CLUSTER-BASED EXTREMAL INFERENCE FOR MULTIVARIATE TIME SERIES

Statistical procedures for inference on extremal properties of a multivariate time series are affected by the underlying extremal dependence structures. Many common time series models exhibit a clustering of extreme values and this will typically affect the variance of estimators which were built for i.i.d. observations. On the other hand, the behavior of quantities of interest, for example marginal distributions of the spectral tail process, is closely related to the overall dependence structure which we see in extremal clusters. We explore how this connection can be exploited to derive new estimators for extremal quantities. This is joint work with Holger Drees, University of Hamburg.