A Dynamic Factor Model
Yu‐Pin Hu, Rouh‐Jane Chou · SSRN Electronic Journal · 2003
_This paper attempts to find the possibilities of simplifying a multiple time series. We consider a dynamic factor model, Z_t=[sum]_i=1-super-m Lambda inf>iX_t-i+G epsilon _t, where Z_t is a k-dimensional Gaussian stationary time series, X_t is an unobservable r-dimensional factor series (K>=r) and epsilon _t is a (K - r) -dimensional white noise series, fully independent of {X_t}. The model is for reducing the dimension of time effect of the series, because the error term is independent of time. This article starts with an example showing that two independent AR(1) series can be generated by one series under this model. Mathematical properties of the model are studied. Its relations with other time series models are also discussed. Copyright 2003 Blackwell Publishing Ltd.