Time-Varying Mixed-Frequency Vector Autoregressive Models *

To simultaneously consider mixed-frequency time series, their joint dynamics, and possible structural changes, we introduce a time-varying parameter mixedfrequency vector autoregressive model (TVP-MF-VAR). To keep our approach from becoming too complex in comparison with time-invariant MF-VARs, we limit time variation to the intercepts and error variances. We estimate our model using two techniques: an approximate one using forgetting factors in the prediction step of the Kalman lter and exponentially weighted moving averages (EWMA) for the error variances; and one based on exact Bayesian inference. Both approaches allow us to evaluate moderately large VARs (up to around 10 variables) in a recursive forecasting exercise in a reasonable amount of time. For eight variables in Germany, we examine the performance of our TVP-MF-VAR model variants and compare them to the MF-VAR of Schorfheide and Song (2015). Our empirical results demonstrate the feasibility and usefulness of our methods, even in the presence of only mild structural changes.