Model selection in sparse high-dimensional vine copula models with application to portfolio risk

Extreme event propagation using counterfactual theory and vine copulas

“…This allows capturing tail dependence across the marginals themselves. We control the fitting performance and the number of parameters included in this second layer using information criteria such as AIC, BIC or sparse modified BIC for vines (mBICV) (Nagler, Bumann and Czado, 2018).…”

“…Modelling extremal behaviour accurately requires flexible dependence structures as formalised by the asymptotic dependence (Wadsworth et al, 2017;Huser and Wadsworth, 2019), regularly varying distributions (Mikosch, 2006;Weng and Zhang, 2012) or sparse structures for high-dimensions (Engelke and Ivanovs, 2021;Engelke and Hitz, 2020). Although the EEP framework remains model-agnostic in its current formulation, vine copulas (Bedford and Cooke, 2002)-a family of hierarchical pairwise graphical copulas-are a balance between adjustable extremal properties (Joe, Li and Nikoloulopoulos, 2010), the ability to capture non-linear relationships (Gräler, 2014;Erhardt, Czado and Schepsmeier, 2015) and scalability capabilities (Nagler, Bumann and Czado, 2018;Nagler, Krüger and Min, 2020). It also offers a conditional sampling mechanism (Bevacqua et al, 2017, App.…”