AlphaFoundry: Factor-Based Quantitative Strategy Engine

AlphaFoundry is a quantitative investment framework that implements a rolling-window strategy to outperform the S&P 500. It combines traditional Fama-French 5-factor analysis with modern machine learning (XGBoost Learning-to-Rank) to forecast excess returns and construct optimized portfolios.

---

📈 Strategy Overview

The system operates on a monthly rebalancing schedule, using a "Walk-Forward" validation process to prevent data leakage.

1. Data Ingestion: Processes daily market data (S&P 500 constituents) and Fama-French factors.

2. Feature Engineering: Computes rolling betas (sensitivity) to Market, Size (SMB), Value (HML), Profitability (RMW), and Investment (CMA) factors.

3. Alpha Generation:

   • Base Model: OLS Rolling Regression.
   • ML Model: XGBoost Ranker trained on 36-month lookback windows.

4. Submission Checklist: Selects Top-K assets (e.g., decile spread) equal-weighted

🏗️ Architecture

The repository contains both the research environment and a production-grade inference API.

• inference.py: FastAPI application serving the trained model.
  • Endpoints: /topk, /health
  • Capabilities: On-the-fly ranking of 500+ assets based on live factor data.
• one.ipynb: Comprehensive research notebook containing the full backtest pipeline, EDA, and model comparison (OLS vs XGBoost).

🚀 Usage

1. Research & Backtesting

Open one.ipynb to view the full end-to-end backtest, including:

• Data cleaning and alignment.
• Factor loading estimation (Rolling OLS).
• Performance metrics (Sharpe Ratio, Max Drawdown, Cumulative Return vs SPY).

2. Production Inference API

To run the ranking engine locally:

# Install dependencies
pip install -r requirements.txt

# Start the API server
uvicorn inference:app --reload

API Example: Get top 50 ranked stocks for the current month:

curl "http://localhost:8000/topk?k=50&n_bins=5"

📊 Performance Benchmark

Results based on 2016-2025 Out-of-Sample Backtest:

Metric	AlphaFoundry (XGB)	S&P 500 (SPY)
Annualized Return	14.2%	11.8%
Sharpe Ratio	0.95	0.85
Max Drawdown	-18.4%	-24.5%

(Note: Performance metrics are based on backtested data and do not guarantee future results.)

🛠️ Requirements

• Python 3.8+
• pandas, numpy, scikit-learn, xgboost
• fastapi, uvicorn
• Fama-French Data (included in data/raw)

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.