some, as opposed to a
of equilibrium in financial markets.27 particular, equilibrium model as an untestable ex-
pression of a preferred researchmethod. If so, Ross'
definitionimplies that marketefficiencyis untestable,
27 Ross (1987) is an exception. Ross proposed as if and that therefore the entire empiricalliterature on
it were self-evident that the intuition of market effi- market efficiency is beside the point. Despite the
ciency is essentially that of no arbitrage,rather than considerable merits of Ross' characterizationof mar-
the martingale model or rational expectations. Be- ket efficiency, it is seen to be at odds with the re-
cause (loosely) any equilibriumprice system implies ceived practice, which emphasizes the testability of
satisfactionof the no-arbitragecondition, and satisfac- market efficiency.

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
1614 Journal of Economic Literature, Vol. XXVII (December 1989)

Merton (1987) went out of his way to em- have always displayed a strong prefer-
phasize that the hypothesis of stock mar- ence for empirical tests in which market
ket rationality is not tautologous: Market efficiency implies the absence of a pat-
efficiency is "not consistent with models tern, such as return autocorrelation tests,
or empirical facts which imply that either over tests that do not have such a charac-
stock prices depend in an important way terization, such as variance-bounds tests.
on factors other than the fundamentals Therefore they have been led to dismiss
* . .or that . . . investors can systemati- out of hand some of the most important
cally identify significant differences be- evidence bearing on market efficiency.
tween stock prices and fundamental Finally, financial economists' preference
value." Yet Terry Marsh and Merton for arbitrage-based over equilibrium-
(1986) interpreted the variance-bounds based arguments (together with the pre-
violations, which would seem to raise dilection noted above for tautologous for-
questions about the empirical validity of mulations of market efficiency) has di-
both these attributes of market effi- verted them from attempting to specify
ciency, as constituting evidence against intellectually coherent alternatives to
the assumed stationarity of dividends market efficiency, and from analyzing the
rather than as conflicting with market ef- econometric properties of these alterna-
ficiency. Apparently when the evidence tives relative to the null hypothesis of
is favorable, market efficiency is sup- market efficiency. Thus they have not se-
ported, but when the evidence is unfa- riously considered the possibility that
vorable, market efficiency is treated as many of the econometric tests that favor
part of the maintained hypothesis, insu- market efficiency have little power to re-
lated from falsification. Another example ject reasonable alternative hypotheses.
of the extreme reluctance, bordering on The foregoing discussion suggests that
inability, of proponents of efficient capital financial economists have not always suc-
markets to acknowledge contrary evi- ceeded in applying to efficient capital
dence is Roll's (1986) "hubris" hypothesis market theory the same high standards
of corporate takeovers, discussed in the of rigor and consistency that they have
preceding section. exhibited in other areas of finance. As a
Several attributes of financial econo- result, they have for the most part been
mists' outlook help explain the extraordi- able to avoid confronting the conclusion
nary durability of the widely held opinion that is warranted by the evidence: How-
that the bulk of the empirical evidence ever attractive (to economists) capital
favors capital market efficiency. As ob- market efficiency is on methodological
served in Section VII, financial econo- grounds, it is extraordinarily difficult to
mists at once insist on the central impor- formulate nontrivial and falsifiable impli-
tance of their contention that asset prices cations of capital market efficiency that
are determined exclusively by funda- are not in fact falsified.
mentals, and at the same time have been Empirical rejection of the martingale
unreceptive to attempts to determine model suggests that there exist trading
empirically whether price changes are in rules that increase expected returns rela-
fact traceable to fundamentals, at least tive to buy-and-hold. It is this implication
until recently. Accordingly, it has been that advocates of market efficiency have
only recently that they have come to ap- always found implausible: Even if it is
preciate that fundamentals appear to ex- conceded that some or most traders act
plain ex post only a small portion of price irrationally, why would rational traders
changes. Further, financial economists not exploit the patterns, and in so doing

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
LeRoy: Efficient Capital Markets and Martingales 1615

