Multivariate extreme value theory exhibits a complex dependence structure which is often described by underlying physical or economical phenomena. This means that observations of data of extremes are usually driven by fewer actual events than number of observations taken. Reducing dimensionality allows to approximately recover this hidden causal structure from the data. However classical statistical methodology is not directly applicable to the setting of extremes, thus there is a need for new methods.

The research project will focus on developing an approach to such a new method, then study its statistical behavior and properties. Furthermore we want to evaluate its performance on datasets from meteorology and finance, with the goal to provide useful new tools for studying extreme events in applications.