The cross-section of Indian stock returns: evidence using machine learning

Abstract

ABSTRACT We test whether 35 stock characteristics can explain the cross-section of stock returns in India. We address the limitations of previous studies by using a comprehensive, survivorship bias free sample of all firms listed on the major Indian stock exchanges from 1994 to 2019. Results from Fama-Macbeth regressions show as many as 14 predictors breaching the significance threshold of t-stats greater than three. We also use machine learning methods to generate rolling one-month ahead out-of-sample forecasts of stock returns for all firms in our sample. We find substantial improvement in forecast accuracy when using machine learning compared to OLS. Further, we run additional tests for understanding the economic significance of our findings. Investment strategies based on model forecasts provide significant returns to investors.