bid them away? The simplest answer to some settings) that the risk takers as a
this question is that optimal trading by whole do better than the risk avoiders,
rational agents will completely reverse even though individual risk takers will
the effects of irrational trades on prices suffer bad outcomes with higher fre-
only if the rational agents are well fi- quency than individuals who are fully ra-
nanced and risk-neutral. The need for tional. Thus irrationality may actually be
substantial wealth on the part of rational rewarded in the aggregate.
agents is obvious: The existence of a Market efficiency is a complex joint hy-
lower bound (zero) on any agent's con- pothesis. Some elements of this joint hy-
sumption in any state implies the exis- pothesis are central to economists' way
tence of bounds on that agent's security of thinking, like rationality and rational
purchases and sales.28 Existence of these expectations, while others are no more
bounds is consistent with rational agents than convenient auxiliary assumptions,
completely offsetting the effect on prices like the martingale model. Rejection of
of irrational agents' trades only if the market efficiency requires that one or
bounds are not binding, which will occur more of these elements of the joint hy-
only if the rational agents' wealth is large. pothesis be replaced. Understandably,
The need for risk neutrality is equally economists have focused their critical at-
obvious: If rational agents are risk- tention on those elements that can be
averse, they will find that the portfolio discarded with the least damage to their
they would have to acquire in order com- research programs. We have already
pletely to reverse the effects of irrational seen that the effort to generalize the mar-
trades imply excessive risk. tingale model to allow for risk aversion
A related argument for full rationality has not succeeded empirically so far.
is sometimes put in sociobiological While it is possible that this work will
terms. Traders who act irrationally will, succeed better in the future than it has
it is suggested, lose wealth on average. in the past, several considerations sug-
Like any group of individuals whose abil- gest that the problems with market effi-
ity to survive and reproduce is impaired ciency go deeper.
by a dysfunctional genetic mutation, the The high volume of trade on organized
irrational agents will eventually disap- securities markets poses a serious prob-
pear from the population. In a series of lem; no minor tinkering with efficient-
recent papers, Bradford De Long et al. markets models seems likely to provide
(1988a, 1988b, 1989a, 1989b), however, an intelligible reason why rational agents
questioned this reasoning. If the irra- would exchange securities as much as
tional behavior of the nonoptimizers con- real-world market participants do. The
sists of taking risks based on unrealisti- willingness of investors to pay for infor-
cally optimistic appraisals of possible mation is equally problematic: As noted
outcomes, this irrationality may have ef- in the introduction, if the purchased in-
fects that are indistinguishable from low formation makes profitable trades possi-
risk aversion. Because in a population of ble, securities markets cannot be infor-
risk-averse agents the average rewards mationally efficient, while if it does,
to risk takers exceed those to risk avoid- agents are irrationally wasting their
ers, the law of large numbers implies (in money. Neither is consistent with effi-
ciency. These considerations suggest that
28It is true that by borrowingagents can acquire
a large number of market participants act
investments that exceed their total assets in value,
but insofar as borrowings are secured by future as if they do not believe that the market
wealth, their amount remains bounded. is efficient. While there may be some

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
1616 Journal of Economic Literature, Vol. XXVII (December 1989)

sense in which securities markets can be that make indiscriminate use of irration-
efficient even though most agents act as ality and nonrational expectations cannot
if they believe them to be inefficient, the impose discipline on economists' think-
argument is far from transparent, to say ing about securities markets. Nonethe-
the least. Regrettably, it appears as if it less, there is no reason in principle to
is the assumptions of rationality and ra- believe that these objections cannot be
tional expectations that require reforma- met. It is a task to which economists,
tion. working with psychologists, would do
The recent literature on cognitive psy- well to turn their attention.
chology (e. g., Kenneth Arrow 1982; Dan- If it is accepted that successful models
iel Kahneman, Paul Slovic, and Amos of securities prices will require a broader
Tversky 1982; Robin Hogarth and Melvin analytical framework than has been
Reder 1987; Mark Machina 1987) pro- adopted up to now, it follows that the
vides a promising avenue for future re- routine use of efficient-markets reasoning
search. Cognitive psychologists have will require reassessment. Some argu-
documented systematic biases in the way ments based on appeal to market effi-
people use information and make deci- ciency will remain valid, while others will
sions. Some of these biases are easy to have to be discarded. The most funda-
connect, at least informally, with securi- mental insight of market efficiency-the
ties market behavior. For example, reminder that asset prices reflect the in-
agents allow their decisions to be dis- teraction of self-interested agents-will
torted by the presence of points of refer- remain. However, the contention that no
ence that should be irrelevant ("anchor- successful trading rule can be based on
ing"). Further, they systematically publicly available information may have
overweight current information and un- to go; it is this strict version of market
derweight background information rela- efficiency that produces the empirical im-
tive to what Bayes' theorem implies. To plications that the evidence contradicts.
be sure, most of the evidence for these The most radical revision in efficient-
biases comes from experiments and ques- markets reasoning will involve those im-
tionnaires. Economists have in the past plications of market efficiency that de-
confidently assumed that these biases pend on asset prices equaling or closely
would disappear in settings where the approximating fundamental values. The
stakes are high, as in real-world securities evidence suggests that, contrary to the
markets. However, this line is beginning assertion of this version of efficient mar-
to wear thin, particularly in light of econ- kets theory, such large discrepancies be-
omists' continuing inability to explain as- tween price and fundamental value regu-
set prices using models that assume away larly occur.29 The implication is that
cognitive biases. there may be a constructive role for gov-
The problem with the cognitive psy- ernment in altering or regulating the op-
chology literature is that it is more suc- eration of securities markets. Those who
cessful in providing after-the-fact expla- think of governments as engines of Pareto
nations for observed behavior than in optimality will interpret the evidence
generating testable predictions. Econo- summarized here as in fact justifying such
mists require from their theories a clear 29 Black(1986),in tacit recognitionof the frequency
statement of what observed phenomena with which major discrepancies between prices and
would be inconsistent with these theo- values occur, defined a market as efficient if price
is within a factor of two of value, and estimated that
ries, and so far this has not been forth- U.S. capital marketsare efficient on the order of 90
coming from the psychologists. Models percent of the time.

