Out-of-Sample Performance of Factor Return Predictors

Using a comprehensive set of 92 equity factors, I re-evaluate the performance of variables that have been shown from prior studies to be good predictors of factor returns. I find that most variables do not provide systematic evidence in favor of factor return predictability, as judged by their poor out-of-sample performance. Model instabilities appear to be the primary reason for the weak performance of individual variables. I overcome these limitations by exploring a variety of shrinkage techniques that combine signals from all predictors. Results show that this approach offers more favorable and conclusive evidence for predictability. The shrinkage approach typically leads to significant economic gains for factor timing strategies in real time