Analysis of Financial Contagion and Prediction of Dynamic Correlations During the COVID-19 Pandemic: A Combined DCC-GARCH and Deep Learning Approach

Abstract

This study aims to combine the use of dynamic conditional correlation multiple generalized autoregressive conditional heteroskedasticity (DCC-GARCH) models and deep learning techniques in analyzing the dynamic correlation between stock markets. First, we examine the contagion effect of the high-risk financial crisis during COVID-19 in the United States on the Latin American stock market using a dynamic conditional correlation approach. The study covers the period from 2014 to 2020, divided into the pre-COVID-19 period (January 2014–February 2020) and the COVID-19 period (March 2020–November 2020), to examine the sudden change in average conditional correlation from one period to the next and identify the contagion effect. The contagion test showed significant contagion between the S&P 500 and Latin American indices, except for Argentina’s MERVAL. Additionally, we applied deep learning models, specifically LSTM, to predict market dynamics and changes in volatility as an early warning system. The results indicate that incorporating LSTM improved the accuracy of predicting dynamic correlations and provided early risk signals during the crisis. This suggests that combining DCC-GARCH with deep learning techniques is a powerful tool for predicting and managing financial risk in highly uncertain markets.