On the Informational Content of Asset Prices

Abstract

This paper presents an information-theoretic measure of theoretical return predictability and informational efficiency. Predictability decreases in relative efficiency, which is a nonparametric statistic of the uncertainty content of an empirical distribution relative to the maximum possible uncertainty. The latter is a benchmark corresponding to the uniform distribution over the chosen information partition. The evolution of RE with sample size is non-monotonic and marked by discontinuities corresponding to the arrival of extreme events. The empirical methodology involves computing rolling estimates of RE, its finite-sample bias and standard error. Using 11 long daily return time series, it is found that interest rate returns are more predictable than stock market returns, which in turn are more predictable than spot FX returns. The framework is also used to rank market crash episodes according to their information revelation.

Keywords: Return predictability, relative e?ciency, entropy, extreme events

JEL Classification: C14, G14