Information Theory and Market Behavior

Jing Chen
School of Business
University of Northern British Columbia
Prince George, BC
Canada V2N 4Z9
Phone: 1-250-960-6480
Fax: 1-250-960-5544
Email: [email protected]
Web: http://web.unbc.ca/~chenj/

We thank David Faulkner, Frank Schmid, Juan Wu and participants of MFA and FMA
conferences for helpful comments.

Abstract

Some recent empirical works indicate that investor performance and market patterns are
primarily information driven instead of a behavioral phenomenon. However, Grossman
and Stiglitz information theory and its variations offer little guidance in identifying
informed investors and in distinguishing between securities with scarce information and
those with widely available information. We show that most empirical evidences about
market behaviors documented in the literature can be explained by a new information
theory generalized from Shannons entropy theory of information. Investor performance
and market patterns are the results of information processing by investors of different
sizes with different background knowledge.
Keywords: information, entropy, market patterns, behavioral finance
JEL classification: G14

There is a growing consensus that security markets are not as efficient as we thought
before. The inefficiency is generally attributed to behavioral biases of investors. Since
many behavioral biases have been documented in the cognitive psychology literature,
almost any patterns in the financial markets can be linked to one or several of these
biases. However, the potentially boundless set of psychological biases that theorists can
use to build behavioral models and explain observed phenomena creates the potential for
theory dredging. (Chan, Frankel and Kothari, 2002) Furthermore, many theories, while
consistent with empirical patterns that they are set out to explain, are not consistent with
other empirical results. For example, while Bloomfield and Hales (2002) find evidence
supportive of behavioral model of Barberis, Shleifer, and Vishny (1998) in a laboratory
setting, Durham, Hertzel and Martin (2005) find scant evidence that investors behave in
accordance with the model using market data. The link between behavioral theory and
investment behavior are often vague. For example, empirical works reveal that small
investors trading activities often hurt their investment return (Hvidkjaer, 2001). This is
usually thought that small investors are more prone to behavioral biases than
professionals, who are better trained (Shanthikumar, 2004). Yet some empirical work
suggests that professionals exhibit some behavioral biases to a greater extent than nonprofessionals (Haigh and List, 2005).
Behavioral models often assume some market sectors or activities are irrational while
other parts are rational. However, they often differ on which part should be considered
irrational. For example, Shleifer and Vishny (2003) develop a new model on corporate
mergers. Mergers in this model are a form of arbitrage by rational managers operating in
inefficient markets. This theory is in a way the opposite of Rolls (1986) hubris
hypothesis of corporate takeovers, in which financial markets are rational, but corporate
mangers are not. In our theory, managers rationally respond to less-than rational markets
(Shleifer and Vishny, 2003, p. 297).
In a recent work, Ivkovic and Weisbenner (2005) questioned whether investor
performance should be interpreted primarily as a behavioral phenomenon or informationdriven. They conclude, at least in their sample, that information is the primary reason of
investment performance. Empirical results also suggest that superior performance of
investors trading local securities and geographically proximate analysts is due to an
information advantage over others (Coval and Moskowitz, 2001; Malloy, 2005).
Kacpercyzk, Sialm and Zheng (2005) documented that, mutual funds whose portfolios
are highly concentrated in industries where the fund managers have informational
advantages perform better than mutual funds with well-diversified portfolios. The
empirical evidences suggest we should explore an information based theory of market
behavior. However, theory offers little guidance in identifying informed investors and in
distinguishing between securities with scarce information and those with widely available
information (Coval and Moskowitz, 2001).
The standard economic theory of information was developed by Grossman and Stiglitz
(1980). This theory is based on rational expectation. It assumes that investors can
accurately assess the value of some information and pay some fixed amount accordingly

to obtain the information. Recently, various models relax the rational expectation
assumption to explain major market patterns. Most of these models rely on some kind of
human psychological biases. The flourishing of behavioral models and interpretations
suggests that a more fundamental theory is needed.
The basic problem of Grossman and Stiglitz theory is that it does not actually model how
human beings process information. More than fifty years ago, Shannon (1948) developed
the entropy theory of information. The major advantage of Shannons theory is that it can
determine how much information a communication system can receive from the source of
information. Recently, a new theory of information was developed to expand Shannons
entropy theory of information into an economic theory (Chen, 2003, 2005). The basic
idea of this theory is to assert that information is the reduction of entropy, not only in a
mathematical sense, as in Shannons theory, but also in a physical sense. The physical
cost of information is highly correlated with economic cost of information. The rules of
information transmission developed in Shannons theory, as mathematical rules, apply
not only to communication systems, but also to all living organisms, including human
beings. In this work, we will show how this new information theory can offer guidance
in identifying informed investors and in distinguishing between securities with scarce
information and those with widely available information and provides a unified
understanding of observed market behaviors documented in increasingly abundant
empirical works.
Several important properties can be derived from this new information theory. First,
information with higher value is in general more costly. This is a direct extension from
Maxwells (1871) thought experiment on an intelligent demon. Second, the amount of
information one can receive is the amount of information generated minus equivocation.
The level of equivocation, which is the measure of information asymmetry, is determined
by the correlation between the source of information and the receiver of information. In
general, how much information one can receive depends on the background knowledge
of a person. Therefore the process of understanding information is a process of learning,
which often takes long time. This result is a direct extension from Shannons theory from
technical communication systems to human cognitive systems. Third, the value of
information is inversely related to the number of people who understand it. For example,
an investor who buys the shares of a company before it becomes popular often earns
higher rate of return than those who buy the shares of the same company when it
becomes hot. Buffet (2001) once commented, What the few bought for the right reason
in 1925, the many bought for the wrong reason in 1929. At the same time, the value of a
companys investment is also affected by how much its competitors understand the
technology and market potential of a product.
As the efficient market theory retreats, many new theories have emerged and more new
theories will emerge to fill the void left behind. With so many new theories around, it is
time to consider criteria for a good theory.
First, a good theory should be consistent with empirical evidences. Not so long ago,
empirical works on market behavior were so few that it was very difficult to distinguish

