Simulation of certain multivariate generalized Pareto distributions (Q1003302)

An algorithm is proposed for a simulation from a generalized pareto distribution (GPD) of logistic type, i.e., for the CDF \( w_\lambda(x_1,\dots,x_d)=1-\left(\sum_{i=1}^d(-x_i)^\lambda\right)^{1/\lambda} \) for \(x_i<0\). It is based on the Shi transformation. A version of the rejection method is described for simulations from GPDs with bounded Pickands densities in a neighbourhood of the origin. It is proposed to simulate unconditionally GPD distributed random vectors using the POT-stability property of GPDs. A test for a hypothesis \(\lambda=\lambda_0\) is considered in which the distribution of the test statistics is approximated via simulations. An application to a hydrological data set is described.