Exceedances over high thresholds: a guide to threshold selection (Q1294765): Difference between revisions
From MaRDI portal
Changed an Item |
Set profile property. |
||
Property / MaRDI profile type | |||
Property / MaRDI profile type: MaRDI publication profile / rank | |||
Normal rank |
Revision as of 03:49, 5 March 2024
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Exceedances over high thresholds: a guide to threshold selection |
scientific article |
Statements
Exceedances over high thresholds: a guide to threshold selection (English)
0 references
10 August 1999
0 references
This article focuses on two problems connected with modeling of exceedance over high thresholds: (a) approximation of distributions of exceedances over a given threshold, (b) selection of an appropriate threshold. The author proposes to model the distribution for \(n\) exceedances \(x_1,\dots,x_n\) over a given threshold by the generalized Pareto distribution (GPD) with density \(h_{\alpha,k}(x)=(1/\alpha)(1-kx/\alpha)^{(1-k)/k}\), where \(0\leq x\leq \infty\) for \(k\leq 0\) and \(0\leq x\leq \alpha/k\) for \(k>0\). In practice, the parameters \(\alpha\) and possibly \(k\) must be estimated from the data set. The author's approach involves optimal bias robust estimators (OBRE) for \((\alpha,k)\) obtained by modification of the algorithm proposed by \textit{M.-P. Victoria-Feser} and \textit{E. Ronchetti} [Can. J. Stat. 22, No.2, 247-258 (1994; Zbl 0801.62099)]. This robust procedure also assigns weights to data points. These weights are used to assess the validity of the GPD model for exceedances over a chosen threshold and can guide threshold selection: one can increase the threshold and reduce the number of data points until all weights are close to one. Such method is used in analysis of two real data sets from hydrology. Goodness of fit tests for GPD are also discussed.
0 references
exceedances over thresholds
0 references
generalized Pareto distribution
0 references
optimal bias robust estimators
0 references
M-estimators
0 references
goodness of fit tests
0 references