Identification and normalization in Markov switching models of \"business cycles\"

Recent work by Hamilton, Waggoner and Zha (2004) has demonstrated the importance of identification and normalization in econometric models. In this paper, we use the popular class of two-state Markov switching models to illustrate the consequences of alternative identification schemes for empirical analysis of business cycles. A defining feature of (classical) recessions is that economic activity declines on average. Somewhat surprisingly however, this property has been ignored in most published work that uses Markov switching models to study business cycles. We demonstrate that this matters: inferences from Markov switching models can be dramatically affected by whether or not average growth in the 'low state' is required to be negative, rather than simply below trend. Although such a restriction may not be appropriate in all applications, the difference is crucial if one wants to draw conclusions about 'recessions' based on the estimated model parameters.