Deep Learning, Predictability, and Optimal Portfolio Returns

TL;DR

This paper demonstrates that deep learning models, especially LSTM networks, significantly improve portfolio performance by better forecasting stock returns, outperforming linear models across various market conditions and constraints.

Contribution

It introduces the use of deep neural networks, including LSTM architectures, for dynamic portfolio selection, showing their advantages over traditional linear models in financial forecasting.

Findings

Deep neural networks outperform linear regressions in portfolio returns.

LSTM models provide incremental benefits with more frequent rebalancing.

Performance gains are robust across different market conditions and constraints.

Abstract

We study the dynamic portfolio selection of an investor who uses deep learning methods to forecast stock market excess returns. In a two-asset allocation problem, deep neural networks -- both feedforward and long short-term memory (LSTM) recurrent architectures -- deliver economically significant gains in terms of certainty equivalent returns and Sharpe ratios relative to linear predictive regressions. These gains are robust to alternative performance measures, the inclusion of transaction costs, borrowing and short-selling constraints, different rebalancing horizons, and subsample splits, and are particularly pronounced during NBER recessions and periods with large return swings. Within the class of neural networks we consider, economic performance is broadly similar across architectures, with the recurrent LSTM specification providing incremental benefits with more frequent rebalancing. Overall, our evidence suggests that exploiting the time-series structure of standard predictor variables via deep learning can generate meaningful portfolio improvements for investors beyond those obtained from linear models.