University of Newcastle upon Tyne

School of Mathematics and Statistics

Statistics Seminars 2005-2006

21 October 2005, L401, 2:00pm

Dave Walshaw

Declustering procedures for extremes: a misleading waste of effort

 Abstract

Current accepted practice for analysing extremes of time series would be to identify clusters of large values and then fit a limiting extreme value model to the cluster maxima, thereby working with a set of approximately independent values.

Here we demonstrate that such procedures lead to systematic bias in the estimation of return levels. If, on the other hand, we take the much simpler approach embodied by fitting the same model to all large values, then the bias is negligible. We are left with the problem that the set of extremes being analysed is no longer independent. We suggest a pragmatic application of existing methodology to make appropriate adjustments to standard errors and confidence intervals. We advocate the resulting procedure as being both simpler to apply, and giving better results, than current practice.
 

Return to timetable for 2005-06 seminars

School of Mathematics and Statistics Homepage