Statistics and Data Science Seminar: Dr Christian Rohrbeck (University of Bath)

Extreme value theory provides a wide range of theoretically justified models for analysing the tail distribution of random vectors. One common approach is to study the marginal distribution and tail dependence structure separately based on a sample collected over time. In this talk, we consider the detection of non-stationarity in the tail dependence structure, a feature we increasingly observe in practice. Specifically, the talk will outline clustering and change point detection methods which are based on a summary of the tail dependence structure and can handle the following scenarios: (i) a random vector observed over time and (ii) random vectors observed at multiple spatial sites and over time.  The methods will be illustrated using environmental and EEG data.

This is joint work with: Euan McGonigle (Southampton), Paddy O’Toole (Bath), Matthew Pawley (ATASS) and Jordan Richards (Edinburgh).