the validity of different theories. In the past several years, however, we witness the
emergence of a growing number of empirical works, which often call into the question of
the ability of existing behavioral theories in explaining broader sets of empirical patterns.
In this work, we will list these empirical results and show how the new information
theory can provide a simple and unified understanding of the empirical patterns. This
theory will resolve some puzzles about the market patterns raised in the recent literature
that could not be answered by existing theoretical frameworks.
Second, a good theory should provide more precise predictive power than the existing
theories. Kahneman and Tverskys prospect theory is an improvement over Simons
bounded rationality theory because it offered some concrete patterns of irrationality
beyond the general statement of bounded rationality. Empirical evidences indicate that
the patterns of trading by small, individual investors differ systematically from that by
large, institutional investors. Yet the existing behavioral theories do not offer any
particular behavioral explanations to this systematic difference (Hvidkjaer, 2001;
Shanthikumar, 2004). The new information theory, however, provide very precise
understanding of the systematic differences between trading patterns of large and small
investors. This information theory states that more valuable information is more costly to
obtain in general. For large investors, it pays to spend a lot of effort and money to
research the fundamentals. For small investors, it doesnt pay to dig into the
fundamentals. They depend on processed and easy to understand information that is
readily available at low cost, such as news from popular media and price movement of
the shares. In this work, we will show that patterns of returns by small and large investors
and patterns of return and trading volumes of stocks can be easily understood from this
information processing perspective.
Third, a good theory should help provide deeper understanding to existing models. For
example, Hong and Steins (1999) results are built on three key assumptions. The first
two assumptions are that traders are classified as newswatchers and momentum
traders according to their information processing abilities. They commented that, the
constraints that we put on traders information-processing abilities are arguably not as
well-motivated by the experimental psychology literature as the biases in Barberis et al.
(1998) or Daniel et al. (1998), and so may appear to be more ad hoc (Hong and Stein,
1999, p. 2145). These assumptions can actually be derived naturally from this new
information theory. Depending on the value of assets under management, different
investors will choose different methods of information gathering with different costs.
Newswatchers are large investors who are willing to pay a high cost to collect private
information and to make a deep understanding of public information. Momentum
traders are investors who spend less cost or effort on information gathering and rely
mainly on easy to understand low cost information such as coverage from popular media
and price momentum signals. Cohen, Gompers and Vuolteenaho (2002) empirically
confirm that institutional investors buy on fundamental news while individual investors
buy on price trends. The third assumption of Hong and Stein (1999) is that private
information diffuses gradually across the newswatcher population. The gradual diffusion
of private information means that people gradually learn about the background
knowledge of information and understand information better over time. This new

information theory provides clear understanding to all three assumptions in Hong and
Steins model.
Empirical evidences show that patterns in the security markets are mainly driven by
patterns of information processing by the investment public. However, this does not mean
that human emotions do not influence market behaviors. It has been shown that most
psychological patterns either reflect the constraints of physical laws or evolutionary
adaptation to process information more efficiently in our evolutionary past (Chen, 2003,
2005). In a word, human emotions are really low cost, but not necessarily unbiased ways
of information processing. By showing that most patterns in asset markets documented in
the literature can be explained by this information theory, the focus of attention can be
directed to search the links between human biases and the phenomena or magnitude of
phenomena in asset markets that could not be explained by the information theory.
Furthermore, human activities, including mental activities, are constrained by physical
laws. These constraints offer initial tests to the plausibility of assumptions in behavioral
theories. For example, from the information theory, new information can only be
understood gradually by human beings. If a behavioral theory suggests investors will
generally overact to new information, we need to examine the empirical evidence with
great caution.
This paper is an update from earlier works Chen (2003, 2004, 2005). The remainder of
the paper is structured as follows. Section I presents an updated version of the
generalized entropy theory of information. This information theory provides natural
measures of the cost of obtaining information and of information asymmetry. Section II
explains how patterns of investor returns and patterns of security returns and trading
volumes are natural results of information processing by the heterogeneous investment
public. It also answers many questions on the gaps between existing behavioral theories
and empirical evidences. Section III discusses how this information theory based model
of investor behavior is related to other models of behavioral finance. Many assumptions
in some of the recent theoretical models can be derived naturally from the entropy theory
of information. Section IV concludes.

I. Generalized entropy theory of information

The value of information is a function of probability and must satisfy the following
properties:
(a) The information value of two events is higher than the value of each of them.
(b) If two events are independent, the information value of the two events will be the
sum of the two.
(c) The information value of any event is non-negative.
The only mathematical functions that satisfy all the above properties are of the form
H ( P) = logb P

(1)
5

where H is the value of information, P is the probability associated with a given event
and b is a positive constant (Applebaum, 1996). Formula (1) represents the level of
uncertainty. When a signal is received, there is a reduction of uncertainty, which is
information.
Suppose a random event, X, has n discrete states, x1, x2, ,xn, each with probability p1, p2,
,pn. The information value of X is the average of information value of each state, that is
n

H ( X ) = p j log( p j )

(2)

j =1

The right hand side of (2), which is the entropy function first introduced by Boltzmann in
the 1870s, is also the general formula for information (Shannon, 1948).
After the entropy theory of information was developed in 1948, its technique has been
applied to many different problems in economic and finance. (Theil, 1967; Maasoumi
and Racine, 2002 and many others) However, the standard economic theory of
information, represented by Grossman and Stiglitz (1980) was not built on the foundation
of entropy theory. An entropy theory based economic theory of information can be
simply stated as:
Information is the reduction of entropy, not only in a mathematical sense, as in
Shannons theory, but also in a physical sense. The rules of information transmission
developed in Shannons theory, as mathematical rules, apply not only to
communication systems, but also to all living organisms.
In the following, we will discuss some distinct properties of this new information theory.
First, information that is more valuable is in general more expensive to obtain. From the
second law of thermodynamics, Maxwell concluded that information of higher value is of
higher physical cost. Since economic cost is highly correlated to physical cost,
(Georgescu-Roegen, 1971; Chen, 2005) more valuable information is in general more
expensive to obtain.
Second, the amount of information one can receive depends on the persons background
knowledge about that particular information. The most important result from Shannons
entropy theory of information is the following formula
R = H ( x) H y ( x)

