Empirical Asset Pricing via Machine Learning

Abstract

We perform a comparative analysis of machine learning methods for the canonical problem of empirical asset pricing: measuring asset risk premiums. We demonstrate large economic gains to investors using machine learning forecasts, in some cases doubling the performance of leading regression-based strategies from the literature. We identify the best-performing methods (trees and neural networks) and trace their predictive gains to allowing nonlinear predictor interactions missed by other methods. All methods agree on the same set of dominant predictive signals, a set that includes variations on momentum, liquidity, and volatility. Authors have furnished an Internet Appendix, which is available on the Oxford University Press Web site next to the link to the final published paper online.

Machine Learning in Systematic Equity Allocation: A Model Comparison

“…The hyper-parameters for neural networks are chosen to define one rather shallow (NN1) and one deeper (NN3) network. The structures are comparable to those of Abe and Nakayama (2018) and the orders of magnitude are also similar to those of Gu et al (2018).…”

“…• Without being "taught" asset pricing theories, the ML algorithms find most of the factor-investing metrics as important variables, even if the feature dataset is large (350 characteristics). This represents a promising step toward an "empirical asset pricing" approach in quantitative equity (see, e.g., Gu et al, 2018). • Multi-factor signals based on trees boosted, and neural networks outperformed the signals based on simple factor-investing type: • This result holds for decile performance and long/short performance analysis using theoretical equal-weight portfolios.…”

“…• Support vector machines (Cao and Tay, 2003;Matías and Reboredo, 2012;Dunis et al, 2013). • Trees and their extensions (Ballings et al, 2015;Patel et al, 2015;Moritz and Zimmermann, 2016;Krauss et al, 2017;Gu et al, 2018). • Other clustering methods (Raffinot, 2018).…”

“…• Other clustering methods (Raffinot, 2018). • Neural networks and their extensions (Dunis et al, 2008;Adeodato et al, 2011;Ballings et al, 2015;Patel et al, 2015;Heaton et al, 2017;Krauss et al, 2017;Fischer and Krauss, 2018;Gu et al, 2018). • Mixes of several of the above (Dunis et al, 2013;Nair et al, 2017).…”

The Effects of Targeting Predictors in a Random Forest Regression Model

“…(2019) recast a classical kernel weighting function as an adaptive weighting function based on RF, Wager and Athey (2018) estimate heterogeneous treatment effects, Ng (2014) employs trees to forecast economic recessions, and Gu et al (2020) use RF to predict future stock returns using numerous firm specific and common predictors.…”