A predictive approach to tail probability estimation (Q1409824): Difference between revisions

The problem of high quantiles estimation by an i.i.d. sample \(X_1\),\dots, \(X_n\) is considered. The authors propose a Bayesian approach where \(\Pr\{X-u \mid X>u\}\) is supposed to be a generalized Pareto distribution and the threshold \(u\) is a Bayesian parameter with some prior distribution. A five levels hierarchical Bayesian model is constructed. MCMC techniques with Gibbs sampler are used to implement the estimation. Results of simulations and a case study of ground-level ozone data are presented.