Seminar: Raphael Huser

Gaussian scale mixtures are constructed as Gaussian processes with a random variance. They have non-Gaussian marginals and can exhibit asymptotic dependence unlike Gaussian processes, which are always asymptotically independent except for perfect dependence. Motivated by the analysis of spatial extremes, we propose a flexible but parsimonious Gaussian scale mixture copula model, which smoothly interpolates from asymptotic dependence to independence. We show how this new model can be fitted to high threshold exceedances using a censored likelihood approach, and we demonstrate that it provides valuable information about tail characteristics. The methodology will then be illustrated with an application to wind speed data in the Pacific Northwest, United States, showing that it adequately captures the data's extremal properties.

 

Collaborators: Thomas Opitz and Emeric Thibaud