Momentum Trading System

A machine-learning-driven momentum strategy built using Python, Auquan Toolbox, and custom backtesting logic. The project downloads historical Level-2 market data, generates features, trains predictive models, and evaluates trade performance using a structured backtesting pipeline.

---

📘 Overview

This project aims to forecast short-term changes in the basis (difference between stock and future prices) and trade based on expected movement. It combines feature engineering, ensemble ML models, and a custom portfolio simulator to evaluate strategy performance.

---

🧰 Features

✔ Data Pipeline

• Auto-download of historical data using Auquan Historical Data API
• Cleans & processes stock and futures order-book information
• Computes mid-price, bid/ask volumes, target variable, etc.

✔ Feature Engineering

Includes multiple predictive indicators:

• Momentum: mom10
• EMA signals: emabasis2, emabasis5, emabasis10
• Volume imbalances: totalaskvolratio, totalbidvolratio
• RSI movement features
• Shifted/normalized basis transformations

✔ Machine Learning

• Extra Trees Regressor (sklearn)
• Achieves high variance score (~0.95 on sample run)
• Smooth predictive output suitable for trading signals

✔ Backtesting System

Tracks:

• Instrument-level & portfolio-level PnL
• Score (normalized performance metric)
• Capital usage
• trades: wins/losses
• Maximum Drawdown
• Variance & volatility of returns
• Portfolio value progression

---

📊 Sample Strategy Output

Sample JSON summary:

{
  "instrument_names": ["MQK"],
  "instrument_stats": [{
      "pnl": { "MQK": 0.11692330000000006 },
      "score": { "MQK": 0.5270070414825269 }
  }],
  "pnl": 0.11692330000000006,
  "trading_days": 3,
  "score": 0.5270070414825269,
  "maxDrawdown": 483.66499999999905,
  "maxPortfolioValue": 11230.483,
  "total_loss": 4699.115000000002,
  "variance": 790.2561871673435,
  "count_profit": 228,
  "capital": 9936.30299999999,
  "capitalUsage": 1324.7970000000041,
  "portfolio_value": 11169.233,
  "total_profit": 5868.347999999998,
  "count_loss": 222
}

Interpretation Highlights

• ~50% win-rate (228 wins / 450 trades)
• Moderate net PnL (after aggregation)
• High drawdown → high-risk strategy
• High variance → volatile performance
• Model signal works but risk controls need improvement

---

🚀 Getting Started

Clone the repository

git clone https://github.com/VarunBanka/Momentum-Trading-System
cd Momentum-Trading-System

Install Dependencies

pip install scikit-learn tensorboardX beautifulsoup4 auquan_toolbox --no-deps

Run the Notebook

Open:

Momentum_Trading_System.ipynb

Run all sections sequentially:

1. Data download
2. Cleaning & preprocessing
3. Feature engineering
4. Model training
5. Validation
6. Backtesting & output analysis

---

📈 Future Enhancements

• Add Stop-Loss/Take-Profit logic
• Add slippage & transaction costs
• Integrate more ML models (XGBoost, LightGBM, LSTM)
• Walk-forward or rolling-window training
• Add visualizations for equity curve + drawdown
• Multi-instrument diversification

---

⚠️ Disclaimer

This project is for educational and research purposes only.

• Nothing here is financial advice.
• Trading carries significant risk, including total capital loss.
• The author & contributors are not responsible for any financial losses.
• Models trained on historical data do not guarantee future performance.
• Use at your own risk.

---

🤝 Contributing

PRs and suggestions are welcome! If you want to add features, please open an issue first.

---

⭐ Support

If you find this project useful, consider giving it a ⭐ on GitHub!

---