Aggregation of Information About the Cross Section of Stock Returns: A Latent Variable Approach

We propose a new approach for estimating expected returns on individual stocks from a large number of firm characteristics. We treat expected returns as latent variables and apply the partial least squares (PLS) estimator that filters them out from the characteristics under an assumption that the characteristics are linked to expected returns through one or few common latent factors. The estimates of expected returns constructed by our approach from 26 firm characteristics generate a wide cross-sectional dispersion of realized returns and outperform estimates obtained by alternative techniques. Our results also provide evidence of commonality in asset pricing anomalies.