Non-Gaussian models for CoVaR estimation

In this paper we show how to obtain estimates of CoVaR based on models that take into consideration some stylized facts about multivariate financial time series of equity log returns: heavy tails, negative skew, asymmetric dependence, and volatility clustering. While the volatility clustering effect is captured by AR-GARCH dynamics of the GlostenJagannathan-Runkle (GJR) type, the other stylized facts are explained by non-Gaussian multivariate models and copula functions. We compare the different models in the period from January 2007 to March 2020. Our empirical study conducted on a sample of listed banks in the euro area confirms that, in measuring CoVaR, it is important to capture the time-varying dynamics of the volatility. Additionally, a correct assessment of the heaviness of the tails and of the dependence structure is needed in the evaluation of this systemic risk measure. (c) 2021 International Institute of Forecasters. Published by Elsevier B.V. All rights reserved.

Automated summary

This paper demonstrates how to estimate CoVaR based on models that consider stylized facts about equity log returns, including heavy tails, negative skew, asymmetric dependence, and volatility clustering. Different models are compared using data from January 2007 to March 2020. The empirical study shows the importance of capturing time-varying dynamics of volatility in measuring CoVaR and highlights the need for an accurate assessment of tail heaviness and dependence structure in evaluating this systemic risk measure.