Ensembling Deep Reinforcement Learning Strategies for Crypto Trading

Summary

This study presents an ensemble approach for intraday cryptocurrency portfolio trading with deep reinforcement learning. It selects candidate models using multiple validation periods, then combines selected policies through a mixture distribution. The goal is to improve generalization in a highly stochastic market environment.

The authors assess performance across granular out-of-sample test periods and periodically retrain models to account for changing financial data. They report improved out-of-sample performance against both a standalone deep reinforcement learning strategy and passive investment. The document does not provide details about assets, costs, implementation, or the size and statistical significance of the gains, so the reported comparison alone does not establish live trading effectiveness.

Key ideas

• Multiple validation periods are used to select models for the ensemble.
• A mixture distribution policy combines the selected reinforcement learning models.
• Performance is examined across shorter out-of-sample periods to assess robustness as markets evolve.
• Periodic retraining is used to address non-stationary financial data.
• The reported ensemble outperforms a standalone deep reinforcement learning strategy and passive investment in out-of-sample comparisons.

Tags

# An Ensemble Method of Deep Reinforcement Learning for Automated Cryptocurrency Trading


# An Ensemble Method of Deep Reinforcement Learning for Automated Cryptocurrency Trading









We propose an ensemble method to improve the generalization performance of trading strategies trained by deep reinforcement learning algorithms in a highly stochastic environment of intraday cryptocurrency portfolio trading. We adopt a model selection method that evaluates on multiple validation periods, and propose a novel mixture distribution policy to effectively ensemble the selected models. We provide a distributional view of the out-of-sample performance on granular test periods to demonstrate the robustness of the strategies in evolving market conditions, and retrain the models periodically to address non-stationarity of financial data. Our proposed ensemble method improves the out-of-sample performance compared with the benchmarks of a deep reinforcement learning strategy and a passive investment strategy.

Shown in full with attribution under the source's licence. Licence: abstract CC0

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.