Information and Competitive Price Systems

Author(s): Sanford J. Grossman and Joseph E. Stiglitz

Source: The American Economic Review , May, 1976, Vol. 66, No. 2, Papers and
Proceedings of the Eighty-eighth Annual Meeting of the American Economic
Association (May, 1976), pp. 246-253
Published by: American Economic Association

Stable URL: https://www.jstor.org/stable/1817229

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 [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

is collaborating with JSTOR to digitize, preserve and extend access to The American Economic
Review

This content downloaded from

201.214.229.228 on Sun, 25 Aug 2024 23:31:51 UTC
All use subject to https://about.jstor.org/terms
 Information and Competitive Price Systems
By SANFORD J. GROSSMAN AND JOSEPH E. STIGLITZ*

Although the price system is conven- variable -q, which can be observed at a
tionally praised as an efficient way of cost, and another, unobservable random
transmitting the information required to variable e:
arrive at a Pareto optimal allocation of re-
(1) r +
sources, the context in which the price
system is usually discussed is not one in where X and e are ,independent, normally
which the informational efficiency of the distributed random variables. Knowing -
price system can be properly evaluated. reduces but does not eliminate the risk as-
Questions of how the price system leads sociated with the asset. The per capita de-
the economy to respond to a new situation, mand, XI, for the asset by those who are
how it conveys information from informed informed of -q will depend both on the price
individuals to uninformed individuals, and of the asset and the value of -.
how it aggregates the different informa-
(2) X = P, 71)
tion of different individuals, are never di-
rectly attacked. We assume that OXI/71 > 0
In a series of papers (Grossman 1975a, Equilibrium each period requires that de-
1975b, Grossman and Stiglitz 1975, and mand equal supply:
Stiglitz 1971, 1974), we have attempted to
remedy this deficiency. It is the object of (3) xXVi(p, X) + (1 - X)XtT(p) - X,
this paper to draw attention to some of the where Xl; is the per capita demand of the
more fundamental implications of our ap- uninformed, Xs is the per capita supply
proach and to use it to assess the meaning and X is the fraction of the individuals who
and validity of the efficient market hy- are informed. Uninformed individuals ob-
pothesis. Although our discussion will ac- serve only price, but from the price they
cordingly focus on the capital market, the may be able to infer -q. For instance, if the
kind of analysis developed here is applica- stock of the resources were fixed, the unin-
ble to any competitive market subject to formed individual can infer that a higher
random shocks. p is associated with a higher -, since an
I. Prices and the Transfer of Information increase in -q increases informed demand,
and thus the price. Since there are no other
The basic idea behind our analysis' may
stochastic elements in this model, there
be illustrated by the following example:
will be precisely one -q corresponding to
Assume there are two assets, one safe
any p. Hence, the conditional distribution
and one risky, and that the return to the
of r given p is the same as the conditional
risky security r, depends on a random
distribution of r given -q. Thus, the price
* This work was supported by National Science
system conveys all the information from
Foundation Grant SOC74-22182 at the Institute for
Mathematical Studies in the Social Sciences (I.MSSS),
the informed individuals to the uninformed.
Stanford University. The authors are also indebted to Now, let us introduce some further ran-
the Dean Witter Foundation for financial support. domness; e.g., in the stock of the risky as-
'See Grossman and Stiglitz (1975) for proofs and
a detailed analysis of the model described by equa-
set or in the demand functions of informed
tions (1)-(4). or uninformed individuals. Then the price
246