(3)

where R is the amount of information one can receive, H is the amount of information a
source sent and Hy(x), the conditional entropy, is called equivocation. Formula (3) shows
that the amount of information one can receive would be equal to the amount of
information sent minus the average rate of conditional entropy. Before Shannons theory,
it was impossible to accurately assess how much information one can receive from an

information source. In communication theory, this formula is used to discuss how noises
affect the efficiency of information transmission. But it can be understood from more
general perspective. The level of conditional entropy Hy(x) is determined by the
correlation between senders and receivers. When x and y are independent, Hy(x) = H(x)
and R = 0. No information can be transmitted between two objects that are independent of
each other. When the correlation of x and y is equal to one, Hy(x) = 0. No information loss
occurs in transmission. In general, the amount of information one can receive from the
source depends on the correlation between the two. The higher the correlation between
the source and receiver, the more information can be transmitted.
The above discussion does not depend on the specific characteristics of senders and
receivers of information. So it applies to human beings as well as technical
communication equipments, which are the original focus in information theory in science
and engineering. However, the laws that govern human activities, including mental
activities, are the same physical laws that govern non-living systems.
Hy(x) in Formula (3) offers the quantitative measure of information asymmetry (Akerlof,
1970). Since different people have different background knowledge about the same
information, heterogeneity of opinion occurs naturally. To understand the value of a new
product or new production system may take the investment public several years. To fully
appreciate the scope of some technology change may take several decades. For example,
the economic and social impacts of cars as personal transportation instruments and
computers as personal communication instruments were only gradually realized over the
path of several decades. This is why individual stocks and whole stock markets often
exhibit cycles of return of different lengths. This property is very different from
Grossman-Stiglitz information theory, where economic agents can recognize the value of
information instantly and pay according to its value.
Third, the same information, when known to more people, becomes less valuable. Figure
1 is a graph of (1), where H is a function of P, the probability of any given event. From
Figure 1, value is a decreasing function of probability. In the standard information theory,
P represents the probability that some event will occur. In this theory, P is generalized to
represent the percentage of people or money that is controlled by informed investors.
When P = 1, -log P =0. Thus the value of information that is already known to everyone
is zero. When P approaches zero, -logP approaches infinity. Therefore, the value of
information that is known to few is very high. The following example will illustrate this
point. Figure 2 shows overnight rate of return and trading volume of shares of WestJet, a
Canadian airline, surrounding the announcement of the bankruptcy of Jetsgo, the main
competitor of WestJet. Jetsgo announced bankruptcy at the evening of March 10, 2005. If
one bought stock at March 10, he would have made a return of 40% overnight. Judging
from the trading volume of March 9, some people did buy WestJet stock before
information was released to the public. After the announcement made the information
public, trading volume was very high and the rate of return is near zero. Figure 2 neatly
illustrates the relation between value of some information and the number of people who
know the information.

It is often said that the cost of information has dropped sharply over the years. But at the
same time, the value of the same type of information has dropped sharply as well.
Information of high value is usually carefully guarded and difficult to detect. For
example, Warren Buffett, who has a very successful record for gaining and using
insightful market information, would not announce to the public which stock(s) he is
going to buy or sell. Animals have discovered this long ago. In those cases where animal
signals really are of mutual benefit, they will tend to sink to the level of a conspiratorial
whisper: indeed this may often have happened, the resulting signals being to
inconspicuous for us to have noticed them. If signal strength increases over the
generations this suggests, on the other hand, that there has been increasing resistance on
the side of the receiver. (Dawkins, 1999, p. 59)
Unlike Grossman-Stiglitz information theory, this information theory is a nonequilibrium theory. It does not assume a company possesses some intrinsic value waiting
to be discovered by the investment public. Instead, the process of understanding the value
of a company by the investment public is accompanied by the process of understanding
the technology and market potential by its competitors, which generally reduce the value
of that particular company. Empirical evidences that we will present later support this
statement.
While this theory could be a natural extension of Shannons entropy theory of
information, many have pointed out that Shannon (1956) would have a different view:
Workers in other fields should realize that that the basic results of the subject are
aimed at a very specific direction, a direction that is not necessarily relevant to such
fields as psychology, economics, and other social sciences. Indeed, the hard core of
information theory is essentially, a branch of mathematics, a strictly deductive system.
(Shannon, 1956)
This orthodox view was reaffirmed recently:
The efforts of physicists to link information theory more closely to statistical physics
were less successful. It is true that there are mathematical similarities, and it is true
that cross pollination has occurred over the years. However, the problem areas being
modeled by these theories are very different, so it is likely that the coupling remains
limited.
In the early years after 1948, many people, particularly those in the softer sciences,
were entranced by the hope of using information theory to bring some mathematical
structure into their own fields. In many cases, these people did not realize the extent to
which the definition of information was designed to help the communication engineer
send messages rather than to help people understand the meaning of messages. In
some cases, extreme claims were made about the applicability of information theory,
thus embarrassing serious workers in the field. (Gallager, 2001, p. 2694)

However, the dissonance between entropy function as a mathematical representation of

information and the practical value of information has long puzzled many people and
recent works have shown that our intuitive concept of information coincides with the
mathematical definition of information as entropy (Bergstrom and Lachmann, 2004;
Adami, 2004). In the following we will provide a more formal argument. If some
decision making process is truly important and is needed again and again in life, it is
highly economical that quantitative modules to be developed in the mind to expedite the
process. For example, predators need routinely to assess their distance from the prey, the
geometry of the terrain, the speed differential between itself and the prey, the energy cost
of chasing down its prey, the probability of success of each chase and the amount of
energy it can obtain from prey to determine whether, when and where to initiate a chase.
There are many other sophisticated functions, such as navigation by migrating birds over
long distances, that need sophisticated mathematical capabilities. Many animals need to
make precise calculations of many of these quantitative problems many times in life. To
reduce the cost of estimation, mathematical models must have evolved in their mind so
that many decision making processes are simplified into parameter estimation and
numerical computation. It is highly likely that, if some function is very important for the
survival of the animal, in the process of evolution, this function will be genetically
assimilated. Since all living organisms need to extract low entropy from the environment
to compensate continuous dissipation and entropy is the only mathematical function to
measure scarcity of resources, it is inevitable that information, which we collect for our
survival, is largely about entropy. It is not a mere coincidence that our intuitive concept
of information and the mathematical definition of information as entropy largely overlap.
The intuitive concept is really a simplified evaluation of a mathematical computation.
If Shannons entropy theory of information is purely a mathematical theory with little
connection with the physical laws, it would be a miracle that information defined as
entropy turns out to have the magic property of formula (3) that handles technical
problems in communication so well. However, once mathematical theories are thought to
be a natural part of our evolutionary legacy, it would be natural for entropy theory of
information to possess this property.
Since this new information theory can be applied to much broader fields than Shannons
theory, it may be called the Generalized Entropy Theory of Information. Empirical
evidences show that Grossman-Stiglitz information theory cannot explain observed
investor behaviors (Barber and Odean, 2000). Other models extended from GrossmanStiglitz theory often have little explanatory power outside a limited scope. We will show,
in the next section, that the generalized entropy theory of information will provide a
unified understanding of the empirical market behaviors documented in the literature.

