Title: Diagnostics for scale-shape models based on robust statistics
Authors: Peter Ruckdeschel - University of Oldenburg (Germany) [presenting]
Nataliya Horbenko - KfW Bankengruppe (Germany)
Matthias Kohl - Hochschule Furtwangen (Germany)
Abstract: Infrastructure and robustness-based diagnostics for scale-shape models available in R package RobExtremes (on CRAN since 08/2018) are presented. These scale-shape models cover amongst others, Pareto, generalized Pareto, generalized extreme value, gamma, and Weibull distributions. Lacking equivariance in the shape, these models call for refined computational techniques to achieve acceptable timings. RobExtremes provides speeded up optimally-robust estimators for these models together with high-breakdown starting estimators. Of course, MLEs and Minimum-Distance-Estimators (MDEs) are also available through R package distrMod. Diagnostics from R package ismev are available for our model fits, as well as the robustness-based diagnostic plots from R package RobAStBase such as outylingness plots, influence curve plots, information plots. In addition, we provide non-parametric confidence bands for qqplots and return level plots. We demonstrate these diagnostics with data from hydrology, hospital length of stay, and finance.