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
LeRoy: Efficient Capital Markets and Martingales 1617

an enlarged role for government. The COOTNER, PAUL H., ed. The random character of

rest of us, however, will continue to re- stock marketprices. Cambridge.MIT Press, 1964.
COPELAND, THOMAS E. AND WESTON, J. FRED. Finan-
ject major changes along these lines, cial theory and corporatepolicy. 2nd ed. Reading,
while acknowledging that the case MA: Addison-Wesley, 1983.
against such changes is not as clear-cut COWLES, ALFRED. "Can Stock Market Forecasters
Forecast?"Econometrica,July 1933, 1(4), 309-24.
as it once seemed. Cox, JOHN; INGERSOLL, JONATHAN E. AND Ross, STE-
PHEN A. "ATheory of the Term Structureof Inter-
REFERENCES est Rates," Econometrica, Mar. 1985, 53(2), pp.
385-407.
ALEXANDER, SIDNEY. "Price Movements in Specula- Cox, JOHN: Ross, STEPHEN A. AND RUBINSTEIN,
tive Markets:Trends or Random Walks?"(1961) MARK. "Option Pricing: A Simplified Approach,"
in COOTNER, 1964(a),pp. 199-218. J. Finan. Econ., Sept. 1979, 7(3), pp. 229-63,
. "Price Movements in Speculative Markets: CROSS, FRANK. "The Behaviorof Stock Prices on Fri-
Trends or RandomWalks, no. 2," (1964) in COOT- days and Mondays,"Financial Analysts J., Nov./
NER 1964(b), pp. 338-72. Dec. 1973, 29(6), pp. 67-69.
ARIEL, ROBERT A. "A Monthly Effect in Stock Re- DANTHINE, JEAN-PIERRE. "Martingale,Market Effi-
turns,"J. Finan. Econ., 1987, 18(1), pp. 161-74. ciency and Commodity Prices," Europ. Econ.
ARROW, KENNETH J. "RiskPerception in Psychology Rev., Oct. 1977, 10(1), pp. 1-17.
and Economics,"Econ. Inquiry, Jan. 1982, 20(1), DEBONDT, WERNER F. M. AND THALER, RICHARD.
pp. 1-9. "Does the Stock Market Overreact?"J. Finance,
BACHELIER, Louis. "Theory of Speculation, in July 1985, 40(3), pp. 793-805.
COOTNER 1964, pp. 17-78 (first published 1900). . "FurtherEvidence on InvestorOverreaction
BANZ, ROLF. "TheRelationshipBetween Return and and Stock Market Seasonality,"J. Finance, July
MarketValueof CommonStocks,"J. Finan. Econ., 1987, 42(3), pp. 557-81.
Mar. 1981, 9(1), pp. 3-18. DE LONG, J. BRADFORD ET AL. "TheSurvivalof Noise
BASU, SANJOY. "InvestmentPerformanceof Common Tradersin FinancialMarkets,"NBER WorkingPa-
Stocks in Relation to Their Price-EarningsRatios: per No. 2715, Sept. 1988a.
A Test of the Efficient MarketHypothesis,"J. Fi- . "Noise Trader Risk in Financial Markets,"
nance, June 1977, 32(3), pp. 663-82. reproduced, HarvardU., 1988b.
_ . "The Relationship Between Earnings' Yield, . "The Size and Incidence of the Losses From
Market Value and Returns for NYSE Common Noise Trading,"J. Finance, July 1989a, 44(3), pp.
Stocks:Further Evidence," J. Finan. Econ., June 681-96.
1983, 12(1), pp. 129-56. . "Positive Feedback Investment Strategies
BLACK, FISCHER. "Noise," J. Finance, 1986, 41(3), and Destabilizing Rational Speculation," NBER
529-43. Working Paper No. 2880, Mar. 1989b.
BREEDEN, DOUGLAS. "An IntertemporalAsset Pric- DIEBOLD, FRANCIS X. "Testingfor Bubbles, Reflect-
ing Model with Stochastic Consumption and In- ing Barriersand Other Anomalies,"J. Econ. Dy-
vestment Opportunities,"J. Finan. Econ., Sept. nam. Control, Mar. 1988, 12(1), pp. 63-70.
1979, 7(3), pp. 265-96. DIMSON, ELROY, ed. Stock market anomalies. Cam-
BROCK, WILLIAM, A. "Asset Prices in a Production bridge: CambridgeU. Press, 1988.
Economy," in The economics of information and DREMAN, DAVID. The new contrarian investment
uncertainty. Ed.: JOHN J. MCCALL. Amsterdam: strategy. NY: RandomHouse, 1982.
North-Holland,1982, pp. 1-42. FAMA, EUGENE F. "EfficientCapital Markets:A Re-
CAMPBELL, JOHN Y. AND KYLE, ALBERT. "Smart view of Theory and EmpiricalWork,"J. Finance,
Money, Noise Tradingand Stock Price Behavior." May 1970, 25(2), pp. 383-417.
Unpublished ms., Princeton U., 1986. . Foundations of finance. NY: Basic Books,
CAMPBELL, JOHN Y. AND SHILLER, ROBERT J. "The 1976a.
Dividend-Price Ratio and Expectations of Future . "Efficient Capital Markets: Reply," J. Fi-
Dividends and Discount Factors," Rev. Finan. nance, Oct. 1976b, 3(4), pp. 143-45
Stud., 1988a, 1, pp. 195-228. FAMA, EUGENE F. AND BLUME, MARSHALL. "Filter
. "Stock Prices, Earnings, and Expected Divi- Rules and Stock Market Trading,"J. Bus., Jan.
dends,"J. Finance, 1988b, 43(3), pp. 661-76. 1966, 39(1), pp. 226-41.
CLARK, RoSS AND ZIEMBA, WILLIAM T. "Playingthe FAMA, EUGENE F. ET AL. "The Adjustmentof Stock
Turn-of-the-YearEffect with Index Futures," Op- Prices to New Information,"Int. Econ. Rev., Feb.
erations Research, Nov./Dec. 1987, 35(6), pp. 799- 1969, 10(1), pp. 1-21.
813. FAMA, EUGENE F. AND FRENCH, KENNETH R. "Per-
COCHRANE, JOHN H. "How Big Is the RandomWalk manent and Temporary Components of Stock
in GNP?"J. Polit. Econ., 1988, 96(5), pp. 893- Prices,"J. Polit. Econ., Apr. 1988a,96(2), pp. 246-
920. 73.
CONSTANTINIDES, GEORGE M. "Habit Formation:A . "Dividend Yields and Expected Stock Re-
Resolution of the Equity Premium Puzzle." Un- turns," J. Finan. Econ., 1988b, 22(1), pp. 3-25.
published ms., U. of Chicago, Dec. 1988. FAMA, EUGENE F. AND MAcBETH, JAMES D. "Risk,