II. Theoretical predictions and empirical evidences

A. Differences in informational advantages of investors of different sizes

From this information theory, information with higher value is in general more costly.
The amount of time and resource an investor will spend depends on the value of her
portfolio. It can be predicted that the rate of return for large, institutional investors will be
higher than small, individual investors because large investors will spend more time and
resources in research. As the market rate of return is the average rate of return of all
investors, we expect large investors outperform market average and small investors
underperform market average. This is supported by empirical evidences. (Wermers,
2000; Barber and Odean, 2000; Cronqvist and Thaler, 2004)
If it is not economical for small investors to spend a lot of effort to study individual
stocks, can they expect superior returns by investing in mutual funds with talented
managers? The information theory predicts that the value generated by the talented
managers will be mainly retained by the managers, who spend time to collect and analyze
information, and not be distributed to mutual fund investors. Wermers (2000)
documented that during the 1977 to 1994 period, mutual funds did generate higher raw
rate of return than the market indices, which indicates researches by mutual fund
managers do uncover valuable information. However, the net rate of return to investors is
an average of 13.3 percent per year, which is the same as the Vanguard fund, the largest
index fund. As the average transaction costs decrease over the years, average expense
ratios increase, making the sum of these two costs remaining relatively constant over the
years. (Wermers, 2000) This shows that it is the mutual fund managers, not the average
investors, who capture the benefit of declining transaction costs. This is consistent with
the model of Berk and Green (2004), where mutual fund managers capture the entire
rents from their performance. In essence, for small investors without special information
advantage or people whose cost of collecting and analyzing information outweighs its
benefit, which are the majority of the population, they cannot expect a higher rate of
return on stocks than the general market. The above result may be called the generalized
efficient market hypothesis: investors without informational advantage can not
outperform the general market.
Odean (1999) documented that the shares individual investors sold outperform the shares
they bought. He attributed this to some unknown biases of investors judgment and
stated, What is more certain is that these investors do have useful information which
they are somehow misinterpreting. (Odean, 1999, p. 1296) The real reason may be a lot
simpler. Chen, Jegadeesh and Wermers (2000) documented that shares bought by mutual
fund managers outperform shares they sold. The differential performance of shares
bought and sold by individual investors is partly due to informational advantages of some
of their trading counterparties. Even if individual investors randomly select buy and sell
orders, most trades that get executed are not in their favor, because some of their
counterparties, such as mutual fund managers, possess valuable information and select to
fill the orders that are in their favor. For example, if an individual investor randomly
select two stocks to sell with limit orders, it is more likely that the stock which will have
a higher rate of return in the future will be bought by more informed investors. Later, we
will demonstrate that small investors dependence on low cost, low value information
does make them prone to bad trading decisions.

10

Barber and Odean (2000) documented that there is very little difference in the gross
performance of households that trade frequently and those that trade infrequently. On
average, shares individual investors sold outperform the shares they bought. (Odean,
1999) If all else are same, more frequently traded investors should earn lower rate of
gross return. So this indicates that more frequently traded investors have better
investment skills than less frequently traded investors. This prediction can be directly
verified from trading records. Barber and Odean cited Carhart (1997) to show that
frequent trading also hurts the performance of mutual funds. However, a more recent and
more detailed study by Wermers (2000) shows that the most traded mutual funds
outperform least traded mutual funds by a wide margin. Barber and Odean attribute
trading activities to overconfidence, a behavioral explanation. It may be simpler and more
consistent with empirical evidences to explain trading as a type of learning activity. Like
all other kinds of learning, trading is costly. The cost of learning is compensated, at least
partially, by knowledge gained from the experience.

B. Differences in informational advantages of investors of different backgrounds

From (3), the ability to understand information depends on the background knowledge of
investors. This indicates that investors will earn higher rate of return if they choose to
invest in securities that they are familiar with. This is supported by some recent empirical
investigations. Professional managers and individual investors local investments
outperform their remote investments. (Coval and Moskwitz, 2001; Ivkovic and
Weisbenner, 2005) Malloy (2005) provides evidence that local analysts are significantly
more accurate than other analysts in forecasting and recommendations. Mutual funds
with high industry concentration, where fund managers can focus on particular industries
they are familiar with, are more successful in selecting securities than diversified funds.
(Kacpercyzk, Sialm and Zheng, 2005) Investors take longer time to understand
information from sources they are less familiar with. Hong, Lim and Stein (2001)
empirically confirm that information from small firms, from firms with low analyst
coverage and from firms with bad news, which managers are reluctant to release,
generally diffuse slower. From Hvidkjaer (2001), the selling pressures on losers generally
are stronger and last much longer than buying pressures on winners, suggesting
information processing is less efficient on bad news.
Since insiders understand information much better than others, they can and will take
advantage of this information asymmetry to pursue certain corporate activities. Shleifer
and Vishny (2003) develop a new model on corporate mergers, in which mergers are a
form of arbitrage by rational managers operating in inefficient markets. They show that
their model has better explanatory power than the behavioral based models and is
supported by new empirical evidences. (Ang and Zheng, 2002) While assuming
managers are rational, they continue to assume financial markets are less-than-rational.
However, form the information theory, there is no need to assume financial markets are
less-than-rational. In the next subsection, we will show how information processing by
investors of different sizes with different background knowledge will generate cycles of
undervaluation and overvaluation.

