Title: The (adaptive) Lasso in the Zoo - Firm characteristic selection in the cross-section of expected returns
Authors: Marcial Messmer - University of St.Gallen (Switzerland) [presenting]
Francesco Audrino - University of St Gallen (Switzerland)
Abstract: The adaptive Lasso is shown to be superior to both the Lasso and OLS in most panel specifications with low signal-to-noise ratio based on Monte Carlo Simulations. The results are robust to heteroskedastic, cross-sectionally correlated and non-gaussian errors. Based on the results of the simulation, we find that cross-sectional returns are highly dimensional. However, most published firm characteristics are rejected as predictors for returns when considered in a multivariate selection analysis. The empirical application, which comprises more than 70 published firm characteristics, constructed based on CRSP/Computstat data from 1962-2014, shows that price related FC, namely, the one and twelve month(s) momentum are among the most robustly selected coefficients for explaining differences in average/expected cross-sectional returns. The results are consistent, along large, mid and small cap stocks and for most sub periods considered.