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
1618 Journal of Economic Literature, Vol. XXVII (December 1989)

Return, and Equilibrium:EmpiricalTests,"J. Po- HARRISON, J. MICHAEL AND KREPS, DAVID M.


lit. Econ., May/June1973, 81(3), pp. 607-36. "Speculative Investor Behavior in a Stock Market
FLAVIN, MARJORIEA. "ExcessVolatilityin the Finan- with Heterogeneous Expectations," Quart. J.
cial Markets:A Reassessmentof the EmpiricalEvi- Econ., May 1978, 92(2), pp. 323-36.
dence, "J. Polit. Econ., Dec. 1983, 91(6), pp. 929- . "Martingales and Arbitrage in Multiperiod
56. Securities Markets," J. Econ. Theory, June 1979,
FRENCH, KENNETH. "StockReturnsand the Weekend 20, pp. 381-408.
Effect,"J. Finan. Econ., Mar. 1980, 8(1), pp. 55- HARRISON, J. MICHAEL AND PLISKA, S. R. "Martin-
69. gales and Stochastic Integrals in the Theory of Con-
FRENCH, KENNETH AND ROLL, RICHARD W. "Stock tinuous Trading," Stochastic Processes and Their
Return Variances:The Arrivalof Informationand Applications, 1981, 11, pp. 21560.
the Reaction of Traders,"J. Finan. Econ., Sept. HENDRICKS, KENNETH AND PORTER, ROBERT H. "Em-
1986, 17(1), pp. 5-26. pirical Study of an Auction with Asymmetric Infor-
GIBBONS, MICHAEL AND HESS, PATRICK. "Day of the mation,"Amer. Econ. Rev., Dec. 1988, 78(5), pp.
Week Effects and Asset Returns,"J. Bus., Oct. 865-83.
1981, 54(4), pp. 579-96. HOGARTH, ROBIN M. AND REDER, MELVIN W. Ra-
GILIBERTO, S. MICHAEL AND VARAIYA, NIKHIL P. "The tional choice: The contrast between economics
Winner'sCurse and Bidder Competitionin Acqui- and psychology. Chicago: U. of Chicago Press,
sitions: Evidence from Failed Bank Auctions,"J. 1987.
Finance, Mar. 1989, 44(1), pp. 59-76. HUANG, CHI-FU AND LITZENBERGER, ROBERT H.
GILLES, CHRISTIAN. "Chargesas Equilibrium Prices Foundationsfor financial economics. NY: North-
and Asset Bubbles,"J. Math. Econ., forthcoming. Holland, 1988.
GILLES, CHRISTIAN AND LERoY, STEPHEN F. "A Note INGERSOLL, JONATHAN E., JR. Theoryoffinancial de-
on The Local ExpectationsHypothesis:A Discrete- cision-making. Totowa, NJ: Rowman and Little-
Time Exposition,"J. Finance, Sept. 1986, 41(4), field, 1987.
pp. 975-79. JAFFE, JEFFREY AND WESTERFIELD, RANDOLPH. "The
. "Econometric Aspects of the Variance- Week-End Effect in Common Stock Returns: The
Bounds Tests." Reproduced, U. of California, International Evidence," J. Finance, June 1985,
Santa Barbara,Jan. 1988a. 40(2), pp. 433-54.
. "Bubbles and Charges." Reproduced, U. of JENSEN, MICHAEL. "The Performance of Mutual
California,Santa Barbara,1988b. Funds in the Period 1945-1964," J. Finance, May
GRAHAM, BENJAMIN AND DODD, DAVID L. Security 1968, 23(2), pp. 389-416.
