Shrinking Factor Dimension: A Reduced-Rank Approach
62 Pages Posted: 23 Jul 2018 Last revised: 17 Oct 2019
Date Written: October 17, 2019
We propose a reduced-rank approach to reduce a large number of factors to a few parsimonious ones. It is designed to explain the cross section of stock returns, more suitable than principal component analysis and partial least squares in factor dimension reduction. Out of 70 factor proxies, we find that the best five combinations of them seem adequate. However, they do not improve much for pricing individual stocks, though they outperform the Fama-French (2015) five factors for pricing industry portfolios as expected. Our results suggest that new factors are wanted to reduce the pricing errors at the firm level.
Keywords: reduced rank, PCA, PLS, factors, factor model, cross section
JEL Classification: G1, G11, G12, G17
Suggested Citation: Suggested Citation