Abstract

The year 2020 presented unprecedented challenges to global financial markets, with extreme volatility patterns emerging due to the COVID-19 pandemic, policy interventions, and economic uncertainties. This study examines the application and performance of neural network-based forecasting models in predicting stock market volatility during this turbulent period. We analyze various neural network architectures including Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), and Convolutional Neural Networks (CNN) in forecasting volatility across major stock indices. Our findings indicate that LSTM models demonstrated superior performance in capturing the non-linear dynamics and sudden regime changes characteristic of 2020's market conditions. The study reveals that neural networks significantly outperformed traditional econometric models like GARCH and EGARCH, particularly during periods of extreme market stress. These results have important implications for risk management, portfolio optimization, and derivative pricing during crisis periods.