The (Adaptive) Lasso in the Zoo - Firm Characteristic Selection in the Cross-Section of Expected Returns

Abstract

We find short-term reversal, the twelve-months momentum and research spending scaled by market-value to be the firm characteristics (FC) most robustly selected by the adaptive Lasso in the US cross-section of stock returns. Moreover, the majority of the 68 FC included in our analysis are not considered. Nonetheless, the return process we identify is multi-dimensional, comprising 14 FC. Additionally, our Monte Carlo Simulations indicate that the adaptive Lasso is superior to Lasso and OLS-based selection in panel specifications with a low signal-to-noise ratio. The results are robust to various assumptions. These findings gain support by an empirical out-of-sample factor analysis.