analysis. NY: McGraw-Hill,1934. . "Risk, the Pricing of Capital Assets, and the
GRANGER, CLIVE W. J. AND MORGENSTERN, OSKAR. Evaluation of Investment Portfolios," J. Bus., Apr.
"Spectral Analysis of New York Stock Market 1969, 42(2), pp. 167-247.
Prices," (1963) in COOTNER 1964, pp. 162-88. JORDAN, JAMES. "On the Efficient Markets Hypothe-
GROSSMAN, SANFORD J. "FurtherResultson the Infor- sis," Econometrica, Sept. 1983, 51(5), pp. 1325-
mational Efficiency of Competitive Stock Mar- 43.
kets," J. Econ. Theory, June 1978, 18(1), pp. 81- KAHNEMAN, DANIEL; SLOVIC, PAUL AND TVERSKY,
101. AMOS. Judgmentunder uncertainty:Heuristicsand
GROSSMAN, SANFORD AND SHILLER, ROBERT J. "The biases. Cambridge: Cambridge U. Press, 1982.
Determinants of the Variabilityof Stock Market KEIM, DONALD B. "Size-Related Anomalies and Stock
Prices," Amer. Econ. Rev., May 1981, 71(2), pp. Return Seasonality," J. Finan. Econ., 1983, 12(1),
222-27. pp. 13-32.
GROSSMAN, SANFORD AND STIGLITZ, JOSEPH E. "Infor- . "Stock Market Regularities: A Synthesis of
mation and Competitive Price Systems," Amer. the Evidence and Explanations," in DIMSON1988,
Econ. Rev., May 1976, 66(2), pp. 246-53. pp. 16-39.
. "On the Impossibility of Informationally Effi- KEIM, DONALD B. AND STAMBAUGH, ROBERT F. "A
cient Markets," Amer. Econ. Rev., June 1980, Further Investigation of the Weekend Effect in
70(3), pp. 393-408. Stock Returns," J. Finance, July 1984, 39(3), pp.
GULTEKIN, MUSTAFA N. AND GULTEKIN, N. BULENT. 819-37.
"Stock Return Anomalies and the Tests of the KENDALL, MAURICE G. "The Analysis of Economic
APT," J. Finance, Dec. 1987, 42(5), pp. 1213- Time-Series, Part I: Prices," (1953) in COOTNER
24. 1964, pp. 85-99.
HANSEN, LARS AND SINGLETON, KENNETH J. "Gener- KEYNES, JOHN MAYNARD. General theory of employ-
alized InstrumentalVariables Estimation of Non- ment, interest and money. NY: Harcourt, Brace,
linear Rational Expectations Models," Econo- 1936.
metrica, Sept. 1982, 50(5), pp. 1269-86. KIM, MYUNG JIG; NELSON, CHARLES R. AND STARTZ,
. "Stochastic Consumption, Risk Aversion, RICHARD. "Mean Reversion in Stock Prices? A Re-
and the Temporal Behavior of Asset Returns,"J. appraisal of the Empirical Evidence." Reproduced,
Polit. Econ., Apr. 1983, 91(2), pp. 249-65. U. of Washington, May 1988.
HARRIS, LAWRENCE. "A TransactionData Study of KLEIDON, ALLAN W. "Variance Bounds Tests and
Weekly and IntradailyPatternsin Stock Returns," Stock Price Valuation Models," J. Polit. Econ.,
J. Finan. Econ., May 1986, 16(1), pp. 99-117. Oct. 1986a, 94(5), pp. 953-1001.

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
LeRoy: Efficient Capital Markets and Martingales 1619

