Predicting Winner and Loser Stocks: A Classification Approach

Abstract

We introduce a novel framework for predicting future winner and loser stocks in the cross-section of U.S. stocks using binary response models. Building upon the traditional Fama-MacBeth setup, we predict winners and losers directly using separate cross-sectional binary response models. We investigate the predictive ability of various classical return-based anomalies and find significant changes and time-varying patterns in these predictive relationships over time. Our results demonstrate that portfolio strategies based on these direct winner and loser forecasts consistently outperform standard benchmarks.

Keywords: cross-sectional forecasting, momentum, reversal, seasonality, multinomial logit model

JEL Classification: C25, C53, G11, G12