Multivariate Hawkes-Driven Dynamic Vine Copulas for Jump Clustering and Systemic Risk Measurement in Financial Markets
- 1Jesse H. Jones Graduate School of Business, Rice University, Houston, TX 77005, USA
- 2School of Economics and Management, Xinjiang University, Urumqi 830046, China
Abstract
Extreme co-movements in asset returns tend to arrive in clusters and to propagate across markets, a feature that static dependence models and diffusion-based volatility models fail to capture jointly. This paper develops a dynamic vine copula framework in which the dependence parameters of each bivariate building block are driven by the intensities of a mutually exciting multivariate Hawkes process fitted to the jump arrivals of the underlying return series. The construction yields a closed-form expression for conditional joint density, preserves the flexibility of regular vines in high dimensions, and links tail dependence directly to the self- and cross-excitation structure of jumps. We establish stationarity and ergodicity conditions for the coupled Hawkes–copula system, we derive the asymptotic properties of a two-stage maximum likelihood estimator, and we propose a Monte Carlo procedure for computing Conditional Value-at-Risk (CoVaR) and Marginal Expected Shortfall under the fitted dependence dynamics. An empirical study based on the daily returns of 11 S&P 500 sector indices and 6 major international equity indices over 2005–2025 shows that the proposed model produces significantly better out-of-sample density forecasts and more accurate 5% tail-risk coverage than time-varying Gaussian, Student-t, and DCC-based copula benchmarks and quantile-regression CoVaR, with the largest gains observed during the 2008 financial crisis and the 2020 pandemic sell-off. The estimated excitation matrix further identifies the financial and energy sectors as the dominant transmitters of jump contagion, a classification that is robust to the jump threshold, to the shape of the excitation kernel, to sub-sample choice and, in the international application, to corrections for non-synchronous trading. The framework offers a tractable tool for regulators and portfolio managers to monitor the build-up of systemic risk in real time.
Related
- Robust model-based clustering via mixtures of multivariate pseudo-Voigt distributions
- Latent-Class Poisson-Lognormal Mixed Models for Clustering Multivariate Longitudinal Count Data
- Salient Bayesian Clustering for Proportional Data via a Multivariate Beta Mixture Model
- Twoblock clustering trees with coskewness-based dimension reduction: recovering piecewise multivariate linear regimes
- Finite Mixtures of Generalized Estimating Equations for Clustering Multivariate Correlated Outcomes
BibTeX
@article{yao2026multivariate,
title = {{Multivariate Hawkes-Driven Dynamic Vine Copulas for Jump Clustering and Systemic Risk Measurement in Financial Markets}},
author = {Jingqi Yao and Xuechao Xia},
journal = {Mathematics},
year = {2026},
month = oct,
doi = {10.3390/math14193587},
url = {https://www.mdpi.com/2227-7390/14/19/3587},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Multivariate Hawkes-Driven Dynamic Vine Copulas for Jump Clustering and Systemic Risk Measurement in Financial Markets
AU - Jingqi Yao
AU - Xuechao Xia
PY - 2026
DA - 2026/10/02
JO - Mathematics
DO - 10.3390/math14193587
UR - https://www.mdpi.com/2227-7390/14/19/3587
AB - Extreme co-movements in asset returns tend to arrive in clusters and to propagate across markets, a feature that static dependence models and diffusion-based volatility models fail to capture jointly. This paper develops a dynamic vine copula framework in which the dependence parameters of each bivariate building block are driven by the intensities of a mutually exciting multivariate Hawkes process fitted to the jump arrivals of the underlying return series. The construction yields a closed-form expression for conditional joint density, preserves the flexibility of regular vines in high dimensions, and links tail dependence directly to the self- and cross-excitation structure of jumps. We establish stationarity and ergodicity conditions for the coupled Hawkes–copula system, we derive the asymptotic properties of a two-stage maximum likelihood estimator, and we propose a Monte Carlo procedure for computing Conditional Value-at-Risk (CoVaR) and Marginal Expected Shortfall under the fitted dependence dynamics. An empirical study based on the daily returns of 11 S&P 500 sector indices and 6 major international equity indices over 2005–2025 shows that the proposed model produces significantly better out-of-sample density forecasts and more accurate 5% tail-risk coverage than time-varying Gaussian, Student-t, and DCC-based copula benchmarks and quantile-regression CoVaR, with the largest gains observed during the 2008 financial crisis and the 2020 pandemic sell-off. The estimated excitation matrix further identifies the financial and energy sectors as the dominant transmitters of jump contagion, a classification that is robust to the jump threshold, to the shape of the excitation kernel, to sub-sample choice and, in the international application, to corrections for non-synchronous trading. The framework offers a tractable tool for regulators and portfolio managers to monitor the build-up of systemic risk in real time.
SN - 2227-7390
ER -