Deep Fundamental Factor Models

TL;DR

This paper introduces deep fundamental factor models that leverage neural networks to capture complex non-linear relationships in asset pricing, providing uncertainty quantification and outperforming traditional linear models.

Contribution

It presents a novel deep learning framework for factor modeling that includes uncertainty quantification and demonstrates superior performance over linear and quadratic models.

Findings

Information ratios are approximately 1.5x greater than OLS models.

Deep models outperform quadratic LASSO in robustness and predictive power.

Uncertainty bands naturally arise from network weights, aiding interpretability.

Abstract

Deep fundamental factor models are developed to automatically capture non-linearity and interaction effects in factor modeling. Uncertainty quantification provides interpretability with interval estimation, ranking of factor importances and estimation of interaction effects. With no hidden layers we recover a linear factor model and for one or more hidden layers, uncertainty bands for the sensitivity to each input naturally arise from the network weights. Using 3290 assets in the Russell 1000 index over a period of December 1989 to January 2018, we assess a 49 factor model and generate information ratios that are approximately 1.5x greater than the OLS factor model. Furthermore, we compare our deep fundamental factor model with a quadratic LASSO model and demonstrate the superior performance and robustness to outliers. The Python source code and the data used for this study are provided.