A1287
Title: Frequency-domain estimation of dynamic factor models
Authors: Giovanni Motta - Columbia University (United States) [presenting]
Michael Eichler - Maastricht University (Netherlands)
Abstract: The generalized dynamic factor model has become very popular in the theory and practice of large panels of time series data. The asymptotic properties of the corresponding estimators have been studied previously. Those estimators rely on Brillinger's dynamic principal components and thus involve two-sided filters, which leads to rather poor forecasting performances. A more recent study derives the asymptotic properties of a semi-parametric estimator of loadings and common shocks based on one-sided filters. However, compared to the model in the previous study, the latter model relies on the additional assumptions that the common components have rational spectral density and admit a finite autoregressive representation. Moreover, the estimator involves several time- and frequency-domain estimation steps. We propose a novel approach to estimate the common components and the common shocks directly in the frequency domain. Our approach does not rely on the assumption of rational spectral density, and our estimation method is computationally simpler and faster.
