Dynamic Factor Analysis with ARMA Factors

In this paper we present a new approach to the specification of dynamic factor models. Our model has three advantages over existing work. Firstly, it is based on a minimal-dimension state-space representation giving some gain in computational efficiency over existing methods. Secondly, it easily accommodates hypothesis tests about the order of the factor-filter. Thirdly, by allowing the factor-filter to have a common polynomial factor, ARMA-factor models may be estimated with little extra computational expense over the AR- factor case. We illustrate the use of our model with an application to business cycle analysis.