CMStatistics 2020: Start Registration
View Submission - CFE
Title: G-identifiability for multivariate AR systems and mixed frequency data: REMIS for the unit root case Authors:  Philipp Gersing - Vienna University of Technology (Austria) [presenting]
Leopold Soegner - Institute for Advanced Studies (Austria)
Manfred Deistler - Vienna University of Technology (Austria)
Abstract: The identification of the model parameters for data observed at mixed frequencies in a Johansen-type error correction model is investigated. Thus, the generic identifiability results for multivariate AR Systems and mixed frequency data are extended to the non-stationary case. We call this approach REMIS (REtrieval from MIxed Sampling frequency). For the blocked process of time-series observed, a canonical state-space representation is derived. By applying the projection approach recently developed in another paper of the authors, we obtain a system in prediction error form. For a regular high-frequency system, the blocked process is regular, and the system in prediction error form is minimal. Given the second moments of the variables observed under mixed frequency, the parameters of the high-frequency system are generically identified from these second moments. This result is established for slow stock variables as well as for variables obtained from a linear aggregation scheme.