Factor Returns and Out-of-Sample Alphas : Factor Construction Matters

Portfolios formed on a time-varying basis from the principal components of the factors compiled by Chen and Zimmerman (CZ) and Jensen, Kelly, and Pedersen (JKP) display large and robust out-of-sample Sharpe ratios, implying that the factors strongly forecast the cross-section of stock returns. However, average Sharpe ratios obtained based on the CZ factors are notably larger than when using the JKP factors. We investigate this divergence, documenting the roles of the weighting methods used to construct factor returns, the number of quantile portfolios employed to construct factor returns, and divergences in the numbers and composition of the factors contained across databases. We also show that factor principal components beyond the first few contribute substantially to the factors’ ability to forecast the cross-section of returns