A1238
Title: Estimation of generalized linear dynamic factor models: The single and the mixed frequency case
Authors: Manfred Deistler - Vienna University of Technology (Austria) [presenting]
Alexander Braumann - Vienna University of Technology (Austria)
Elisabeth Felsenstein - Vienna University of Technology (Austria)
Diego Fresoli - Vienna University of Technology (Austria)
Lukas Koelbl - Vienna University of Technology (Austria)
Abstract: We consider generalized linear dynamic factor models in a stationary framework; the observations are represented as the sum of two uncorrelated component processes: the so-called latent process, which is obtained from a dynamic linear transformation of a low dimensional dynamic factor process and which shows strong dependence of its components, and the noise process, which shows weak dependence of the components. The latent process is assumed to have a singular rational spectral density. In high dimensional time series often the univariate component series are available at different sampling frequencies. This is called mixed frequency observations. We discuss estimation, first for the single frequency case, consisting of the following steps: 1) Denoising, i.e. obtaining estimates of the latent variables and of the static factors from the observations, e.g. via a PCA or a Kalman filtering procedure described; 2) Estimation of a VAR for the static factors, e.g. by AIC and Yule Walker estimators; 3) Estimation of the dimension of the dynamic factors via a PCA on the innovations of the VAR. This procedure is generalized to the mixed frequency case. Several overall estimation procedures, both for the single and for the mixed frequency case, are evaluated and compared by Monte Carlo simulations.
