Extremal Dependence of Time Series

Holger Drees
University of Hamburg

By now the statistical analysis of independent extreme risks is well developed. In contrast, the understanding of the statistical inference on the dependence between extreme events over time is rather patchy, despite the fact that the total risk is often strongly influenced by the clustering behaviour of extremal events. For example, the risk of huge accumulated losses in a financial investment is larger if days with large losses tend to occur in clusters.

We present a systematic approach to the analysis of the extremal serial dependence of such time series using empirical process theory. Particular attention is turned to the bias which is known to often cause serious misjudgment of the clustering behaviour and hence of the total risk.