Change in Regime and Markov Models

Abstract

In this paper we point out that using a two-state Markov chain to describe change in regime makes it difficult to interpret the model since there is a bias towards frequent shifts. However, by using a finite Markov chain with a transition matrix satisfying certain restrictions it is possible to circumvent the difficulty and at the same time use the established procedures for estimation and filtering. The methods are applied to a couple of time series from the Norwegian quarterly national accounts.