Abstract: We derive a collection of reference prior distributions for a Bayesian analysis under the three-parameter generalized extreme value (GEV) distribution. These priors are based on an established formal definition of oninformativeness. They depend on the ordering of the three parameters, and we show that the GEV is unusual in that some orderings fail to yield proper posteriors for any sample size. We also consider a reparametrization that explicitly regards a return level estimation, which is the most common goal of a GEV analysis, to be the most important inferential task. We investigate the properties of the derived priors using a simulation, and apply the priors to an analysis of a fire threat index in California.

Key words and phrases: Noninformative priors, objective Bayes, posterior normality.