B0562
Title: Huesler-Reiss extremal graphical models
Authors: Alexandros Grosdos - University of Augsburg (Germany) [presenting]
Frank Roettger - TU Eindhoven (Netherlands)
Jane Coons - Oxford University (United Kingdom)
Abstract: Extreme value theory aims to explain statistical phenomena whose occurrence probability is very low, such as extreme weather phenomena or financial crises. Extremal graphical models are an exciting and rapidly growing new direction in graphical modelling that allows the study of extremes in higher dimensions. The algebraic structures of the Huesler-Reiss distributions underlying these models are emphasized. The connection between the Gaussian ideal of an undirected model and the corresponding extremal counterpart is established. Moreover, the maximum likelihood degree is studied for these models and contrasted with the Gaussian case. Furthermore, coloured graphical models are used in this setting and the results using real-world data are illustrated.
