New index of coincident indicators: A multivariate Markov switching factor model approach

This paper extends the univariate Markov switching unobserved component model to the multivariate Markov switching factor model of coincident economic indicators. An approximate ML method is developed to estimate the model. The extracted Markov switching factor may be interpreted as the coincident index and is comparable to the Stock-Watson index which differs only by the Markov switching component. Using four constituent series of the DOC coincident index for the period January 1960 to June 1992, the proposed model generates recessionary and expansionary periods which are remarkably consistent with the NBER chronology of the business cycles from the noisy monthly data. This paper also finds that duration dependence does matter.