Please rotate your device to landscape mode to view the charts.

Empirical Asset Pricing via Machine Learning

Journal: Review of Financial Studies

Date: 20200501

Author: Gu, Shihao; Kelly, Bryan; Xiu, Dacheng

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.

Link: Google Scholar

Key Findings

Neural Networks Excel at Return Prediction

Neural networks consistently outperform traditional methods, with NN3-NN5 producing positive R² for every portfolio analyzed. The 3-layer neural network (NN3) achieves monthly out-of-sample R² of 0.40% for individual stocks.

Shallow Learning Beats Deep Learning

Neural networks with 3 hidden layers perform optimally for stock return prediction, with performance declining as more layers are added. This differs from other fields where deeper networks typically perform better.

Strong Economic Performance

Machine learning portfolios deliver substantial economic gains. Neural network-based market timing strategy achieves Sharpe ratio of 0.77 vs 0.51 for buy-and-hold. Long-short portfolio based on NN4 achieves Sharpe ratio of 1.35.

Model Performance Comparison

  • Neural networks (NN1-NN3) achieve highest predictive R² values
  • Traditional OLS performs poorly with negative R²
  • Tree-based methods (RF, GBRT) show competitive performance

Portfolio Strategy Performance

  • Neural network models achieve highest Sharpe ratios, peaking at 1.35 for NN4
  • All machine learning methods outperform basic buy & hold strategy
  • Tree-based methods show strong risk-adjusted performance

Important Predictive Signals

  • Price trend signals (momentum, reversal) are most influential
  • Liquidity measures form second most important category
  • Risk measures (beta, volatility) also show strong predictive power

Contribution and Implications

  • Demonstrates significant economic value of machine learning in asset pricing
  • Identifies optimal neural network architecture for financial prediction tasks
  • Provides framework for combining multiple machine learning approaches
  • Shows importance of nonlinear interactions in asset pricing models

Data Sources

  • Model Comparison Chart: Based on Table 1 showing monthly out-of-sample R² values
  • Sharpe Ratio Chart: Constructed from Table 8 risk-adjusted performance metrics
  • Predictor Importance Chart: Synthesized from variable importance analysis in Section 2.3 and Figure 5