This content downloaded from

201.214.229.228 on Sun, 25 Aug 2024 23:31:51 UTC
All use subject to https://about.jstor.org/terms
VOL. 66 NO. 2 INFORMATION AND MARKET STRUCTURE 247

may be high because -q is high, but it may an individual will eventually observe that
be high because the supply of the risky as- the frequency distribution of returns, con-
set is low, or because informed individuals' ditional on the observable variables, is dif-
demand functions have shifted upwards. ferent from the subjective distribution,
Hence, corresponding to any p, there is a and accordingly, ought to revise his ex-
distribution of possible values of -q. The pectations.
price system conveys some information, As there are costs of obtaining informa-
but does not transmit all the information tion, the marginal individual who chooses
from the informed to the uninformed: on to become informed must be indifferent to
average, when the price is high, the return being informed or uninformed, i.e., the in-
is high (i.e., -q and price are correlated) but crement in expected utility from becoming
the price is a noisy signal; that is p and -i informed is exactly offset by the cost of
do not contain the same information the information. In making this calcula-
about r. tion, individuals assume that a change in
Assume that the source of randomness their information (and hence in their de-
is the supply of the risky asset. (We shall mands) would have no effect on prices.
use this example through the rest of the This is an adaptation to this context of
paper.) Then, from (3), the equilibrium the conventional Nash equilibrium hy-
price will depend on -q and the stock of the pothesis of competitive equilibrium theory.
risky asset, Xs; write p = p(Q, Xs). Solve Since when no one is informed, the price
for -q as a function of (p, Xs) as, say, system conveys no information, the value
ti=t(p, Xs). Using (1): of information about -i is likely to be high;
when almost everyone is informed, the
(4) r = t(p, XS) + E.
price system is very informative, so the
The distribution of (Xs, E) induces a dis- value of knowing -q precisely is low. Thus,
tribution on r for a given p. Since the un- provided the costs of information are posi-
informed observe r and p, they come to tive but not too high, equilibrium entails a
learn the conditional distribution of r fraction, X*, of the population being in-
given p. When they observe a p, they use formed-that X which generates a price
this distribution to determine the expected solution to (3) such that the marginal in-
utility from purchasing a given amount of dividual finds the expected utility to being
the risky asset; Xi, is chosen to maximize informed equal to the expected utility of
expected utility. This is how the unin- being uninformed.
formed individual's demand function in Some striking features of the equilibrium
(3) is derived. Finally for this to be an which we have modeled should be noted.
equilibrium, for all -q and Xs, p= p(Qq,First,Xs) it provides a resolution of the follow-
must be a solution to (3). Such an equilib- ing classical conundrum. If markets are
rium entails rational, self-fulfilling expecta- perfectly arbitraged all the time, there are
tions. never any profits to be made from the ac-
This is a reasonable condition for long- tivity of arbitrage. But then, how do
run equilibrium. If this condition is not arbitragers make money, particularly if
satisfied (and the stochastic process de- there are costs associated with obtaining
scribing the returns is stationary),2 then information about whether markets are
2 One can argue that the limitation of our analysis
to stationary stochastic processes is not a serious within a particular event those characteristics which
limitation; economic theory is concerned with identi- it has in common with other events which have
fying, describing, and explaining regularities in eco- occurred. It is these regularities that are described by
nomic processes. Economic theory attempts to identify the stationary stochastic process.

This content downloaded from

201.214.229.228 on Sun, 25 Aug 2024 23:31:51 UTC
All use subject to https://about.jstor.org/terms
248 AMERICAN ECONOMIC ASSOCIATION MAY 1976

