Generalized Linear Dynamic Factor Models: The Single and the Mixed Frequency Case

• Activity Number: 385
• Type: Topic Contributed
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Business and Economic Statistics Section
• Author(s): Manfred Deistler*+ and Brian D.O. Anderson and Elisabeth Felsenstein and Alexander Filler and Bernd Funovits and Mohsen Zamani
• Companies: Vienna University of Technology and Australian National University and Vienna University of Technology and Uniqa and University of Vienna and Australian National University
• Keywords: High Dimensional Time Series ; Generalized Dynamic Factor Models ; Singular AR Systems ; (Generalized) Yule-Walker Equations ; Mixed Frequency

Abstract

We consider generalized linear dynamic factor models. These models have been developed recently and they are used for high dimensional time series in order to overcome the ''curse of dimensionality''. We present a structure theory with emphasis on the zeroless case, which is generic in the setting considered. Modelling of the latent variables is decomposed into two steps, first the transformation of the latent variables to static factors by a linear static transformation. Then, in the second step, modelling of the static factors as a possibly singular autoregressive process. The (generalized) Yule-Walker equations are used for parameter estimation. The Yule-Walker equations do not necessarily have a unique solution in the singular case, and the resulting complexities are examined with a view to find a stable and coprime system. Finally, some preliminary results for the mixed frequency case are presented.