Diverging Roads: Theory-Based vs. Machine Learning-Implied Stock Risk Premia

Abstract

We assess financial theory-based and machine learning methods to quantify stock risk premia and investigate the potential of hybrid strategies. In the low signal-to-noise environment of a one-month investment horizon, we recommend to rely on a theory-based strategy that exploits the information in current option prices, especially if the risk premium estimate is to be updated at a high frequency. At the one-year horizon, the theory/option-based strategy and an ensemble of neural networks, two notably different methodologies, perform comparably well. A random forest can improve on the theory-based method, provided that a sufficiently long training period is used. In an effort to connect the opposing philosophies, we identify the use of a random forest to account for the approximation errors of the theory-based approach towards measuring stock risk premia as a promising hybrid strategy. It combines the advantages of two diverging roads in the finance world.

Keywords: stock risk premia, option prices, machine learning

JEL Classification: C53, C58, G12, G17