Quasi-conjugate Bayes estimates for GPD parameters and application to heavy tails modelling (Q2488471)

Bayesian estimation of the parameters \(\alpha>0\) and \(\beta>0\) of a Generalized Pareto Distribution (GPD) with density \[ f_{\alpha,\beta}(y)=(\alpha/\beta) \left( 1+y/\beta \right)^{-\alpha-1}, \quad y\geq 0, \] by an i.i.d. sample is considered. A quasi conjugate prior is constructed based on the representation of the GPD PDF as a mixture of exponential densities with Gamma mixing distribution. A Markov chain Monte Carlo procedure with Gibbs sampler for posterior calculations is described. Results of simulations and applications to real data sets are considered.