Bayesian Inference for Markov-switching Skewed Autoregressive Models

Abstract

We examine Markov-switching autoregressive models where the commonly used Gaussian assumption for disturbances is replaced with a skew-normal distribution. This allows us to detect regime changes not only in the mean and the variance of a specified time series, but also in its skewness. A Bayesian framework is developed based on Markov chain Monte Carlo sampling. Our informative prior distributions lead to closed-form full conditional posterior distributions, whose sampling can be efficiently conducted within a Gibbs sampling scheme. The usefulness of the methodology is illustrated with a real-data example from U.S. stock markets.