Market Microstructure Forecasting with Deep RL
Published:
An end-to-end system for predicting short-horizon price movements from order book snapshots and using those predictions to inform a deep reinforcement learning agent that executes trades. Performance is evaluated via event-driven backtesting against historical market data.
Key components: Order book feature engineering from L2 snapshot data · Supervised forecasting model for short-horizon price direction · DRL agent (PPO/DQN) trained on predicted signals · Event-driven backtesting engine with realistic transaction costs and slippage
Stack: Python, PyTorch, reinforcement learning, quantitative finance, backtesting