Outperforming Naive Diversification Using Stock Level Information

Abstract

We construct mean-variance portfolios using a factor model approach. We show the importance of portfolio allocation for large unbalanced equity data sets using the full CRSP database. We compare the performance of our portfolio construction methodology to the 1/N naive diversification strategy, standard shrinkage procedures, and alternative factor model estimation. We document significant out-of-sample performance improvement in terms of Sharpe ratios, turnover and certainty equivalent. We show that it is due to improved expected returns estimation coming from the 2-pass regression approach.

Keywords: Mean-Variance, factor model, optimization, 2 stage least square

JEL Classification: G11