Dynamic stock-decision ensemble strategy based on deep reinforcement learning.

Area of Science:

• Computational Finance
• Artificial Intelligence
• Machine Learning

Background:

• Stock markets are complex and volatile, necessitating advanced trading agents for investor profitability.
• Existing deep reinforcement learning models have limitations in dynamic market adaptation.

Purpose of the Study:

• To develop and evaluate two novel stock trading decision-making methods.
• To improve investor returns and portfolio management through enhanced algorithmic trading.

Main Methods:

• Proposed a nested reinforcement learning (Nested RL) method integrating three deep reinforcement learning models (A2C, DDPG, SAC).
• Introduced a weight random selection with confidence (WRSC) strategy to leverage agent strengths.
• Validated algorithms on U.S., Japanese, and British stock data.

Main Results:

• The proposed ensemble strategy (Nested RL and WRSC) demonstrated superior performance.
• Achieved higher annualized returns, cumulative returns, and Sharpe ratios compared to baseline methods.
• Indicated improved portfolio management with increased profits under equivalent investment risk.

Conclusions:

• Nested RL and WRSC methods offer significant advantages for algorithmic stock trading.
• These strategies can effectively assist investors in navigating complex market environments.
• The ensemble approach enhances profitability and risk management in portfolio management.