11

C. Information processing by heterogeneous investors and market patterns

A persistent pattern in the security market is the price continuation in short to medium
run and the reversal of return in the long run. (DeBondt and Thaler, 1985; Jegadeesh and
Titman, 1993) Several models have been developed to explain this pattern. (Barberis,
Shleifer and Vishny,1998; Daniel, Hirshleifer and Subrahmanyam,1998; Hong and
stein,1999) However, these models are based on some ad hoc assumptions and could not
explain other patterns. (Lee and Swaminathan, 2000; Hvidkjaer, 2001) For example, the
return patterns are often accompanied by distinct patterns of trading volume. However,
existing theories of investor behavior do not fully account for all of the evidence.
none of these models incorporate trading volume explicitly and, therefore, they cannot
fully explain why trading volume is able to predict the magnitude and persistence of
future price momentum. (Lee and Swaminathan, 2000, p. 2066)
Lee and Swaminathan (2000) and Hvidkjaer (2001) presented many empirical results
regarding return, trading volume and patterns of investment decisions by traders of
different sizes. Many of these results, especially those in Hvidkjaer (2001), are not
presented in a systematic way because of the lack of a proper theoretical framework. In
the following, we will show how the generalized entropy theory of information offers a
coherent and comprehensive understanding of observed market patterns and the
underlying dynamics. Hvidkjaers work has gone through several significant revisions.
Our discussion will be based on his new version from July, 2004. This part is mainly
adapted from Chen (2004).
To an investor, the choice of information gathering is a matter of cost and the ability to
understand some information depends on her background. More valuable information is
more costly to obtain in general. For large investors, it pays to spend a lot of effort and
money to research the fundamentals. For small investors, it doesnt pay to dig into the
fundamentals. They depend on processed and easy to understand information that is
readily available at low cost, such as news from popular media and price movement of
the shares. Cohen, Gompers and Vuolteenaho (2002) confirmed that institutional
investors buy on fundamental news while individual investors buy on price trends.
Whether an investor will be able to understand certain information also depends on her
particular background, which determines her level of equivocation in receiving that
information. Information processing by investors of different sizes with different
backgrounds will generate the return and trading volume cycle that is similar to Lee and
Swaminathans (2000) momentum life cycle. In the following, we will illustrate the
patterns of return and trading volumes of a stock of a typical company from the
information processing perspective.
Suppose a company develop a new technology, which is expected to bring the company
high profit in the future. From (3), those who are familiar with the technology and
company will have low equivocation in receiving the information. They understand the

12

significance of the information first and buy the company shares. Since they are a small
number of people, the buying is of low volume. This corresponds to the beginning of the
low volume winner stage in momentum life cycle. From table IX of Lee and
Swaminathan (2000), the return on equity does improve over the next three years for low
volume winner, which shows that the investors in this stage do have accurate perception
about the future. From Figure 3 of Hvidkjaer (2001), the buying pressures from both
large trades and small trades in this period trend up gradually, signaling the gradual
diffusion of information. The buying pressure from large trades are higher than the
buying pressure from small trades, which shows that large traders as a group are better
informed than small traders.
As the technology goes through various stages from R&D to production, the potential
becomes clearer to more people. This means that the level of equivocation gradually
reduces to more people, which sustains buying interest and share prices increase
gradually. As the technology becomes adopted in production and profit figures become
public, the level of equivocation decrease further and the pool of investors increases
further. Eventually, both the sustained increase of stock price and stable pattern of profit
increase, which are very easy to understand by the general public, attract large amount of
buyers, which results in high trading volume and pushes the stock prices further up. This
corresponds to the high volume winner stage in momentum life cycle. From Figure 2,3 of
Hvidkjaer (2001), there is a steady and higher buying pressure among large traders than
in low volume winner stage, signaling a consensus of bullish sentiment from informed
investors. Because of this consensus, the return of this stage is extremely high. (Lee and
Swaminathan, 2000, Table IV) From Table IX of Lee and Swaminathan (2000), the
return on equity is very high for high volume winners. However, the high return of the
company will attract the attention of not only investors but also competitors, which will
try to produce same or similar products for this high profit market.
From Figure 1, the value of some information that is known to everyone is zero. As the
good news reaches most investors, the security is probably already fully or over priced.
Among the increased pool of investors, more and more investors understand very little of
the fundamentals behind the technology and depend on easy to understand signals such as
coverage from popular media and stock price movement to make trading decisions. For
this group of investors, they will stop buying only when the opinion of public media
changes and the trend of price increase reverses significantly. As stock price keeps
increasing, momentum trading becomes highly profitable, which will eventually push the
share prices higher than the fundamental value. Since large investors spend more
resources in investment, they are generally better informed than small investors. As share
prices become highly overvalued relative to fundamentals, large investors start to unload
positions while small investors keep buying. As the selling pressure from large investors
becomes greater than the buying pressure from small investors, the trend of price increase
reverses to price decrease. (Hvidkjaer, 2001, Figure 2) This is the period of high volume
losers in momentum life cycle. In the period of high volume loser, as competition
intensifies, return on equity drops sharply from previous years. (Lee and Swaminathan,
2000, Table IX).

13

As the pattern of price drop becomes clear, more and more people joined the selling.
After large investors and some of the small investors have finished unloading the
positions, the volume of trading will decline, which is the period of low volume loser in
momentum life cycle. This period is characterized by active selling of small investor.
(Hvidkjaer, 2001) Since small investors are typically slow to understand information,
their active selling, after the selling by large investors, signals the selling is overextended,
which indicates the low volume losers will rebound and earn high future return in
general. From the operation point of view, this is the worst time for the company.
Overcapacity of a once high profit margin industry pushes down the return on equity
further from the high volume loser stage. There are probably some layoffs of labor and
write off of capital. But the return on equity will gradually regress toward normal level.
(Lee and Swaminathan, 2000, Table IX).
The above paragraphs describe the patterns of information processing and trading when
the initial news is positive. When the news is negative, a similar pattern exists at opposite
directions. Since there is a cost shorting stocks and there are many institutional
constraints on shorting stocks, short selling is much more difficult than buying stocks.
With good news, there are many potential buyers. With a bad news, the sellers are largely
confined to existing share holders. So overreaction is less strong on bad news. The
statistical results, which are the average of all phenomena, mainly reflect the action from
good news instead of bad news. With this observation in mind, we can discuss the
following:
The Hong and Stein (1999) model predicts that momentum profit should be larger for
stocks with slower information diffusion. If we make the assumption that scarcity of
trading leads to insufficient diffusion of information, then the Hong and Stein model
would predict a greater momentum effect among low volume stocks. Our result indicate
this to be true among winners but not among losers. That is, low volume winners have
greater momentum, but low volume losers actually have less momentum. (Lee and
Swaminathan, 2000, p. 2062)
The information theory indicates that low trading volume may reflect either a lack of
understanding of some new information or a lack of information. From the above
discussion, there are two types of low volume losers. The first, which is an average
representative from Lee and Swaminathans statistical results, is part of a cycle that is
triggered by some good news. It has experienced the cycle of low volume winner, high
volume winner, high volume loser and low volume loser. The low volume loser period
represents the end of an information processing cycle. This is why it exhibits less
momentum. The other type is the low volume loser period at the beginning of an
information processing cycle that is triggered by a bad news. (Hong, Lim and Stein,
2001) These low volume losers do exhibit strong momentum in low rate of return. So the
apparent inconsistency of both views can actually be reconciled with a more detailed
analysis from information theory.
The above analysis shows that securities often experience cycles of underreaction and
overreaction as the result of investors information processing. What determine the level

