While my early research focussed on the implications of the
consumption-based model, I always remembered the advice of my
dissertation advisor, Gene Fama, "you must hedge your research portfolio
especially with respect to this model." This particular research
line ended with my JF92
with Wayne Ferson. In this paper, we explored numerous twists to
the basic model and were unable to salvage the standard framework
to explain both the time-series and cross-section of expected returns.
While I believe that this is a good paper, its impact is limited by
the move in the profession away from the consumption model. Nevertheless,
I learned a tremendous amount from this research program (and in
particular from Wayne Ferson) that led to some other asset pricing
frameworks.
3. Asset Pricing Frameworks
3.1 Domestic
My JFE89, "Time-Varying Conditional Covariances in Asset Pricing Tests"
draws heavily from my joint work with Ferson. In trying to salvage a
linear consumption model, we tried many different formulations of
the risk function. In my JFE89, I apply some of these formulations
(and develop new ones) to versions of the Sharpe-Lintner asset pricing
model as well as alternative multifactor models. The advantage of
the formulation is that it provides for general covariance dynamics without
explicitly parameterizing these dynamics. As such, the instrumental
variables approach provided an alternative to other approaches (in
particular to the GARCH family of models).
Ferson and I jointly studied an approach that incorporated time-varying
risk and returns in our JPE91.
The genesis of this paper was the controversy regarding the attribution
of predictability in asset returns: was it due to rational asset
pricing implications or was it due to market imperfections and/or irrational
behavior on the part of investors? We noted that most of the
asset pricing tests had been conducted within the paradigm of
constant expected returns and risks. Our contribution was to study
a multifactor model that allowed for general time-variation in asset returns
and risk. We detailed an economic interpretation of the changes in economic
risk premiums through the business cycle. We measured the "statistical
predictability" of asset returns (i.e., from projecting asset returns
on some, perhaps data-mined, set of predictors) and compared this
predictability with that implied by asset pricing specification. We found
that 85% of the predictability could be attributed to the asset pricing
dynamics. While we could not eliminate the possibility of market
imperfections causing some predictability, I think that the strength
of our results tilted the balance of opinion toward rational asset
pricing interpretations.
My final examination of domestic asset pricing question incorporates
investigators' prior beliefs into the problem ( with Guofu Zhou in JFE90). This paper examines
unconditional asset pricing restrictions.
However, the twist is that information (in the form of the investigator's
prior belief) are incorporated into the analysis. The paper proposes
a methodology for a full Bayesian analysis of asset pricing models.
We use Monte Carlo integration to evaluate complicated posterior densities.
This is a methodology paper. While Bayesian approaches are very appealing,
there has been a hesitation to apply these methodologies because of
the intense requirement of computational resources. I envision future
researchers using our ideas as Bayesian analysis becomes more
common place.
Ferson and I provided a comprehensive examination of expected international
returns in our RFS93. We addressed two issues. First, we abandoned the
multistep
estimation approach that we used in our JPE91. Second, we presented
a multifactor framework which generalized the single factor results
of my JF91. We found that the multifactor model could account
for most of the predictable variation in asset returns. We found that
most of the variation was being driven by a single world factor
(consistent with my JF91). However, we also found an important,
but smaller role for a foreign currency factor.
Next, I moved to emerging capital markets. These high profile markets had
little academic attention mainly because of the costs associated
with obtaining the data. In return for presenting a paper at a World
Bank conference, the International Finance Corporation gave me their
data. My "Predictable Risk and Returns in Emerging Markets" RFS95
provides a comprehensive analysis of these data. Within the framework
of a world asset pricing model, I present sharp rejections. In
particular, in terms of the usual risk measures, these markets have
essentially zero risk. As a result, their returns are "too high"
to be consistent with the world asset pricing models. The models also
fail to account for predictability of these returns.
The RFS95 explored a world asset pricing model applied to emerging
equity markets. That is, the risk was measured with respect to world
factor benchmarks. This model assumes that each capital market is
perfectly integrated into world capital markets. But this is one of
three possible approaches to the problem. The second involved the
assumption that the capital market was completely segmented (then
risk is measured with respect to local factors). The third is
the assumption of partial integration. Those that implemented
this third approach fixed the degree of integration through time
(constant influence of world and local factors).
This third model provided the origins of my collaboration with Geert
Bekaert. At the World Bank conference, we talked about generalizing
the partial integration framework to allow for the possibility that
the degree of integration changes through time. This squared with our
intuition that some countries had become more integrated through time
(like Thailand) and other countries less integrated (like Zimbabwe).
We provided the first attempt to capture time-variation market
integration in . Of course, in a general model of time-varying integration, there
may be demand to hedge changes in integration over time. This is
not modeled in our framework and we make the reader aware of this
limitation. Nevertheless, our paper presents a number of interesting
results. In addition, we provide an alternate characterization of our
work as tracing the influence of world vs. local factors on time-varying
expected returns.
Another limitation of the JF95 is that the influence of the world and
local factors through the variance is not modeled. Bekaert and
I recently realized that given the persistence in variances and covariances,
it might be more fruitful to examine a similar framework with respect
to variances and covariances. In (working paper), we present a world factor model that allows
for changing
influences of world vs. local factors on both the mean returns and
the variance dynamics. This paper also presents an attempt to understand
why variances differ across countries and why variances shift through time.
