Statistical Arbitrage Engine

A quantitative statistical arbitrage engine that identifies mean-reverting pairs and executes market-neutral strategies with machine learning enhancements.

Pair Discovery:

• Cointegration testing (Engle-Granger, Johansen)
• Correlation matrix with rolling windows
• Sector/industry pair clustering
• Cross-asset pair discovery (crypto, stocks, forex)
• Dynamic pair rebalancing

Signal Generation:

• Z-score based entry/exit signals
• Bollinger Band spread signals
• Kalman filter for dynamic hedge ratios
• Half-life estimation for mean reversion
• Regime detection (trending vs mean-reverting)

ML Enhancement:

• LSTM for spread prediction
• Random forest for regime classification
• Feature engineering (volume, volatility, momentum)
• Walk-forward optimization
• Overfitting protection with cross-validation

Execution:

• Dollar-neutral position sizing
• Beta-adjusted hedging
• Gradual entry/exit (TWAP/VWAP)
• Portfolio-level risk management
• Margin and leverage optimization

Backtesting:

• 5-year historical data included
• Transaction cost modeling
• Sharpe ratio, Sortino, max drawdown
• Monte Carlo simulation

Tech Stack:

• Python + NumPy + pandas + scikit-learn
• Zipline/Backtrader backtesting
• Interactive Brokers / CCXT execution
• Jupyter notebooks included
• Docker deployment