Skip to contents

Prediction error C-criterion threshold selection method

Source: R/thselect.R
thselect.pec.Rd

This function computes the non-robust Pareto prediction error of Dupuis and Victoria-Feser (2003), termed C-criterion, for the Hill estimator of the shape parameter. The threshold returned is the value of the threshold, taken from order statistics, that minimizes the average prediction error.

Usage

thselect.pec(xdat, kmax)

Arguments

xdat

vector of observations

kmax

maximum number of order statistics to consider. Default to sample size if left unspecified.

Value

a list with the number of exceedances k, the chosen threshold thresh0 and the corresponding Hill estimator shape estimate shape.

References

Dupuis, D.J. and M.-P. Victoria-Feser (2003). A Prediction Error Criterion for Choosing the Lower Quantile in Pareto Index Estimation, University of Geneva, technical report, https://archive-ouverte.unige.ch/unige:5789.