Jonathan Atyeo

Bayesian modelling of multivariate extremes using a multivariate Gaussian tail model

Modelling multivariate extreme values has proven to be difficult in the past due to the difficulty in accounting for different forms of dependence at extreme levels. One model which has been developed recently is based on the marginal transformation of the tail of a multivariate Gaussian distribution (Bortot et al., 2000). This model however proved to be difficult to implement for high dimensional problems. We propose an alternative Bayesian Method for the modelling of Multivariate Extreme Values using the Gaussian Tail Model. The performance of the method is assessed by a simulation study, and the method is applied to extreme sea level data at Newlyn, a port in South West England. We show the Bayesian method to offer a very similar fit to the current inference procedures with the potential to be easily applied in higher dimensions.

Bortot, P., Coles, S.G. and Tawn, J.A. (2000) The multivariate Gaussian tail model: an application to oceanographic data. Appl. Statist. 49, 31-49.