Time series factor analysis model: Factors generated by autoregression and moving average process

The dynamic factor analysis model (Molenaar, 1985) which is one of the generalizations of the p-technique factor analysis model, can explain the lagged covariance structure among observed variables. Hershberger, Corneal, and Molenaar (1994) showed that the dynamic factor model can be easily evaluated within a structural equation modeling (SEM) program such as LISREL. In this paper, an alternative time series model containing the latent factors which are generated by the autoregression and moving average (ARMA) process is proposed. This model, which has been named the time series factor analysis model, can also be easily evaluated with a SEM program. The application of this model to the leading index, the coincident index and the lagging index of the Japanese economy revealed a latent common factor series generated by considerable autoregression.