Volatility Spillovers and Contagion During Major Crises: An Early Warning Approach Based on a Deep Learning Model

Abstract This paper contributes to the ongoing debate on nature and characteristics of volatility transmission channels major crash events in international stock markets between 03 July 1997 09 March 2021. Using dynamic conditional correlations (DCC) for clustering, GARCH-BEKK direction disturbances, Diebold-Yilmaz spillover index level contagion, finds that climbs external shock transmissions have long-lasting impacts domestic due contagion effect during crisis periods. The findings also reveal heavier magnitude financial stress is transmitted Asian countries via Hong Kong market. Additionally, degree spillovers advanced emerging equity smaller compared pure or markets, offering a window opportunity market participants terms portfolio diversification risk management applications. Furthermore, study introduces novel early warning system created by integrating DCC with state-of-the-art deep learning model predict global COVID-19 crisis. experimental analysis long short-term memory network evidence verifying bursts generating signals high accuracy before 12-month period. provides supplementary information decision-making process practitioners, as well indicative facilitates assessment vulnerability policymakers.