Enhancing Alpha Signals from Trade Ideas using Supervised Learning

We present a methodology that uses supervised learning techniques, in particular, 'random forests' and 'gradient boosting trees' to assess the probability of success on trade ideas. We use features related to the underlying stock's characteristics, the characteristics of the contributor of the idea and the trade idea itself, as well as the characteristics of the stock related market microstructure information. Our model improves on average the ex post performance of the long ideas by more than 1% and the performance of the short ideas by more than 2%