already perfectly arbitraged? The conven- formed. The informed traders make prices
tional answer is that, when markets are reflect true values, and the uninformed can
not arbitraged, there are profits to be simply take advantage of these services
made, and so equilibrium must entail per- provided by the informed. In our analysis
fect arbitrage; the profits accrue in the this is not true. Indeed, it is only because
process of responding to some unspecified prices do not accurately represent the true
disequilibrium. A particular example of worth of the securities (i.e., the informa-
this classical conundrum is presented by tion of the informed is not fully conveyed
the efficient market hypothesis, which ar- through the price system, to the unin-
gues the prices on capital markets reflect formed) that the informed are able to earn
all the relevant information instantaneously. a return to compensate them for the costs
We resolve this paradox by arguing that associated with the acquisition of the in-
there are constantly new shocks to the formation.
economy; although each of these shocks Those empirical tests of the weak ver-
may have certain individual character- sion "efficient market hypothesis" which
istics-the company president may be show there are no gains to be made from
sick, a machine may break down-from looking at current prices and the past per-
the point of view of an analysis of market formance of the security provide support
behavior, we are interested not in these for our model, which assumes uninformed
individual characteristics, but in how these traders have rational expectations. But
shocks affect market returns; and we postu- contrary to strong versions of the efficient
late that we can describe the occurrence of market hypothesis, prices do not fully re-
these different shocks, in terms of their ef-flect all available information, in particu-
fects on returns, by a stationary stochastic lar, that of the informed; the informed do
process. The capital market must con- a better job in allocating their portfolio
tinually adjust to these shocks. We have than the uninformed. "Efficient markets"
formulated an equilibrium notion which theorists state that costless information is
explicitly takes account of the economy's a sut/icient condition for prices to fully re-
response to these various shocks. Others flect all available information (Eugene
have described this as a disequilibrium Fama, p. 387). They are not aware that it
situation, but have been unable to say is a necessary condition as well. But this is
much about it. a reductio ad absurdum, since prices are
In the structure we have developed, the important only when information is costly.
market never fully adjusts. Prices never (See Friedrich A. Hayek and Grossman
fully reflect all the information possessed 1975b.) Thus, an individual who throws
by the informed individuals. Capital mar- darts at a dartboard to allocate his port-
kets are not efficient, but the difference is folio will not do as well as the informed in-
just enough, to provide the revenue re- dividual;3 what can be decided by a toss
quired to compensate the informed for of the coin is not the allocation of the port-
purchasing the information. The equilib-
rium fraction of informed traders X* is de- It is still true that if individuals were all identical
and purchased the "market basket" of securities, the
termined jointly with the informativeness
uninformed would do as well as the informed. Here
of the price system in such a way as to we assume that the kind of information to make that
generate a competitive return to arbitrage. feasible is not available. If individuals differ in their
attitudes towards risk, or in their information struc-
Perfect arbitrage has one important im- tures, even when such a strategy is feasible, it may not
plication-not all traders need to be in- be optimal.

This content downloaded from

201.214.229.228 on Sun, 25 Aug 2024 23:31:51 UTC
All use subject to https://about.jstor.org/terms
VOL. 66 NO. 2 INFORMATION AND MARKET STRUCTURE 249

folio but whether to be informed or unin- remarks with more than a hint of disap-
formed. I proval; our analysis suggests that this may
A second important characteristic of our be unwarranted. It may be more efficient
analysis is that there is no proper separa- for some individuals to obtain information
tion between demand and supply. An in- from others-through the price system or
crease in supply leads to a lowering of the by other mechanisms-rather than obtain
price; since lower prices on average corre- it directly.
spond to states in which returns are lower,
II. Prices as Aggregators
the lowering of the price leads to a lowering
of the evaluation of the risky security by So far, we have discussed equilibrium in
the uninformed individuals, and hence of markets where prices convey information
their demand. One cannot describe the from the informed to the uninformed. In
equilibrium meaningfully in any period in some market situations, different individ-
terms of independently drawn demand and uals have different information, and then
supply schedules, because the demand the price system may serve to aggregate
curves depend on the probability distribu- their information. That is, the demands
tion of supply. This has the further conse- for a risky security of an individual are af-
quence that an increase in price may ac- fected by his information; total demand
tually increase demand; the presumption and accordingly equilibrium market prices
for a downward sloping demand curve is thus depend on the information of all the
much weaker when individuals judge qual- individuals. In this sense the market price
ity by price. aggregates the various pieces of informa-
Still a third important and related ob- tion.
servation is that prices, in our model, are A simple example may make this clearer.
serving two functions: not only are they Assume there are a large number of iso-
being used to clear markets in the conven- lated farmers. Each knows the size of his
tional way, but they convey information. own crop, yi. The size of the crop on any
In this sense, the models we have formu- farm at any date is described by
lated are closely related to George Aker-
(5) yi -a+ Ei
lof's lemons' model and to Akerlof (1973)
and Stiglitz' (1975) analysis of labor where (ei, Ej) are uncor
markets. independent, normally distributed random
The discussion so far has focused on the variables with means (&, 0) and variances
decision of whether to be informed or un- (a2, a2), respectively. Thus, if Y= yi,
informed. There is an alternative way of then E(Y| yi) is just a linear function of yi,
looking at this question, which may shed i.e., E(Y yi)=hl+Inh2yi. Assume that
some light on an old question discussed by there is a linear demand curve for the
John M. Keynes (p. 156). He suggested crop, so
that the stock market might be viewed as
(6) Y = a-bP,
a beauty contest, where the participants
are not concerned with judging who is the where P8 is the spot price next period.
most beautiful woman, but with judging Then the subjective distribution of P8 is
who the other judges will believe is the normal, with mean (a-E[YIyi])lb and a
most beautiful woman. Keynes made these
I E [ YI |yI = nf[Yt + (1 - IY)il11l + I2Yi
' This is true only if no one has a comparative a2(n-1)
where -y =
advantage in acquiring information. n(a 2 + a2)

