Measuring information decay in financial markets

Abstract

We propose a general method to quantify the rate of information decay. We do so by leveraging the ability of contemporary machine learning methods to efficiently extract information from data. In an application to financial markets we find that stock return predictability closely follows an asymptotic exponential decay process in which the value of information decays at around 7% per month. This suggests that a one month delay in using information is associated with a 7% decrease in expected returns. The speed of information decay does not vary substantially across stocks with different sizes or liquidity. However, if investors restrict their investment universe to stable stocks, the information decay in the first month is around 60% but close to zero thereafter. Our findings can assist institutional investors by quantifying the expected costs of trading at a lag. Institutional investors need to invest resources to acquire and process information on a timely basis. Using our method investors can quantify the expected benefit of more timely trading, thereby determining the optimal level of investment in these processes.

Keywords: finance, information decay, long-run returns, return predictability, machine learning

JEL Classification: C44, C53, D80, G12, G17