Characteristics-based reversals: Exploiting the gap between predicted and realized returns

This paper investigates whether discrepancies between predicted and realized return deciles — based on cross-sectional firm characteristics — contain information about future returns. Using 94 characteristics, we classify each stock into a predicted decile and compare it to its realized decile each month. A long-short strategy that exploits these deviations yields high raw returns. However, the performance is primarily driven by small, illiquid, high-volatility stocks, leading to substantial transaction costs that limit practical implementation.

To address this, we propose a refinement using within-industry peer comparisons, which improves net performance. These decile-level divergences are not merely noise but reflect latent pricing errors that reverse over time. This mechanism differs from traditional reversal strategies that rely on past return behavior, instead capturing return divergences between predicted and realized rankings. Our findings offer a complementary lens on short-term price corrections, especially in markets where trading frictions delay the adjustment of prices to fundamentals.