Rate of convergence for the generalized Pareto approximation of excess distributions (Q2745758)

The aim of the present note is to study the convergence rate for the approximation of a generalized Pareto distribution and to establish an optimality condition for choosing normalizing functions. The author considers the distribution function \(F\) in the maximum domain of attraction of an extreme value distribution with parameter \(\gamma\), \(H_\gamma\). Then \(F_\gamma\), the distribution function of the excesses over \(u\), is proved to converge to the distribution function \(G_{\gamma,\sigma(u)}\) of a generalized Pareto distribution, for an appropriate (positive) normalizing function \(\sigma(u)\), when \(u\) tends to the right end-point (upward boundary) of \(F\).NEWLINENEWLINENEWLINEThe main result of the paper is to study the rate of the uniform convergence to zero for the generalized Pareto approximation \(F_u(x) - G_{\gamma,\sigma(u)}(x + u - \alpha(u))\), when \(u\) tends to the right endpoint of \(F\), and \(\alpha(u)\) and \(\sigma(u)\) are two appropriate normalizing functions. The rate of convergence is proved to be optimal in the following sense: necessary and sufficient conditions are obtained ensuring that another choice of the two normalizing functions provides the same convergence rate.