Accounting for the threshold uncertainity in extreme value estimation (Q2463692)

It is well known that the distribution of exceedance over a high threshold can be approximated by a generalized Pareto distribution (GDP). The aim of this paper is to determine an appropriate threshold for such approximations. The data is assumed to be an i.i.d. sample with a PDF which is peacewise constant below some threshold \(\alpha\) and is GDP over \(\alpha\). A Bayesian approach is used for the estimation of the parameters. Results of simulations and application to river Nidd data are presented.