S&P 500 Forecasting with Machine Learning

This project explores the predictive power of machine learning models in forecasting returns of the S&P 500 index over short- to medium-term horizons. The framework includes rigorous feature engineering, model training and evaluation, and out-of-sample backtesting. The goal is to assess the viability of ML-based return forecasts in a real-world trading context.

---

Project Overview

• Objective: Identify which ML algorithms (e.g., Ridge, Lasso, XGBoost) offer consistent predictive power on equity index returns.
• Approach:
  • Predict log returns of the S&P 500 using daily data.
  • Evaluate forecasts for 1-day, 1-week, and 1-month ahead horizons.
  • Apply rolling-window backtesting to simulate live performance.
  • Analyze economic value through cumulative returns and Sharpe ratios.
• Scope: Combines financial signal processing, time-series modeling, and ML performance evaluation in a fully reproducible pipeline.

---

Models and Techniques

• Regression Models: Ridge, Lasso, Decision Tree Regressor, XGBoost
• Feature Engineering:
  • Technical indicators (RSI, MACD, moving averages)
  • Lagged returns, volatility, momentum factors
  • Calendar/time features
• Backtesting: Expanding and rolling-window forecasts with out-of-sample evaluation

---

Repository Structure

• data/: contains match CSV and Excel files
• notebooks/: main analysis notebook
• reports/: final presentation (PDF)
• LICENSE
• README.md
• requirements.txt

Tools & Libraries

• Python 3.10+
• pandas, numpy, scikit-learn, xgboost
• matplotlib, seaborn, yfinance
• statsmodels (for comparison with ARIMA/GARCH)

---

Feel free to connect via LinkedIn or view more projects on GitHub.