New Estimators for the Extremal Index and Other Cluster Characteristics

摘要

New methods for identifying clusters of extreme values are proposed that exploit additional knowledge of the trajectory of the process around extreme events. These methods lead directly to new estimators of cluster characteristics, such as the extremal index, which are shown to have both substantially reduced bias and greater insensitivity to cluster identification parameters than existing methods. The methods are illustrated for a range of theoretical examples and by applications to environmental and financial time series.