14

of underreaction and overreaction? It depends on how much we understand the

fundamentals. If the fundamentals are easy to understand by many people, both initial
underreaction and eventual overreaction will be small. If the fundamentals are difficult to
understand, mispricing can be substantial. We can have a look at glamour stocks.
Glamour stocks are from companies with high earning growth. This means they have
very few potent competitors, which generally indicates the lack of deep understanding
about the particular products or production systems. That is to say, there is a high level of
information asymmetry between the companies of glamour stocks and the general public.
Initially, these types of companies are underpriced because few people understand them.
However, the solid earning growth of these companies makes the share prices grow
continuously, generating clear technical signals. The clarity of technical signals and
vagueness of fundamental information will eventually cause high level of overreaction.
Statistical results show that stocks undergoing price momentum over longer period will
exhibit higher level of reversal. (Lee and Swaminathan, 2000, Table I) Economy wide,
great speculative bubbles are generally associated with new era or new economy,
when the general public is touched by the profound influence of technology
breakthroughs while having little understanding of the underlying mechanisms. (Shiller,
2000)
In the following, we will answer the three questions posed by Lee and Swaminathan at
the end of their paper.
First, the asymmetry in the timing of momentum reversals between winners and losers
remains a puzzle. We show that low volume losers rebound quickly and outperform high
volume losers with the next three to 12 months. However, it takes low volume winners
longer (more than 12 months) to significantly outperform high volume winner. We know
of no explanation for this timing difference. (Lee and Swaminathan, 2000, p. 2067)
From our analysis of information processing cycle, the low volume winner stage is the
gradually understanding of fundamental news about a firm. Since the understanding of
fundamentals is very costly and generally take very long time, it will take long time for
low volume winners to significantly outperform high volume winners. The high volume
loser stage is when large investors are already well informed about the overpricing and
are active sellers. The price at this stage is supported by active buying of small investors,
which mainly respond to popular media coverage and technical signals. (Hvidkjaer,
2004) Since coverage from popular media and technical signals, which are information
with low cost and low value, are easier to understand than details about fundamentals, the
price adjustment at this stage is much faster.
Second, with the possible exception of the disposition effect from the behavioral
literature, we know of no explanation for why trading volume should decline when firms
fall out of favor.
The volume of trading reflects how many investors believe they can make profitable
trades. When stocks are out of favor, few people believe they can make a profit buying
these stocks. Hvidkjaers detailed analysis shows that losing stocks do experience

15

consistent selling pressures over a long period of time. The low volume of trading when
firms fall out of favor reflects one fundamental asymmetry in security trading. For a
stock, there are always more potential buyers than potential sellers, who are largely
existing share holders.
Finally, we find it remarkable that measures as readily available as past returns and
trading volume can have such strong predictive power for returns. Why this
information is not fully reflected in current prices is another puzzle we leave for future
research.
From (3), how much information we can understand depends on our background
knowledge about the information and how much weight we assign to the information.
From the efficient market theory, trading volume carries very little information. So little
weight was given to the idea that trading volume might contain valuable information,
which inhibited the research on this direction in the past.
D. On equity premium
Mehra and Prescott (1985) observed that the large size of risk premium on US equities
can not be explained by the standard general equilibrium models and called it a puzzle.
Among much research generated by this observation, two approaches are relevant to our
study. One approach attributes the high risk premium to loss aversion by investors.
Barberis and Thaler (2003) provided a survey of works along this line. The other is
survivorship bias proposed by Jorion and Goetzmann (1999).
From the generalized entropy theory of information, how long a pattern persists depends
on the cost of learning. It is often very costly to gain a deep understanding of a company
or an industry, especially when the industry is new. The equity premium puzzle,
however, is a very simple pattern on financial data, which, once discovered, can be
understood very easily by the investment public. The strategy of profiting from the high
equity premium is easy to implement and of low risk. This indicates that the pattern of
high equity premium, if it does exist, is a short term one.
Whether high equity premium is a pattern or a result of selection bias can be answered by
more comprehensive data. Jorion and Goetzmann (1999) documented that among all the
equity markets around the world in the past century, the US market had the highest
return. They argued that US market had the highest return because US was the most
successful economic system in the world in the last century. For 1921 to 1996, U.S.
equities had the highest real return of all countries, at 4.3 percent, versus a median of 0.8
percent for other countries. The high equity premium obtained for U.S. equities appears
to be the exception rather than the rule. (Jorion and Goetzmann,1999, p. 953) Their
conclusion is consistent with the generalized entropy theory of information.

