Anomaly return predictability using deep learning asset pricing
Working paper, 2019
This paper studies anomaly return predictability across deciles using a set of fifty anomaly variables built using individual stock characteristics. I construct deciles and study their predictability using their own past information, other macroeconomic variables, and limit-to-arbitrage variables. I find that some anomalies are persistent and that there are some predictors which help to forecast the decile portfolio returns. Deciles predictability is not uniform across anomaly variables and predictors. Namely, all deciles are not uniformly predictable but extreme deciles seem to be more often predictable. Stock variance, dividend yield, and dividend price ratio are strong predictors for decile portfolio returns. Most importantly, hedge portfolios are often predictable by the TED spread and Amihud illiquidity measure, which indicate that trading frictions may explain the persistence of these portfolio returns. Furthermore, I use the rich set of five hundred anomaly portfolios to investigate their prediction properties using Deep learning techniques.
Recommended citation: Stéphane, N Dri. (2019). "Anomaly return predictability using deep learning asset pricing ." Working paper.