This content downloaded from

201.214.229.228 on Sun, 25 Aug 2024 23:31:51 UTC
All use subject to https://about.jstor.org/terms
250 AMERICAN ECONOMIC ASSOCIATION MAY 1976

variance which is independent of yi, a2 there is no market; but without a market;

Since individuals differ in their expecta- their beliefs will differ. This paradox can
tions, there is an incentive to set up a fu- be put another way. If the market aggre-
tures market. Assume all individuals have gated their information perfectly, individ-
constant absolute risk aversion, k. Then uals' demands would not be based on their
their demand for "futures" Yf is given by own information, but then, how would it
(where Pf is the futures price) :6 be possible for markets to aggregate in-
formation perfectly?
F a-E( Y I y) _ lf So far, we have discussed some of the
basic properties of our approach to equi-
(7) Yf= -= b - |+ yi librium when information is costly. These
models can also be used to address conven-
and the market equilibrium requires tional questions related to existence, com-
parative statics, and welfare.
(8) (8)()=
0O= E y Yi=k--
it{a-hi--h2y~
- b Pf +
III. Existence of Equilibrium Market
Using (6), we obtain the result that the Breakdown and Thinness

futures price is a liniear function of the spot Both Akerlof (1970) and Grossman
price:
(1975a) argue that in markets where prices
Pf = h3 + I-1'Ps convey information between informed and
uninformed traders, there is a possibility
It is a perfect aggregator of the informa- of market breakdown associated with a
tion collected by the different individuals, dwindling in the amount of trading. The
i.e., by observing Pf, one can make a per- example of the stock market presented
fect prediction of the quantity available in above showed that this could indeed hap-
the market and Ps.7 pen: if the price system were fully informa-
But there is a fundamental problem; if, tive, there would be no differences in be-
as one would expect, individuals eventu- liefs; and if there were no differences in
ally come to realize that the futures price beliefs, there would be no trade; but then
is a perfect predictor of the future spot it appears that it is prices in markets in
price, then they will no longer base their which there are no trades which leads to
demands on their own information, but uniformity of beliefs. Although this prob-
rather base it solely on the market in- lem would be alleviated if prices did not
formation. Since the futures price predicts perfectly convey information from the in-
the spot price perfectly (with zero vari- formed to the uninformed or if there were
ance) there is no need for hedging and motivations for trade other than differ-
there will be no trade. But without trade, ences in information (e.g., differences in
attitudes towards risk or in endowments),
Profits are 7 (Pf - P,) + P,yi. Then under
markets still might be thin, i.e., there
normality and constant absolute risk aversion k, the
individual maximizes would be a small volume of trade, and
hence markets may be far from perfectly
Y (Pf - E[Pa)i ] ) + ] (P | yi)yi - 2(yi Yi) arbitraged.
Solving for the optimal iF'
Situations where markets might be thin
or nonexistent need to be distinguished
f EI[PSyjI -Pf from those in which equilibrium does not
I i == Yi + kaIt exist. In the absence of noise, with costly
If 15# nk 2lbea,,, then 11X #O.
information, an (Nash) equilibrium does