III. The relation with other models of behavioral finance

16

Recently, several behavioral models provide frameworks to interpret the short to

intermediate term momentum and long term reversal of return. In this section, we will
discuss the relation between the generalized entropy theory of information and these
models. Daniel, Hirshleifer and Subrahmanyam (1998) explain momentum in terms of
both initial and delayed overreaction, while Barberis, Shleifer and Vishny (1998) and
Hong and stein (1999) explain momentum in terms of initial underreaction and followed
by delayed overreaction.
From the information theory, the absorbing of a new information is a gradual process, in
which the equivocation gradually reduces. So stock prices generally undereact to new
information initially, which is confirmed by empirical evidences (Hvidkjaer, 2001). This
is consistent with the models of Barberis et al. (1998) and Hong and stein (1999). In the
following, we will make further analysis of these two models.
Barberis et al. (1998) utilize the concept of conservatism to understand underreaction.
Conservatism states that individuals update their beliefs slowly in the face of new
information. This property is a natural result from formula (3). Barberis et al. (1998)
attribute overreaction to representativeness heuristic. People rely on a limited number of
heuristic principles which reduce the complex tasks of assessing probabilities and
predicting values to simpler judgmental operations (Tversky and Kahneman, 1974,
p.1124). Many investors dont want to spend tremendous resource to research
fundamentals. They rely on a limited number of heuristic principles, such as technical
signals and opinions from popular media, which reduce the complex tasks of assessing
probabilities and predicting values of stocks to simpler judgmental operations with low
cost. As we analyzed in the last section, this reliance on simple heuristic principles leads
to overreaction in the asset market.
Hong and Steins (1999) results are built on three key assumptions. The first two
assumptions are that traders are classified as newswatchers and momentum traders
according to their information processing abilities. They commented that, the constraints
that we put on traders information-processing abilities are arguably not as wellmotivated by the experimental psychology literature as the biases in Barberis et al. (1998)
or Daniel et al. (1998), and so may appear to be more ad hoc (Hong and Stein, 1999, p.
2145). These assumptions can actually be derived naturally from the entropy theory of
information. Depending on the value of assets under management, different investors will
choose different methods of information gathering with different costs. Newswatchers
are large investors who are willing to pay a high cost to collect private information and to
make a deep understanding of public information. Momentum traders are investors
who spend less cost or effort on information gathering and rely mainly on easy to
understand low cost information such as coverage from popular media and price
momentum signals. Cohen, Gompers and Vuolteenaho (2002) show that institutional
investors buy on fundamental news while individual investors buy on price trends. The
third assumption of Hong and Stein (1999) is that private information diffuses gradually
across the newswatcher population. The gradual diffusion of private information means

17

that the number of people who enjoy low level of equivocation on some information
gradually increases.
Both the reduction of equivocation of a representative investor and the increase of
number of investors who have low level of equivocation on information contribute to the
gradual reduction of underreaction, which generates momentum. Both representativeness
heuristic and momentum trader can generate overreaction, which will lead to eventual
reversal. The information theory can further distinguish the models of Barberis et al.
(1998) and Hong and stein (1999). In Barberis et al. (1998), a representative investor
make trading decisions. In Hong and stein (1999, 2003), investors are heterogeneous.
Investor heterogeneity can be understood naturally because of the different background
of the investors and different cost that different investors are willing to pay to gather
information. Empirical evidences show that investor heterogeneity exists in financial
markets and plays an important role in the formation of trading patterns (Hvidkjaer,
2001).
From (3), the understanding of information depends on the background knowledge.
Investors take longer time to understand information from sources they are less familiar
with. Hong, Lim and Stein (2001) empirically confirm that information from small firms,
from firms with low analyst coverage and from firms with bad news, which managers are
reluctant to release, generally diffuse slower. From Hvidkjaer (2001), the selling
pressures on loser generally are stronger and last much longer than buying pressures on
winners, suggesting information processing is less efficient on bad news.
After discussing the existing behavioral models, Lee and Swaminathan summarized,
existing theories of investor behavior do not fully account for all of the evidence.
none of these models incorporate trading volume explicitly and, therefore, they cannot
fully explain why trading volume is able to predict the magnitude and persistence of
future price momentum (Lee and Swaminathan, 2000, p. 2066). Trading volume, on the
other hand, is an integral part of the model of investment behavior based on the
generalized entropy theory of information. This model answers many questions on the
gaps between existing theories and empirical evidences.
IV. Conclusion

Efficient market theory states that financial markets can process information instantly.
However, common sense tells us that information of higher value is in general more
costly to obtain and it often takes long time for people to understand the meaning and
significance of some information. The newly developed generalized entropy theory of
information naturally incorporates this common sense. We show that this new
information theory provides a simple and unified understanding of the empirical patterns
documented in the literature. It is hoped that future empirical works will help further
distinguish different theories about market behaviors.

18

References

Adami, Christoph, 2004, Information theory in molecular biology, working paper.

Akerlof, Geroge, 1970, The Market for 'Lemons': Quality Uncertainty and the Market
Mechanism, Quarterly Journal of Economics 84, 488-500.
Applebaum, David, 1996, Probability and Information, an integrated approach,
(Cambridge University Press, Cambridge).
Barber, Brad, and Terrance Odean, 2000, Trading is hazardous to your wealth: The
common stock investment performance of individual investors, Journal of
Finance 55, 773-806
Barberis, Nicholas, Andrei Shleifer, and Robert Vishny, 1998, A model of investor
sentiment, Journal of Financial Economics 49, 307-345.
Barberis, Nicholas and Robert Thaler, 2003, A survey of behavioral finance, in
Handbook of the economics of finance, edited by G. Constantinides, M. Harris
and R. Stulz.
Bennett, Charles H., 1988. Notes on the History of Reversible Computation. IBM Journal
of Research and Development, 32, 16-23.
Bergstrom, Carl and Lachmann, Michael, 2004, Shannon information and biological
fitness, Working paper.
Berk, Jonathan, and Richard Green, 2004, Mutual fund flows and performance in rational
markets, Journal of Political Economy 112, 1269-1296.
Bloomfield, Robert, and Jeffrey Hales, 2002, Predicting the next step of a random walk:
Experimental evidence of regime-shifting beliefs, Journal of Financial
Economics 65, 397-414.
Buffet, Warren, 2001, Warren Buffett on the stock market, Fortune Investors Guide,
with Carol Loomis, Dec 10, 80-94
Carhart, Mark, 1997, On persistence in mutual fund performance, Journal of Finance 52,
57-82.
Chan, Wesley S., Richard M. Frankel and S.P. Kothari, 2002, Testing Behavioral Finance
Theories Using Trends and Sequences in Financial Performance, working paper.
Chen, Jing, 2003, An Entropy Theory of Psychology and its Implication to Behavioral
Finance, Financile Studievereniging Rotterdam Forum, 6, 26- 31
Chen, Jing, 2004, Generalized Entropy Theory of Information and Market Patterns,
Corporate Finance Review, 9, No. 3, 23- 32.
Chen, Jing, 2005, The physical foundation of economics: An analytical thermodynamic
theory, World Scientific, Singapore, forthcoming.
Chen, Hsiu-Lang, Narasimhan Jegadeesh, and Russ Wermers, 2000,The Value of
Active Mutual Fund Management: An Examination of the Stockholdings and
Trades of fund managers, Journal of Financial & Quantitative Analysis 35, 343368.
Cohen, Randolpg, Paul Gompers, and Toumo Vuolteenaho, 2002, Who underreact to
cash-flow news? Evidence from trading between individuals and institutions,
Journal of Financial Economics 66, 409-462.
Coval, Joshua, and Tobias Moskowitz, 2001, The geography of investment: Informed
trading and asset prices, Journal of Political Economy 54, 1-39.

