Stock market trading via actor-critic reinforcement learning and adaptable data structure

Area of Science:

• Computational Finance
• Machine Learning
• Algorithmic Trading

Background:

• Stock market investments face challenges due to political, economic, and social volatility.
• Developing accurate, efficient investment models is crucial for minimizing capital loss.

Purpose of the Study:

• To propose an innovative, short-term, automatic investment model using reinforcement learning (RL).
• To enhance the learning and accuracy of investment agents in volatile markets.
• To reduce capital loss during trading operations.

Main Methods:

• Application of a reinforcement learning (RL) model with an actor-critic neural network.
• Utilizing rectified linear unit (ReLU) neurons for specialized agent generation.
• Implementation of an adaptable data window structure for improved learning.

Main Results:

• The RL model demonstrated reduced average losses: 0.03% in Euro, 0.25% in gold, and 0.13% in crude oil.
• Achieved more efficient trading and minimized investment losses across different time periods.
• Reduced the model's overall learning time.

Conclusions:

• The proposed RL model offers an effective solution for short-term, automated trading.
• The adaptable data window structure enhances agent performance in foreign exchange markets.
• This approach contributes to minimizing financial losses in dynamic trading environments.