Operational risk quantification traditionally assumes independence across categories, underestimating stress-period losses. We challenge this paradigm by developing a unified framework addressing discrete-continuous loss data, heavy-tailed severity, and complex risk dependence. Analyzing 1.99 million operational loss events from a major Brazilian bank (2020–2024), we combine Extreme Value Theory (EVT) for severity tails, spliced distributions, and Regular Vine (R-Vine) copulas with jittering for frequency dependence. Our results show dependence increases economic capital at the 99.9% level by 42.87% over five years. Backtesting reveals the independence model systematically fails during stress periods, while the Vine copula model maintains coverage. These findings confirm that presumed diversification benefits are often illusory, though negative dependencies can occasionally reduce aggregate risk.
Comissão Organizadora
Comissão Científica
Ferramenta completa de submissão, avaliação e publicação de anais para eventos científicos.