19

Cronqvist, Henrik and Thaler, Richard, 2004, Design choice in privatized social-security
systems: learning from the Swedish experience, American Economic Review, 94,
424-428
Daniel, Kent, David Hirshleifer and Avanidhar Subrahmanyam, 1998, Investment
psychology and Security market under and overreactions, Journal of Finance 53,
1839-1885.
DeBondt, Werner and Richard Thaler, 1985, Does the stock market overreact? Journal of
Finance 40, 793-805.
Durham, Gregory, Michael Hertzel, Michael, and Spencer Martin, 2005, The market
impact of trends and and sequences in performance: new evidence, Journal of
Finance, 60, 2551-2569.
Fama, Eugene, 1998, Market efficiency, long-term returns and behavioral finance,
Journal of Financial Economics 49, 283-307.
Gallager, R. 2001, Claude E. Shannon: A Retrospective on his life, work, and impact,
IEEE Trans. IT, vol 47,#6,
Georgescu-Roegen, Nicholas, 1971. The entropy law and the economic process. Harvard
University Press, Cambridge, Mass.
Grossman, Sanford and Joseph Stiglitz, 1980, On the impossibility of informationally
efficient markets, American Economic Review, 70, 393-408.
Gulko, Les, 2002, The entropy theory of bond option pricing, International Journal of
Theoretical and Applied Finance, 5, 355-383
Haigh, Michael, and John List, 2005, Do professional traders exhibit myopic loss
aversion? An experimental analysis, Journal of Finance, Forthcoming.
Hong, Harrison and Jeremy Stein, 1999, A unified theory of underreaction, momentum
trading and overreaction in asset markets, Journal of Finance 54, 2143-2184.
Hong, Harrison and Jeremy Stein, 2003, Differences of opinion, short-sale constraints
and market crashes, Review of financial studies, 16, 487-525.
Hong, Harrison, Terence Lim and Jeremy Stein, 2000, Bad news travel slowly: Size,
analyst coverage, and the profitability of momentum strategies, Journal of Finance
55, 265-295.
Hvidkjaer, Soreen, 2001, A trade-based analysis of momentum, Working paper.
Ivkovic, Zoran, and Scott Weisbenner, 2005, Local does as local is: Information content
of the geography of individual investors common stock investments, Journal of
Finance, 60, 267-306.
Jegadeesh, Narasimhan and Sheridan Titman, 1993, Returns to buying winners and
selling losers: Implications for stock market efficiency, Journal of Finance 48, 6591.
Jorion, P. and Goetzmann, W. 1999, Global Stock Markets in the Twentieth Century,
Journal of Finance, 54, 953-980.
Kacperczyk, Marcin, Clemens Sialm, and Lu Zheng, 2005, On the industry concentration
of actively managed equity mutual funds, Journal of Finance, 60, 1983-2011.
Lee, Charles and Bhaskaran Swamminathan, 2000, Price momentum and trading volume,
Journal of Finance 55, 2017-2069.
Maasoumi, Esfandiar and Jeff Racine, 2002, Entropy and predictability of stock market
returns, Journal of Econometrics, 107, 291-312.

20

Malloy, Christopher, 2005, The geography of equity analysis, Journal of Finance, 60,
719-755.
Maxwell, James, 1871, Theory of heat, (Green & Co., Longmans, London)
Odean, Terrance, 1999, Do Investors Trade Too Much, American Economic Review 89,
1279-1298.
Roll, Richard, 1986, The hubris hypothesis of corporate takeover, Journal of Business 59,
197-216.
Shannon, Claude, 1948, A mathematical theory of communication. The Bell System
Technical Journal, 27, 379-423, 623-656.
Shannon, Claude, 1956, The Bandwagon (Editorial), IEEE Transactions Information
Theory, volume 2, p.3
Shannon, Claude and Weaver, Warren 1949. The mathematical theory of communication,
Urbana: The University of Illinois Press.
Shanthikumar, Devin, 2004, Small Trader Reactions to Consecutive Earnings Surprises,
working paper.
Shiller, Robert, 2000, Irrational exuberance, Princeton, N.J. : Princeton University
Press.
Shleifer, Andrei, and Robert Vishny, 2003, Stock market driven acquisitions, Journal of
Financial Economics 70, 295-311.
Theil, Henri 1967, Economics and information theory. Amsterdam: North-Holland
Publishing Company.
Tversky, Amos and Daniel Kahneman, 1974, Judgment under uncertainty: Heuristics and
biases, Science 185, 1124-1131.
Wermers, Russ, 2000, Mutual fund performance: An empirical decomposition into stockpicking talent, style, transaction costs, and expenses, Journal of Finance 55,
1655-1703.

21

Figure Captions

Figure 1: Information value and probability

Figure 2: Overnight rate of return and trading volume of WestJet stock
surrounding the date when Jetsgo announced bankruptcy

22

Value

0
0.01

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Probability

23

12000000

45.00%
40.00%

10000000

35.00%
30.00%

8000000
25.00%
20.00%
6000000
15.00%

Volume
Return

10.00%
4000000
5.00%
0.00%

2000000

-5.00%
0

-10.00%
4-Mar

7-Mar

8-Mar

9-Mar

10-Mar 11-Mar 14-Mar 15-Mar 16-Mar

24