._ "Bias in Small Sample Tests of Stock Price Stock Market Volatility,"J. Finance, July 1985,
Rationality,"J. Bus., Apr. 1986b, 59(2), pp. 237- 40(3), pp. 677-87.
61. MARSH, TERRY AND MERTON, ROBERT C. "Dividend
. "The Probability of Gross Violations of a Variability and Variance Bounds Tests for the Ra-
PresentValueVarianceInequality:Reply,"J. Polit. tionality of Stock Market Prices," Amer. Econ.
Econ., Oct. 1988a, 96(5), pp. 1093-96. Rev., June 1986, 76(3), pp. 483-98.
. "Bubbles, Fads and Stock Price Volatility McAFEE, R., PRESTON AND MCMILLAN, JOHN. "Auc-
Tests:A PartialEvaluation:Discussion,"J. Finance tions and Bidding,"J. Econ. Lit., June 1987, 25(2),
1988b, 43(3), pp. 656-59. pp. 699-738.
LAKONISHOK,JOSEF AND LEVI, MAURICE. "Weekend MCCLOSKEY, DONALD N. The writing of economics.
Effects on Stock Returns: A Note,"J. Finance, June NY: Macmillan, 1987.
1982, 37(3), pp. 883-89. MEAD, WALTERJ.; MOSEIDJORD, ASBJORNAND SOREN-
LAKONISHOK, JOSEF AND SMIDT, SEYMOUR. "Volume SEN, PHILIP E. "The Rate of Return Earned by
and Turn-of-the-YearBehavior,"J. Finan. Econ., Lessees under Cash Bonus Bidding for OCS Oil
Sept. 1984, 13(3), pp. 435-55. and Gas Leases," Energy J., Oct. 1983, 4(2), pp.
. "Are Seasonal Anomalies Real?: A Ninety- 37-52.
Year Perspective," Rev. Finan. Stud., Winter . "Competitive Bidding under Asymmetrical
1988, 1(4), pp. 403-25. Information:Behavior and Performance in Gulf
LERoY, STEPHEN F. The determination of stock of Mexico Drainage Lease Sales, 1959-1969,"
prices. Unpub. PhD dissertation, U. of Pennsylva- Rev. Econ. Statist., Aug. 1984, 66(3), pp. 505-
nia, 1971. 08.
. "Risk Aversion and the Martingale Property MEHRA, RAJNISH AND PRESCOTT, EDWARD C. "The
of Stock Prices,"Int. Econ. Rev., June 1973, 14(2), Equity Premium: A Puzzle," J. Monet. Econ.,
pp. 436-46. Mar. 1985, 15(2), pp. 145-61.
. "Efficient Capital Markets: Comment," J. MERTON, ROBERTC. "AnIntertemporalCapitalAsset
Finance, Mar. 1976, 3(1), pp. 139-41. Pricing Model," Econometrica, Sept. 1973, 41(5),
. "ExpectationsModels of Asset Prices:A Sur- pp. 867-87.
vey of Theory,"J. Finance, Mar. 1982, 37(1), pp. . "On the Current State of the Stock Market
185-217. Rationality Hypothesis," in Macroeconomicsand
. "Efficiency and the Variability of Asset finance: Essays in honor of Franco Modigliani.
Prices," Amer. Econ. Rev., May 1984, 74(2), pp. Ed.: STANLEY FISCHER ET AL. Cambridge: MIT
183-87. Press, 1987.
LERoY, STEPHEN F. AND LACIVITA, C. J. "Risk MICHENER, RONALD W. "VarianceBounds in a Sim-
Aversion and the Dispersion of Asset Prices," J. ple Model of Asset Pricing,"J. Polit. Econ., Feb.
Bus., Oct. 1981, 54(4), pp. 535-47. 1982, 90(1), 166-75.
LERoY, STEPHEN F. AND PARKE, WILLIAM C. "Stock MILGROM, PAUL AND STOKEY, NANCY. "Information,
Price Volatility:An Inequality Test Based on the Tradeand Common Knowledge,"J. Econ. Theory,
GeometricRandomWalk." Reproduced,U. of Cal- Feb. 1982, 26(1), pp. 17-27.
ifornia, Santa Barbara,May 1988. MILLER, EDWARD M. "Risk,Uncertainty,and Diver-
LERoY, STEPHEN F. AND PORTER, RICHARD D. "The gence of Opinion," J. Finance, Sept. 1977, 32(4),
Present-Value Relation: Tests Based on Implied pp. 1151-68.
VarianceBounds,"Econometrica,May 1981, 49(3), MOSSIN, JAN. "Equilibriumin a Capital Asset Mar-
pp. 555-74. ket," Econometrica, Oct. 1966, 34(4), pp. 768-
LINTNER, JOHN. "The Valuation of Risk Assets and 83.
the Selection of RiskyInvestments in Stock Portfo- NASON, J. "The Equity Premium and Time-Varying
lios and Capital Budgets," Rev. Econ. Stat., Feb. Risk Behavior." Reproduced, Federal Reserve
1965, 47(1), pp. 13-37. Board, 1988.
Lo, ANDREW W. AND MACKINLAY, A. CRAIG. "Stock NICHOLSON, FRANCIS. "Price Ratios in Relation to
MarketPrices Do Not Follow RandomWalks:Evi- Investment Results," Financial Analysts J.., Jan.!
dence from a Simple SpecificationTest," Rev. Fi- Feb. 1968, 24(1), pp. 105-09.
nan. Stud., Spring 1988, 1(1), pp. 41-66. NIEDERHOFFER, VICTOR AND OSBORNE, M. F. M.
LUCAS, ROBERT E., JR. "Some International Evidence "Market Making and Reversal of the Stock Ex-
on Output-Inflation Tradeoffs," Amer. Econ. Rev., change," J. Amer. Statist. Assoc., Dec. 1966,
June 1973, 63(3), pp. 326-34. 61(316), pp. 897-916.
. "Asset Prices in an Exchange Economy," OHLSON, JAMES. "Risk-Aversionand the Martingale
Econometrica,Nov. 1978, 46(6), pp. 1429-45. Property of Stock Prices: Comments," Int. Econ.
MACHINA, MARK. "Choice under Uncertainty:Prob- Rev., Feb. 1977, 18(1), pp. 229-34.
lems Solved and Unsolved,"J. Econ. Perspectives, POTERBA, JAMES AND SUMMERS, LAWRENCE. "The
Summer 1987, 1(1), pp. 121-54. Persistence of Volatilityand Stock MarketFluctua-
MANDELBROT, BENOIT. "Forecasts of Future Prices, tions," Amer. Econ. Rev., Dec. 1986, 76(5), pp.
Unbiased Markets, and 'Martingale' Models," J. 1142-51.
Bus., Jan. 1966, 39(1), pp. 242-55. . "Mean Reversion in Stock Prices: Evidence
MANKIW, N. GREGORY;ROMER, DAVID AND SHAPIRO, and Implications,"J. Finan. Econ., -Oct. 1988,
MATrHEW D. "An Unbiased Reexamination of 22(1), pp. 27-59.

