Commodity Risk Factors: A Machine Learning Approach
53 Pages Posted:
Date Written: June 27, 2019
To which extent are financial market returns predictable? Standard linear approaches à la Fama & French (1992) are widespread. Yet, they have difficulties in addressing this question as implicit assumptions undermine their return predicting potential. We employ tree-based methods to overcome these limitations and attempt to empirically approximate an upper bound for the predictability of commodities futures returns. Out-of-sample, we find that up to 3.74% of one-month ahead returns are predictable --- more than a 10-fold increase from linear risk factor approaches. Our findings hint at the importance multi-way interactions and non-linearities acquire in the data. They imply that new factors should be tested on their ability to add explanatory power to an ensemble of existing factors.
Keywords: Machine Learning, Empirical Asset Pricing, Ensemble Methods, Return Forecasting
JEL Classification: G11, G12, G13, G14
Suggested Citation: Suggested Citation