The Misalignment between R2 and Sharpe Ratio: A Perspective from the Heterogeneity in Stocks

Abstract

It is a common practice to evaluate the predictability of asset pricing models by model R². However, optimizing R² does not necessarily lead to optimal Sharpe ratios of long-short portfolios. This Misalignment stems from heterogeneity in stocks. We show that optimizing R² of a model with a homogenous factor loading function typically leads to larger pricing errors for tail stocks. We introduce resampling methods to rebalance the weights of tail stocks, which improves the pricing accuracy of the tail stocks and significantly increases Sharpe ratios of long-short portfolios. The resampling methods capture more near-arbitrage opportunities and perform better during crisis periods.

Keywords: Heterogeneity, Factor loading function, Portfolio optimization, Imbalanced learning, Machine learning