Second-Order and Extremal-Dependence Asymptotics of the Tail Value-at-Risk Sensitivity to an Additional Heavy-Tailed Cyber-Loss Component

François Xavier Ouedraogo *

Université Joseph KI-ZERBO, 03 BP 7021, Ouagadougou 04, Burkina Faso.

Frédéric Bere

École Normale Supérieure, 01 BP 1757, Ouagadougou 01, Burkina Faso.

Kiswendsida Mahamoudou Ouedraogo

Université Joseph KI-ZERBO, 03 BP 7021, Ouagadougou 04, Burkina Faso.

Pierre Clovis Nitiema

Université Thomas SANKARA, 04 BP 8938, Ouagadougou 04, Burkina Faso.

*Author to whom correspondence should be addressed.


Abstract

We study how the Tail Value-at-Risk, denoted TVaR, of a heavy-tailed cyber-loss portfolio reacts when a new digital component with loss \(S\) is added to an existing system with loss \(L\). When \(L\) and \(S\) are regularly varying with the same index \(\beta>1\), we obtain three results. First, under asymptotic negligibility of the joint tail, the capital surcharge \(\Delta \mathrm{TVaR}_\alpha=\operatorname{TVaR}_\alpha(L+S)-\operatorname{TVaR}_\alpha(L)\) is asymptotically equivalent to \(\left\{(1+k)^{1 / \beta}-1\right\} \operatorname{TVaR}_\alpha(L)\), where \(k\) is the tail-balance constant. Second, for independent components with second-order Pareto tails, the surcharge equals this first-order term plus an explicit constant that depends on the means of \(L\) and \(S\), up to \(o(1)\). This constant explains the slow convergence observed in simulations and is not negligible at regulatory confidence levels. Third, under multivariate regular variation, which allows extremal dependence with \(\lambda_U>0\), the factor becomes \(c_\mu^{1 / \beta}-1\), where \(c_\mu\) is computed from the exponent measure. We prove the sharp bounds \((1+k)^{1 / \beta}-1 \leq c_\mu^{1 / \beta}-1 \leq k^{1 / \beta}\), which are attained under asymptotic independence and under asymptotic comonotonicity respectively. We also correct and extend the aggregation results for fixed and random numbers of weakly dependent severities, and we exhibit a common-shock model of dependent severities that satisfies the required Kesten-type domination. A reproducible simulation study with Pareto and Burr margins and with Gaussian, FGM, Clayton, Gumbel and Student copulas confirms the results.

Keywords: Tail Value-at-Risk, cyber risk, regular variation, multivariate regular variation, second-order asymptotics, risk aggregation


How to Cite

Ouedraogo, François Xavier, Frédéric Bere, Kiswendsida Mahamoudou Ouedraogo, and Pierre Clovis Nitiema. 2026. “Second-Order and Extremal-Dependence Asymptotics of the Tail Value-at-Risk Sensitivity to an Additional Heavy-Tailed Cyber-Loss Component”. Journal of Advances in Mathematics and Computer Science 41 (11):105-20. https://doi.org/10.9734/jamcs/2026/v41i112221.

Downloads

Download data is not yet available.