Smoothing sample extremes with dynamic models (Q2488453)

A state space model is considered in which the data possess a generalized extreme value distribution (GEV) conditional on a fixed state vector. The state dynamics is described by a Gaussian transition model. E.g., the location parameter of the GEV can be described by a first order random walk. A Monte Carlo technique based on resampling is proposed for evaluation of the smoothing (posterior) distribution of the state vector. An MCMC technique is proposed for goodness-of-fit inference. Analysis of athletic records and monthly maximum temperatures are considered as examples.