The second project is with John Graham. In we study the asset allocation recommendation of 237
newsletter
strategies. The data are newsletters' suggested investment proportions
in equity and cash (Treasury bills or CDs). We provide the first analysis
of the performance of these newsletters and whether the letters have
any ability to "time the market" (increase equity weight before the
market rises and decrease weight before market declines). Consistent with
my other research, we try to control for market wide time-varying expected
returns and risks.
The third project is with Ravi Bansal,
. We cast the challenging performance evaluation problem in the
context of Hansen and Richard
(1987) and Hansen and Jagannathan (1991). We propose an alternative
method of assigning risk to strategies. We detail a trading
strategy which uses three benchmark assets which have futures
contracts trading on them. We show that this strategy, which is based
on publicly available information, "beats the market" when the
traditional performance evaluation criteria are used. This paper
is chocked full of interesting results which hopefully will affect
the practice of performance measurement in the future.
The fourth project is with Arman Glodjo,
My collaboration with Glodjo began when he was a student in the
Computer Science department. I quickly realized that some of the
techniques used for data compression and caching problems were
ideal for the study of time-varying expected returns at the high
frequency level. In this paper, we present a remarkably simple
framework to detect variation in returns.
The fifth project is with Bruno Solnik and Guofu Zhou, "What Determines Expected International Asset
Returns?"
In contrast to my previous work, this project uses a latent factor
technique to study stock and bond returns. The advantage of this
technique is that you need not specify the risk factors in advance.
We offer a number of new twists. First, we recover the fitted risk
premiums and characterize their time-variation with respect to the
world business cycle. Second, we show that similar forces drive
expected stock and bond returns. Third, we study the role of
exchange rate risk in expected returns.
The sixth project is with Christopher Kirby, "Analytic Tests of Linear Factor Models."
The paper provides a general framework for deriving analytical
test statistics for a wide variety of models. Consistent with
my previous work, among these models are those which allow for
time-varying risk and returns.
The seventh project is with Siddique, "Conditional
skewness in asset pricing tests."
We believe that conditional skewness has long been
overlooked by the mean-variance paradigm. We examine a
model that incorporates skewness and assess its ability to
explain both the time-series and cross-section of expected
returns. Our results suggest that some of recent evidence
that fundamental variables explain the cross-section of expected
returns is due to these variables proxying for conditional skewness.
The eighth project is my working paper with Bekaert, "Emerging Equity Market Volatility,"
which I have already detailed.
I also have an exciting project in the early stages with Bernard Dumas.
We study predictability and time-varying risk and try to relate
both to fundamental economic conditions within each country. While
the usual approach is to specify financial variables which proxy
for time-varying expected returns, we develop economic leading and
coincident indicators for a number of countries. We extract both
local business cycles and the world business cycle and use this
information to try to understand asset dynamics.
In a way, my research has come full circle. I started with an
economic model where implications about economic activity were
recovered from expected returns. I then used models relying exclusively
on financial information to understand variation in expected returns.
I am now back to the stage where I am now trying to understand the
linkages between the financial markets and the real economy. While
I might be back where I started, I have learned much on my journey.
3.2 International
I used the framework of my JFE89 to study expected international returns
and risk in "The World Price of Covariance
Risk,"
published in JF91. I was very nervous about my initial submission to
the JF because the framework of the paper appeared too similar to my
JFE89. However, this paper was one of the first attempts to study
international asset prices in a conditional (time-varying returns and
risk) framework. Using data through mid 1989, I found that a
single factor model did a reasonable job of accounting for the
time-series and cross-section of expected returns with the
exception of one country - Japan. In that country, the expected returns
were too high given the risk.
4. Other markets
My research theme has been extended to other markets as well. In my work
with Roger Huang, we study time-varying RFS91. We also have a working paper which
studies the volatility impact of
the size and type of Federal Reserve open market operation
(see "Market
Volatility
Prediction and the Efficiency of the S&P 100 Index Option Market" JFE92
and ,
we study the implications of time-varying volatility in the
options market. We present a model which forecasts implied volatility of
the S&P 100 index option and develop trading strategies to assess whether
the predictability leads to arbitrage profits. 5. Research in progress
I have eight projects in process that are targeted for top journals.
The first is with Ferson,
The goal of this paper is to provide a unified framework for assessing
risk and asset selection in a global context. We provide a new framework
for evaluating country risk. Our results also provide a possible
interpretation to the recent domestic evidence that fundamental
variables (like price-to-book ratios) impact the cross-section of
expected returns. Our framework provides a theoretically consistent way
for fundamental information to affect expected returns through the
risk function.
6. On the horizon
Most of my effort is being spent on either revising working papers or
getting papers ready for submission. There is one project in which
I would like to mention. "Financial Market Integration and Economic
Development" with Geert Bekaert is an attempt to bridge finance and
economic development. It would be reasonable to ask the question
after reading our emerging markets work: What are the economic
implications of having an integrated or segmented capital market? This
paper attempts to assess the implications. Drawing heavily on the
tools we developed in our JF95 and the working paper, we try to
determine what determines both the time-series and cross-section of
real economic growth.
