From Volatility to Profits: A GARCH and Random Forest-Driven Trading Framework

Yash Deshpande -

Researcher

Utkarsh Babaria

Faculty Advisor

Dr. Ionut Florescu

Summary

This study introduces a hybrid trading framework that integrates volatility forecasting using the GJR-GARCH model with machine learning-based market prediction via Random Forest. The objective is to leverage the predictive power of both methods to design a data-driven trading strategy that improves market predictability and profitability.

The framework applies volatility forecasting using GJR-GARCH to a panel of equities, incorporating these forecasts into a Random Forest classifier to predict market movement probabilities. A trading strategy is then developed using these predictions and tested on historical financial data.

Key Findings

  • The GJR-GARCH model effectively captures volatility clustering and leverage effects, improving market risk assessment.
  • The Random Forest classifier enhances market movement prediction, leveraging financial and macroeconomic data.
  • The trading strategy based on these predictions outperforms traditional benchmarks, showing higher cumulative returns and reduced drawdowns.

Future Implications

The study demonstrates the potential of combining econometric modeling and machine learning for financial forecasting. Future improvements could include:

  • Expanding data sources (e.g., sentiment analysis, news, social media trends).
  • Exploring alternative machine learning models (e.g., XGBoost, LSTMs).
  • Optimizing risk management techniques for better performance.

This research highlights the importance of hybrid approaches in trading, showing that integrating traditional financial models with modern AI techniques can enhance trading strategies and market prediction accuracy.