Jean-François Dupuy

Titre : "Bayesian estimation of the tail index and extreme quantiles in a semi-parametric heavy-tailed distribution" 

Résumé:

 "Bayesian estimation of the tail index of a semi-parametric heavy-tailed distribution is addressed when data are randomly right-censored. 
Maximum a posteriori and mean posterior estimators are constructed for various prior distributions of the tail index. Convergence of the posterior distribution of the 
tail index to a Gaussian distribution is established. Finite-sample properties of the proposed estimators are investigated via simulations. Tail index estimation requires
 selecting an appropriate threshold for constructing relative excesses. A Monte Carlo procedure is proposed for tackling this issue. 
An estimator of extreme quantiles is also proposed, based on the achieved tail index estimator. Finally, all proposed estimators are illustrated on a medical dataset."