Research: Comparative Performance of GARCH and Deep Learning Models in Financial Volati...

Research question: Do deep learning models outperform traditional GARCH models in forecasting financial market volatility across major asset classes?

Background & Motivation: Accurate volatility forecasting is fundamental in finance for risk management, option pricing, and asset allocation. Traditionally, models like GARCH have been widely used due to their statistical robustness and interpretability. However, the rise of deep learning has introduced new, flexible approaches that may capture non-linear patterns present in financial time series.

Research Gap: While several studies have compared GARCH and deep learning methods for volatility forecasting, there is limited empirical evidence across diverse asset classes using standardized datasets and rigorous evaluation metrics. The comparative performance, robustness, and practical implications of these models remain underexplored.

Approach & Expected Contribution: This study will systematically evaluate GARCH and deep learning models (such as LSTM and CNN) on historical price data from equities, FX, and commodities. The methodology involves training and testing models on standardized datasets, comparing forecast accuracy using statistical and economic criteria, and analyzing model robustness. The expected contribution is a comprehensive, evidence-based assessment guiding model selection for volatility forecasting.

Why It Matters: Enhanced volatility forecasts improve financial decision-making for institutional and individual investors. Understanding the strengths and limitations of both model types informs practitioners and researchers, supporting the adoption of more effective risk-management tools in both academia and industry.