Reinforcement Learning Trading Bot
Figure 1: Stock market data visualization
Figure 2: Model predictions for 2018 trading period
Figure 3: Prediction results showing model performance
Overview
Implemented autonomous trading system using OpenAI's reinforcement learning algorithms, deployed on live cryptocurrency exchange.
Challenge
Apply RL to financial markets (stocks and crypto). Create realistic trading environment for agent training. Represent market data effectively for neural networks. Deploy and run safely on live exchange.
Solution & Approach
Implemented custom OpenAI Gym environment for trading. Visual representation: historical chart data + portfolio state. Agent actions: buy, sell, hold. Extensive experimentation with state representations and embeddings. Deployed on BitMEX exchange with live capital. Ran for multiple months across several accounts.
Outcome & Impact
Achieved profitability on live exchange. Ran for multiple months across accounts. Gained understanding of RL fundamentals. Learned importance of information representation for AI. Most difficult part of RL: environment implementation. Buy-and-hold strategy is difficult to beat.
Technologies Used
Key Highlights
- ✓Achieved profitability on live cryptocurrency exchange
- ✓Custom OpenAI Gym environment for trading
- ✓Ran for multiple months across accounts
- ✓Visual representation with historical charts
- ✓Learned importance of RL environment design