This content downloaded from

201.214.229.228 on Sun, 25 Aug 2024 23:31:51 UTC
All use subject to https://about.jstor.org/terms
VOL. 66 NO. 2 INFORMATION AND MARKET STRUCTURE 251

not exist,8 since when one is informed, but not social returns, gains that some in-
every individual believes he can become dividuals make at the expense of others;
informed, increase his expected utility and on the other hand, since some information
not affect the market price. However, is conveyed by the price system, if that in-
when a positive fraction of the population formation is socially useful, those who pur-
becomes informed the price system is fully chase information generate a positive ex-
informative, so it does not pay anyone to ternality to those who do not. See Jerry
purchase the information.9 R. Green (1973) and Stiglitz (1971). Even
if there were no differential information,
IV. Welfare
the price distribution does depend on the
The evaluation of the efficiency of the state of information. To return to our ex-
market in situations such as those analyzed ample of Section I, since when everyone is
in this paper is a subtle and difficult ques- fully informed, price varies with - and XS,
tion. It is not obvious what the appropri- while when no one is informed, price varies
ate comparisons ought to be. Two alterna- only with Xs, it would not be surprising if
tive approaches might be delineated. In information increases the variance of
the reformist approach, we take as given prices. Increased price variability is likely
the market structure, including the mecha- to lead to increased uncertainty about the
nisms for information transmittal. We ask value of one's endowments, and this is
simply, are there too many or too few in- likely to lower expected utility. In one
formed individuals, or, is it desirable to example we have analyzed in detail, where
have an information tax or subsidy? Al- individuals have constant absolute risk
though it is easy to show that the market aversion utility functions and randomly
solution is not, in general, efficient, it is assigned endowments (all individuals hav-
difficult to ascertain whether there is too ing, however, the same endowment dis-
little or too much information acquisition. tribution), every one is better off if no one
There are several effects, operating in dif- is informed than if all are informed.'0
ferent directions: some of the gains arising Finally, if the return to holding an asset
from ditgerential information are private for a period is the dividend plus the capital
gain, the increased variability in price of
In the case where information is costless, an equi- the risky asset makes the risky asset
librium exists; among the set of prices which might
riskier; thus, while in general, information
clear the market, there is a particular price function
which clears it at zero trade and conveys all the rele- reduces the riskiness of a risky asset, this
vant information, and this may be considered to be is at least partially offset by this general
an equilibrium. There is no obvious mechanism for
equilibrium effect.
sustaining this particular set of prices, and this is a
serious limitation. More fundamental questions are raised
" There could not exist an equilibrium in which by the choice of alternative approaches to
trade occurred even if an individual had a monopoly
power over information. For then the uninformed
individuals would observe that they would do better in This is a consequence of the unavailability of
not trading with the monopolist than trading with him, endowment insurance. This result has some important
and the information-monopolist would simply deter- implications for a question which until now has not
mine equilibrium market prices. (See Stiglitz 1974.) been satisfactorily resolved: Can there be destabilizing
Thus Jack Hirshleifer's classic analysis is not that of speculation ? In this context, we interpret that to
a competitive stock market with rational consumers. mean: Can the attempt to engage in intertemporal
If his analysis refers to a market in which there is a arbitrage lead to higher price variability which is
monopolist in information, his results require irration- associated with lower utility? The answer is yes, and
ality on the part of other consumers in the market. indeed such attempts at intertemporal arbitrage can
If his analysis refers to a market in which the market lower welfare. This occurs, in our constant absolute
for information-acquisition is competitive, then the risk aversion example, because by the portfolio separa-
results discussed in the text apply. tion theorem, information has no allocative role.

