A Bayesian Markov-Switching for Smooth Modelling of Extreme Value Distributions

Markov-switching models are attractive for analysing time series that exhibit different stochastic processes along different periods, and where the regime-switching is controlled by an unobservable Markovian process. Model flexibility can be enhanced considering regime-specific distributions, whose distributional parameters may be modelled using smooth functions of covariates. Here, we propose a two-state Markov-switching model using full Bayesian inference and accounting for extreme value modelling. The proposal is illustrated by analysing energy prices.