Title: Robust Estimators in Generalized Pareto Models

Abstract: We study robustness properties of several procedures for joint estimation of shape and scale in a generalized Pareto model. The estimators we primarily focus on, MBRE and OMSE, are one-step estimators distinguished as optimally-robust in the shrinking neighborhood setting, i.e.; they minimize the maximal bias, respectively, on a specific such neighborhood, the maximal mean squared error. For their initialization, we propose a particular Location-Dispersion (LD) estimator, kMedMAD, which matches the population median and kMAD (an asymmetric variant of the median of absolute deviations) against the empirical counterparts. These optimally-robust estimators are compared to maximum likelihood, skipped maximum likelihood, Cramer-von-Mises minimum distance, method of median, and Pickands estimators. To quantify their deviation from robust optimality, for each of these suboptimal estimators, we determine the finite sample breakdown point, the influence function, as well as the statistical accuracy measured by asymptotic bias, variance, and MSE - all evaluated uniformly on shrinking neighborhoods. These asymptotic findings are complemented by an extensive simulation study to assess their finite sample behavior.