M.Phil.,
Department of Economics, Yale University, May 2002
M.A., Department of Economics, Catholic University of Louvain, July 2001, highest honors
M.A., Department of Economics, Yale University, December 1999
B.A., Department of Economics, Catholic University of Louvain, September1997, high honors |
Irrational
Exuberance, Alliance for Lifelong Learning, Online Instructor, Spring 2002
Mathematics and Science Tutor, Pierson College, Yale, Spring 2001
Financial Markets, Head Teaching Assistant, Yale College, Spring 2001
American Economic History, Teaching Assistant, Yale College, Fall 2000 |
The observation
that stock markets are predictable, and thus inefficient, is a stylized fact of modern
finance (Cochrane, 1999, Campbell, 2000). A theoretically attractive forecasting variable
is the dividend yield, defined as the dividend-over-price ratio. When regressing returns
on the dividend yield, much applied work (e.g. Campbell and Shiller, 1998, 2002) has
documented positive and statistically significant coefficients, and more so for long than
short run returns. This evidence suggests that a low dividend yield is likely to be
followed by lower returns, even when measuring returns over a horizon as long as ten
years. Similarly, when regressing long run returns on past long returns, the slope
coefficient is negative and significant, suggesting that stock returns exhibit mean
reversion in the long run.
However, the assumptions guaranteeing the statistical validity of these long run
regressions are poorly understood, and the exact asymptotic distribution of the estimators
is often unknown under the alternative of stock market predictability. The dissertation
fills up this gap in the literature by providing a theoretical and econometric analysis of
these forecasting regressions. In particular, it is demonstrated that the empirical
proposition that stock markets are highly predictable is only valid under stringent and
unrealistic assumptions on the time series properties of dividend yield and long run
overlapping return data (Chapters 1 and 3). Moreover, when using the dividend yield as a
forecasting variable, the evidence against the efficient market hypothesis stands up only
if one is willing to assume a particular type of dividend policy model (Chapter 2).
Chapter 1: A Spectral Domain Analysis of Mean Reversion Tests
This chapter develops a robust technique for dealing with overlapping data in long run
autoregressive regressions, when the degree of overlap is large compared to the sample
size. It is shown that a (time domain) regression with overlapping data naturally
transforms into a narrow band spectral regression around the long run frequency. This
insight enables us to derive the asymptotic distribution and convergence rates for the
least squares estimator, both under the null and the alternative of mean reversion. In
particular, it is shown that the Richardson and Stock (1989) proposition that the least
squares estimator becomes inconsistent when the overlap is large compared to the sample
size, results from the low convergence rate of the estimator when the amount of
(independent) data is small. Moreover, the spectral regression approach, when combined
with the bandwidth selection techniques of Andrews (1991) or Newey and West (1994), yields
more efficient estimators, which converge to normal distributions, both under the null and
the alternative. These insights extend to variance ratio tests and allow us to introduce a
new variance ratio test that compares high and low frequency components in the data.
Simulations show that the spectral approach performs well in small samples, where
traditional techniques lead to biased and size-distorted tests. In the empirical
application, this new technique is used to test for mean reversion in S&P 500 stock
returns.
Chapter 2: Examining the Statistical Properties of Financial Ratios
This chapter questions the theoretical premise that the dividend yield and price earnings
ratio ought to have predictive power for the aggregate stock market. By referring to the
Modigliani and Miller (1961) theorem, which states that dividend policy is irrelevant, it
is shown that the dividend yield can have any time series property, including the presence
of a unit root, unless one is willing to assume a specific type of dividend policy model
such as Lintner (1956). On the other hand, the price-earnings ratio for the aggregate
market is proved to be stationary under the hypothesis that markets are efficient.
Applying cointegration methodology to S&P 500 data, I find that dividends and prices
are cointegrated, but with a (long-run) elasticity of about 1.5, confirming the finding of
Barsky and Delong (1993) that prices seemingly overreact to dividends, even in the long
run. Taking account of this finding, I demonstrate that the regression of capital gains or
dividend growth on dividend yields suffers from severe misspecification bias, with the
bias going against the efficient market hypothesis. Contrary to the dividend yield, the
hypothesis that the price-earnings ratio is stationary cannot be rejected, even when
including the recent boom and bust in the stock market. Furthermore, simple tests of
market efficiency based on earnings do not reject the efficient market hypothesis.
Chapter 3: Long Run Regressions: Pitfalls and Tests
The third chapter studies in detail the pitfalls of using long run (multivariate)
regressions with overlapping data to test the efficient market hypothesis. Extending the
framework of Richardson and Stock (1989), I derive the analytic form of the bias in the
slope coefficient when regressing an overlapping variable on a highly persistent variable
such as the dividend yield. The bias depends on the degree of overlap in the dependent
variable, and therefore helps to explain the low and insignificant regression coefficients
typically found in short run return regressions, and the large and highly significant
slope coefficients in long run regressions. To overcome the bias problem, I propose a
slight modification to the dividend yield regression, and show how to apply the techniques
developed in Chapter 1 to construct an efficient test of the (long run) efficient market
hypothesis.
Other Papers
Liquidity Risk in Financial Markets
This paper constructs a simple overlapping generations model of an asset market with a
stochastic number of traders. The unpredictable nature of the population creates liquidity
risk in the price of the asset, since the holder of the asset faces the risk that he or
she might not find a buyer when selling the asset. As a result, equilibrium prices can
diverge significantly from their fundamental values, despite the absence of risk in the
fundamentals of the economy. The discrepancy between prices and fundamental values depends
on the thinness of the market, the beliefs about the future liquidity of the market and
the myopia of the traders. The model sheds light on a number of financial anomalies; in
particular, it provides an explanation for the excess volatility of asset prices (Shiller
1981), since prices do not only depend on the fundamentals but also on their (uncertain)
resale value. It also shows how high stock returns can be compatible with a low
correlation of returns with consumption, providing an explanation for the equity premium
puzzle identified by Mehra and Prescott (1985). Finally, the underpricing of closed-end
mutual funds (Malkiel 2000) is consistent with the higher risk, and therefore the higher
required return, from holding an asset suffering from inferior liquidity. |