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
1620 Journal of Economic Literature, Vol. XXVII (December 1989)

REINGANUM, MARC. "Misspecificationof Capital As- ._ "Proof That Properly Discounted Present
set Pricing: EmpiricalAnomalies Based on Earn- Values of Assets VibrateRandomly,"Bell J. Econ.,
ings' Yields and MarketValues,"J. Finan. Econ., Autumn 1973, 4(2), pp. 369-74.
Mar. 1981, 9, pp. 19-46. SCHWERT,G. WILLIAM."Sizeand StockReturns,and
. "A Direct Test of Roll's Conjecture on the Other Empirical Regularities," J. Finan. Econ.,
Firm Size Effect,"J. Finance, Mar. 1982, 37(1), June 1983, 12(1), pp. 3-12.
pp. 27-35. SHARPE,WILLIAMF. "CapitalAsset Prices:A Theory
. "The Anomalous Stock Market Behavior of of MarketEquilibriumunder Conditionsof Risk,"
Small Firms in January:EmpiricalTests for Tax- J. Finance, Sept. 1964, 19(4), pp. 425-42.
loss Selling Effects,"J. Finan. Econ., June 1983, SHILLER,ROBERTJ. "The Volatility of Long-Term
12(1), pp. 89-104. Interest Rates and Expectations Models of the
RICARDO, DAVID. Principlesof political economyand Term Structure,"J. Polit. Econ., Dec. 1979, 87(6),
taxation. London:Aldine, [1817] 1960. pp. 1190-1219.
RIETZ, THOMAS A. "The Equity Risk Premium: A . "The Use of Volatility-Measures in Assessing
Solution,"J. Monet. Econ., July 1988, 22(1), pp. Market Efficiency,"J. Finance, May 1981a, 36(2),
117-31. pp. 291-304.
ROBERTS, HARRY V. "Stock-Market'Patterns'and Fi- . "Do Stock Prices Move Too Much to Be
nancial Analysis: Methodological Suggestions," Justified by Subsequent Changes in Dividends?"
(1959) in COOTNER 1964, pp. 7-16. Amer. Econ. Rev., June 1981b, 71(3), pp. 421-
ROGALSKI, ROBERT. "New Findings RegardingDay- 36.
of-the-WeekReturnsOver Tradingand Non-Trad- . "Stock Prices and Social Dynamics," Brook-
ing Periods:A Note,"J. Finance, Dec. 1984, 39(5), ings Pap. Econ. Act., 1984, 2, pp. 457-98.
pp. 1603-14. . "The Probability of Gross Violations of a
ROLL, RICHARD. "VasIst Das? The Turn-of-the-Year Present Value Variance Inequality," J. Polit.
Effect and the Return Premia of Small Firms," Econ., Oct. 1988, 96(5), pp. 1089-92.
J. Portfol. Manage., Winter 1983, 9(2), pp. 18- SIMS, CHRISTOPHER J. "Martingale-LikeBehavior of
28. Prices and Interest Rates." Discussion Paper No.
"Orange Juice and Weather,"Amer. Econ. 205. U. of Minnesota Center for Economic Re-
Rev., Dec. 1984, 74(5), pp. 861-80. search, 1984.
. "The Hubris Hypothesis of Corporate SINGLETON,KENNETHJ. "Specificationand Estima-
Takeovers,"J. Bus., Apr. 1986, 59(2), pp. 197- tion of Intertemporal Asset Pricing Models," in
216. Handbookof monetaryeconomics.Eds.: B. FRIED-
"R2,"J. Finance, July 1988, 43(2), pp. 541- MAN AND F. HAHN, Amsterdam:North-Holland,
66. 1987.
ROSENBERG, BARR; REID, KENNETH AND LANSTEIN, SLEZAK,STEVEL. "The Effect of Market Interrup-
RONALD. "Persuasive Evidence of Market Ineffi- tions on Equilibrium Asset Return Distributions
ciency," J. Portfol. Manage., Spring 1985, 11(3), in Dynamic Economies with AsymmetricallyIn-
pp. 9-16. formed Traders." Reproduced, U. of California,
Ross, STEPHEN A. "The ArbitrageTheory of Capital San Diego, 1988.
Asset Pricing,"J. Econ. Theory, Dec. 1976, 13(3), SMITH,ADAM.Themoneygame. NY:Random House,
pp. 341-60. 1968.
. "Return, Risk, and Arbitrage,"in Risk and SOLOW,ROBERTM. "TechnicalChange and the Ag-
return infinance, Vol. I. Eds.: IRWIN FRIEND AND gregate Production Function," Rev. Econ. Statist.,
JAMES L. BICKSLER. Cambridge:Lippincott, 1977, Aug. 1957, 39(3), pp. 312-20.
pp. 189-218. STIGLER, GEORGE J. AND BECKER, GARY S. "De Gusti-
. "A Simple Approach to the Valuation of Risky bus Non Est Disputandum," Amer. Econ. Rev.,
Streams," J. Bus., July 1978, 51(3), pp. 453- Mar. 1977, 67(2), pp. 76-90.
75. SuN. MERS, LAWRENCE. "OnEconomicsand Finance,"
. "The Interrelations of Finance and Econom- J. Finance, July 1985, 40(3), pp. 633-35.'
ics: Theoretical Perspectives,"Amer. Econ. Rev., . "Does the Stock Market Rationally Reflect
May 1987, 77(2), pp. 29-34. Fundamental Values?"J. Finance, July 1986, 41(3),
ROZEFF, MICHAEL S. AND KINNEY, WILLIAM R., JR., pp. 591-601.
"Capital Market Seasonality: The Case of Stock THALER, RICHARD H. "Anomalies:The JanuaryEf-
Returns,"J. Finan. Econ., Oct. 1976, 3(4), pp. fect," J. Econ. Perspectives, Summer 1987a, 1(1),
379-402. pp. 197-201.
RUBINSTEIN, MARK. "SecuritiesMarketEfficiencyin . "Anomalies: Seasonal Movements in Secu-
an Arrow-Debreu Economy," Amer. Econ. Rev., rity Prices II: Weekend, Hoiliday, Turn of the
Dec. 1975, 65(5), pp. 812-24. Month, and Intraday Effects,"J. Econ. Perspec-
SALYER, KEVIN D. "Risk Aversion and Stock Price tives, Fall 1987b, 1(2), pp. 169-78.
VolatilityWhen Dividends Are Difference Station- . "Anomalies: The Winner's Curse,"J. Econ.
ary," Econ. Letters, 1988, 28(3), pp. 251-54. Perspectives, Winter 1988, 2(1), pp. 191-202.
SAMUELSON, PAUL A. "ProofThat Properly Antici- THIEL, STUART E. "Some Evidence on the Winner's
pated Prices Fluctuate Randomly,"Ind. Manage. Curse," Amer. Econ. Rev., Dec. 1988, 78(5), pp.
Rev., 1965, 6, pp. 41-49. 884-95.

