Portfolio risk assessment using multivariate extreme value methods

This paper presents a model for the joint distribution of a portfolio by inferring extreme movements in financial markets. The core of our proposal is a statistical model for the tail of the joint distribution that attempts to capture accurately the data generating process through an extremal modelling for the univariate margins and for the multivariate dependence structure. The model addresses several features of financial returns by encompassing methods from both econometrics and extreme value theory, and hence, taking into account the asymmetric behaviour of extreme negative and positive returns, and the heterogeneous temporal as well as cross-sectional lead-lag extremal dependencies among portfolio constituents. The model facilitates scenario generation for future returns through extrapolation beyond the empirical observations upon which portfolio risk assessment is based. We provide empirical evidence for the proposed model by an application to stock market returns for the G5 economies.