This content downloaded from

201.214.229.228 on Sun, 25 Aug 2024 23:31:51 UTC
All use subject to https://about.jstor.org/terms
252 AMERICAN ECONOMIC ASSOCIATION MAY 1976

information acquisition, e.g., a comparison bitrage process which will, of necessity, be

between the decentralized process of the imperfect because of the costs of arbitrage
capital market and a centralized process. as discussed in this paper, and (2) econ-
This, in some sense, was the central ques- omies where prices and hence allocations
tion of the Lange-Lerner-Taylor-Hayek are the outcome of a centralized allocative
debate. mechanism which will, of necessity, be im-
Although this earlier debate was pre- perfect because of the costs of monitoring
sumably about the informational efficiency bureaucrats.
of alternative organizational structures, Thus, although we cannot provide an
models in which the systems had to adjust answer to whether a centralized or decen-
to new information were not formulated; tralized organization is more efficient,
rather it was argued that if the informa- without more knowledge of the costs of
tion were to be the same, the allocation operating a centralized informational mech-
would be the same, and thus, a comparison anism, what we have established is that
of alternative organizations came down to the conventional formulations of this ques-
issues like a comparison of cost differen- tion are misleading if not incorrect.
tials arising from different patterns of in-
formation flows, or different speeds of con-
REFERENCES
vergence. Our analysis has suggested that
a decentralized economy is likely to be G. Akerlof, "The Market for 'Lemons':
Qualitative Uncertainty and the Market
characterized by individuals having di/ger-
ential information, that the separation in Mechanism," Quart. J. Econ., Aug. 1970,
the earlier discussion of information and 89, 488-500.
, "A Theory of Information and Labor
allocative questions is inappropriate, and
Markets," mimeo, Univ. of California,
that alternative informational structures
Berkeley, presented at the National Science
will be characterized by different real allo-
Foundation-National Bureau of Economic
cations. In particular, Grossman (1975b)
Research Conference on the Economics of
formalized Hayek's contention that prices Information, Princeton 1973.
are aggregators of information. There it E. Fama, "Efficient Capital Markets: A Re-
was proved that if prices are sufficient view of Theory and Empirical Work," J.
statistics, the competitive economy where Fin., 25, 383-417.
traders have diverse information generates J. R. Green, "Information, Efficiency and
allocations that cannot be improved upon Equilibrium," disc. pap. no. 284, Harvard
by a central planner with all the informa- Institute of Economic Research, Mar. 1973.
tion. However such markets do not provide S. Grossman, "The Existence of Futures
incentives for information acquisition for Markets, Noisy Rational Expectations, and
Informational Externalities," technical re-
the reasons given earlier. Thus only mar-
port no. 182, IMSSS, Stanford Univ., Sept.
kets with noise will exist in equilibrium and
1975a.
these markets will not produce prices
"On the Efficiency of Competitive
which are perfect aggregators. In this case
Stock Markets where Traders have Diverse
a central planner with all the information
Information," technical report no. 183,
can improve on the competitive equilib- IMSSS, Stanford Univ., Sept. 1975b, J.
rium. Thus in our view the Lange-Lerner- Fin., forthcoming, May 1976.
Taylor-Hayek debate comes down to the and J. E. Stiglitz, "On the Impos-
fundamental distinction between econ- sibility of Informationally Perfect Markets,"
omies where: (1) prices and hence alloca- paper presented to Dallas meetings of the
tions are the outcome of a competitive ar- Econometric Society, Dec. 1975.

This content downloaded from

201.214.229.228 on Sun, 25 Aug 2024 23:31:51 UTC
All use subject to https://about.jstor.org/terms
VOL. 66 NO. 2 INFORMATION AND MARKET STRUCTURE 253

F. A. Hayek, "The Use of Knowledge in J. Stiglitz, "Perfect and Imperfect Capital

Society," Amer. Econ. Rev., Sept. 1945, Markets," paper presented to New Orleans
35, 519-30. meeting of the Econometric Society, Dec.
J. Hirshleifer, "The Private and Social Value 1971.
of Information and the Reward to Incentive - "Information and Capital Markets,"
Activity," Amer. Econ. Rev., Sept. 1975, mimeo, Oxford Univ. 1974.
61, 562-74. , "Markets for Heterogeneous Labor
J. Keynes, The General Theory of Employ- with Imperfect Information," mimeo, Stan-
ment, Interest and Money, New York 1964. ford Univ. 1975.

This content downloaded from

201.214.229.228 on Sun, 25 Aug 2024 23:31:51 UTC
All use subject to https://about.jstor.org/terms