Stock Trading with Deep Reinforcement Learning

Overview

This project explores the use of Deep Reinforcement Learning (DRL) to develop and evaluate stock trading strategies. The project is based on cutting-edge research and aims to leverage advanced DRL techniques to optimize trading decisions.

Objectives

• Implement and compare various DRL agents for stock trading.
• Evaluate the performance based on return, volatility, and Sharpe ratio.
• Include transaction costs in the trading environment for realistic simulations.

Environment Setup

1. Install Dependencies:

   • Install necessary packages such as gymnasium, numpy, pandas, matplotlib, and stable-baselines3.

2. Clone the Repository:

   • Clone the GitHub repository and navigate to the project directory.

Steps to Run the Project

1. Prepare the Stock Data:

   • Ensure your stock data is in a suitable format (e.g., CSV files).
   • Load the data into the environment.

2. Define the Trading Environment:

   • Create a custom trading environment (StockTradingEnv) that includes transaction costs.
   • Define the observation and action spaces for the environment.

3. Implement DRL Agents:

   • Implement various DRL agents using algorithms such as PPO, A2C, DDPG, SAC, TD3.
   • Train the agents on the custom trading environment.

4. Evaluate the Agents:

   • Calculate daily returns, standard deviation, and Sharpe ratio for each agent.
   • Compare the performance of the agents based on these metrics.

5. Plot the Results:

   • Visualize the performance metrics using bar charts.
   • Highlight the best-performing agents based on the Sharpe ratio.

Results

• DDPG Agent showed the highest return and Sharpe ratio, despite higher volatility.
• TD3 Agent and Ensemble Agent also performed well, balancing return and risk.
• PPO Agent underperformed, indicating the need for further optimization.

Key Takeaways

• Advanced DRL algorithms like DDPG and TD3 can significantly enhance trading strategies.
• Including transaction costs makes simulations more realistic.
• Continuous improvement and testing are essential for refining these models.

Research Paper

This project was inspired by the research paper Deep Reinforcement Learning for AutomatedStock Trading: An Ensemble Strategy. The paper provides a comprehensive overview of using DRL in financial markets and guided the development of the agents and environment in this project.