A recommender system for active stock selection

Abstract

The goal of this report is to equip equity portfolio managers with a new tool to assist them in the crucial task of narrowing down a broad universe to a list of stocks to be analysed in depth. We explore a number of alternative approaches to building a recommender system, i.e. a predictive model which generates stock recommendations based on observable characteristics and previous investor behaviour. The empirical analysis uses data on a large set of global active mutual funds, observed between 2005 and 2016, to calibrate the models and test their predictive ability out of sample. Our main conclusion is that a simple dimension reduction technique achieves the best compromise between precision and recall. Moreover, our recommender system displays good predictive power, particularly when used to forecast future buy trades.