We examine return predictability with machine learning in 46 international stock markets. We calculate 148 stock characteristics and use them to feed a repertoire of different models. The algorithms extract predictability mainly from simple, yet popular, factor types—such as momentum,...