Realistic Confidence Intervals for Non-Stationary Extreme Value Statistics.

ExtremeCI

The ExtremeCI R package provides versatile algorithms to efficently infer confidence intervals for extreme value statistics, using extreme value theory extrapolation and the profile likelihood. These intervals capture the uncertainty spread of extreme estimates more realistically than most alternative methods, especially for return levels (high quantiles) and for the shape parameter, whose uncertainties are typically highly asymmetric (Coles, 2001). For return levels and extreme quantiles, the profile likelihood method requires reparametrization of the likelihood function. With nonstationary models, this reparametrization is nontrivial and requires repetition for each local interval, which was not possible with alternative software. This package provides a framework to construct both stationary and nonstationary models with novel confidence endpoint search procedures based on binary search, which do no require a prespecified range. The CIs can also be inferred from weighted samples (work in progress). This package is motivated by Zeder et al. (2023) and by Pasche et al. (2026).

Installation

To install the development version of ExtremeCI from R, run

# install.packages("devtools")
devtools::install_github("opasche/ExtremeCI")

Package created by Olivier C. PASCHE
Research Institute for Statistics and Information Science,
University of Geneva (CH), 2025.
Supported by the Swiss National Science Foundation.