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions
LeRoy: Efficient Capital Markets and Martingales 1621

TINIC, SEHA M. AND WEST, RICHARD R. "Riskand Price Volatility,"Econometrica,Jan. 1988a, 56(1),
Return:Januaryvs. the Rest of the Year,"J. Finan. pp. 37-61.
Econ., Dec. 1984, 13(4), pp. 561-74. . "Bubbles, Fads and Stock Price Volatility
. "Risk,Return, and Equilibrium:A Revisit," Tests: A Partial Evaluation,"J. Finance, 1988b,
J. Polit. Econ., Feb. 1986, 94(1), pp. 126-47. 43(3), pp. 639-60.
TIROLE, JEAN. "Onthe Possibilityof SpeculationUn- WILLIAMS, J. B. The theory of investment value.
der Rational Expectations,"Econometrica, Sept. Cambridge:HarvardU. Press, 1938.
1982, 50(5), pp. 1163-81. WORKING, HOLBROOK. A Random Difference Series
VARAN, HAL R. "Divergenceof Opinionin Complete for Use in the Analysis of Time Series,"J. Amer.
Markets:A Note," J. Finance, Mar. 1985, 40(1), Statist. Assoc., 1934, 29, pp. 11-24.
pp. 309-17. . "The Investigation of Economic Expecta-
. "Differences of Opinion in Financial Mar- tions," Amer. Econ. Rev., May 1949, 39(3), pp.
kets," in Financial risk: Theory, evidence and im- 150-66.
plications. Ed.: COURTENAY C. STONE. Boston: . "Note on the Correlation of First Differences
Kluwer Academic Publishers, 1989, pp. 3-37. of Averages in a RandomChain," (1960) in COOT-
WEST, KENNETH D. "DividendInnovationsand Stock NER 1964, pp. 129-31.

This content downloaded from 141.233.160.21 on Fri, 06 Nov 2015 04:52:06 UTC
All use subject to JSTOR Terms and Conditions