[Back]

M. Deistler:
"VAR Models and Mixed Frequency Data";
Talk: RSISE Seminar (Australian National University), Canberra, Australien (invited); 2013-11-14.

This lecture is concerned with identifiability of an underlying high frequency multivariate AR system from mixed frequency observations. The main results show that on a generic subset of the paramter space identifiability holds. We deal with AR systems with nonsingular and singular innovation variance matrices, the latter being important for dynamic factor models. Such models, which are used for high dimensional time series, and their properties will be discussed. We analyze the case of stock and flow variables. If we have identifiability, the parameters and thus all second moments of the output process at the high sampling frequency can be estimated consistently from mixed frequency data. Then linear least squares methods for forecasting, nowcasting and interpolation of nonobserved output variables can be applied.