Estimating the conditional extreme-value index under random right-censoring

In extreme value theory, the extreme-value index is a parameter that controls the behavior of a cumulative distribution function in its right tail. Estimating this parameter is thus the first step when tackling a number of problems related to extreme events. In this paper, we introduce an estimator of the extreme-value index in the presence of a random covariate when the response variable is right-censored, whether its conditional distribution belongs to the Fréchet, Weibull or Gumbel domain of attraction. The pointwise weak consistency and asymptotic normality of the proposed estimator are established. Some illustrations on simulations are provided and we showcase the estimator on a real set of medical data.