Firm Characteristics and Expected Stock Returns

We analyze the joint out-of-sample predictive ability of a comprehensive set of 299 firm characteristics for cross-sectional stock returns. We develop a cross-sectional out-of-sample R2 statistic that provides an informative measure of the accuracy of cross-sectional return forecasts in terms of mean squared forecast error. To improve cross-sectional return forecasts based on a large number of firm characteristics, we propose an E-LASSO approach that implements shrinkage in a flexible manner. Our new approach produces significant cross-sectional out-of-sample R2 gains on a consistent basis over time and provides the most accurate out-of-sample estimates of cross-sectional expected returns to date. The E-LASSO approach also generates substantial economic value in the context of long-short portfolios. Finally, we present evidence that more characteristics work better than fewer with respect to forecasting cross-sectional stock returns