We use supervised learning to identify factors that predict the cross-section of returns and maximum drawdown for stocks in the US equity market. Our data run from January 1970 to December 2019 and our analysis includes ordinary least squares, penalized linear regressions, tree-based models, and neural networks. We find that the most important predictors tended to be consistent across models, and that non-linear models had better predictive power than linear models. Predictive power was higher in calm periods than in stressed periods. Environmental, social, and governance indicators marginally impacted the predictive power of non-linear models in our data, despite their negative correlation with maximum drawdown and positive correlation with returns. Upon exploring whether ESG variables are captured by some models, we find that ESG data contribute to the prediction nonetheless.