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

Background and Context

Research Focus

This study performs a comparative analysis of machine learning methods for measuring asset risk premiums and stock return predictability.

Data Coverage

The analysis covers 30,000 stocks over 60 years (1957-2016), using 94 stock characteristics and 8 macroeconomic predictors.

Methodology

Researchers compare 13 different machine learning approaches including neural networks, regression trees, and linear models for predicting stock returns.

Neural Networks Outperform Traditional Methods in Portfolio Returns

  • Neural networks (NN1-NN4) achieve the highest predictive accuracy with R² values around 0.40%
  • Traditional linear regression (OLS) performs poorly with negative R²
  • Tree-based methods (RF, GBRT) show intermediate performance

Machine Learning Models Excel at Market Timing

  • Neural network-based timing strategy improves Sharpe ratio by 26 percentage points
  • Demonstrates real economic value of machine learning predictions
  • Shows substantial improvement over passive buy-and-hold strategy

Neural Network Depth vs Performance

  • Performance peaks at 3 hidden layers (NN3)
  • Adding more layers does not improve predictions
  • Suggests "shallow" learning is optimal for stock returns

Long-Short Portfolio Performance

  • Neural network strategy achieves highest Sharpe ratio of 1.35
  • More than doubles performance of traditional strategies
  • Demonstrates economic significance of machine learning approaches

Contribution and Implications

  • Machine learning methods, particularly neural networks, significantly improve stock return predictions compared to traditional approaches
  • The improvements translate into meaningful economic gains for investment strategies
  • Results validate the growing adoption of machine learning in the investment industry

Data Sources

  • Performance comparison chart based on Table 1 monthly R² values
  • Market timing Sharpe ratio comparison based on Table 6 data
  • Predictor importance visualization based on Figure 5 rankings
  • Neural network depth comparison based on Table 1 NN1-NN5 results
  • Long-short portfolio performance based on Table 7 Sharpe ratios