https://github.com/ferrangarciarovira/sp500-ml-forecasting

Forecasting S&P 500 returns using ML models across multiple time horizons (1-day, 1-week, 1-month). Includes feature engineering, rolling-window backtesting, and performance evaluation to assess predictive power and trading utility of each model.
https://github.com/ferrangarciarovira/sp500-ml-forecasting

backtesting data-science feature-engineering financial-forecasting investment-strategies machine-learning python quantitative-finance regression-models sp500 time-series

Last synced: 5 months ago
JSON representation

Forecasting S&P 500 returns using ML models across multiple time horizons (1-day, 1-week, 1-month). Includes feature engineering, rolling-window backtesting, and performance evaluation to assess predictive power and trading utility of each model.

• Host: GitHub
• URL: https://github.com/ferrangarciarovira/sp500-ml-forecasting
• Owner: ferrangarciarovira
• License: other
• Created: 2025-05-06T11:25:26.000Z (5 months ago)
• Default Branch: main
• Last Pushed: 2025-05-06T11:45:30.000Z (5 months ago)
• Last Synced: 2025-05-06T12:49:14.697Z (5 months ago)
• Topics: backtesting, data-science, feature-engineering, financial-forecasting, investment-strategies, machine-learning, python, quantitative-finance, regression-models, sp500, time-series
• Homepage:
• Size: 15.6 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# 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)

---

## Authors

- Ferran García Rovira

- Gerard Álvarez

- Joshua Gerstner

- Jiaren Fu

Feel free to connect via [LinkedIn](https://www.linkedin.com/in/ferrangarciarovira/) or view more projects on [GitHub](https://github.com/ferrangarciarovira).