A1637
Title: REMIS (Retrieval from Mixed Sampling Frequency) for generalised dynamic factor models
Authors: Philipp Gersing - University of Vienna (Austria)
Manfred Deistler - Vienna University of Technology (Austria)
Leopold Soegner - Institute for Advanced Studies (Austria)
Christoph Rust - WU Vienna University of Economics and Business and University of Regensburg (Germany) [presenting]
Abstract: Representation and estimation theory for Generalised Dynamic Factor Models with mixed frequency data is provided. We suppose a GDFM for the underlying high-frequency processes where the spectrum of the common component is rational. We look at the structure of the stacked process running on the slow frequency sampling rate containing all observable outputs. With this approach, we aim to build ``information efficient'' methods for denoising, parameter estimation and factor extraction with observations under mixed sampling frequency. We prove that the blocked process is again a GDFM with rational spectrum in the common component and define a canonical state-space realisation for the blocked common component. We analyse the relationship between the dynamic and static factor spaces of the underlying high- and low-frequency processes and the blocked process and show under which conditions the high-frequency dynamic factor space can be retrieved from mixed frequency data. Besides forecasting, imputation of the slow/aggregated variables is also a possible application of our results. We demonstrate our methods in an empirical application on forecasting macroeconomic variables with a high dimensional panel of quarterly and monthly observed US-macroeconomic time series.
