ABSTRACT
Dynamic factor analysis (DFA) is a latent variable modeling technique that joins together the power of multivariate time series analysis and factor analysis. In short, DFA refers to the factor analysis of multivariate time series, the purpose of which is to understand the common, latent sources of change in a set of observed variables across time. Factor analytic approaches to serially dependent data are not new. Earlier approaches were discussed by, among others, Ahamad (1967), Priestly, Subba Rao, and Tong (1973), Brillinger (1975), Geweke (1977), Geweke and Singleton (1981), Sims (1981), Box and Tiao (1977), and Velu, Reinsel, and Wiehern (1986). What distinguishes the current formulation of DFA from other formulations is its acknowledged dependence on the factor analytic tradition embodied by the work of Cattell (1952). This tradition views factor analysis not so much as a data reduction technique but as a method for understanding the covariance structure underlying the data: The factors or components extracted from data do not simply summarize the data, they explain it in some meaningful way.
