Résumé
The aim of this thesis is the estimation of a high quantile function. We consider a sample of n random variables independent and identically distributed on a bounded support. Our goal is to estimate the extreme quantile of the conditional cumulative distribution function of Y given X=x. To construct this estimator, we first propose a new estimator of the extreme value index. To construct our estimators, we only use the number of points exceeding random thresholds. We show that the estimators are weakly consistent and we make a simulation study.