We present a system for combining the different types of predictions given by a wide category of mechanical trading rules through statistical learning methods (boosting, and several model averaging methods like Bayesian or simple averaging methods). Statistical learning methods supply better...