A Machine Learning Approach to Risk Factors: A Case Study Using the Fama–French–Carhart Model

Abstract

Factor models are by now ubiquitous in finance and form an integral part of investment practice. The most common models in the investment industry are linear, a development that is no doubt the result of their familiarity and relative simplicity. Linear models, however, often fail to capture important information regarding asset behavior. To address the latter shortcoming, the authors show how to use random forests, a machine learning algorithm, to produce factor frameworks that improve upon more traditional models in terms of their ability to account for nonlinearities and interaction effects among variables, as well as their higher explanatory power. The authors also demonstrate, by means of a simple example, how combining the random forest algorithm with another machine learning framework known as association rule learning can produce viable trading strategies. Machine learning methods thus show themselves to be effective tools for both ex post risk decomposition and ex ante investment decision-making.