A0379
Title: An observation-driven mixed-frequency VAR model
Authors: Heiner Mikosch - ETH Zurich (Switzerland) [presenting]
Maurizio Daniele - ETH Zürich, KOF Swiss Economic Institute (Switzerland)
Stefan Neuwirth - ETH Zurich - KOF Swiss Economic Institute (Switzerland)
Abstract: The aim is to introduce an observation-driven mixed-frequency vector autoregression (MFVAR) model. A general MFVAR framework is developed based on vector stacking, and it is shown how the model can be reformulated into a closed-form representation that permits analytical estimation. Additionally, a Bayesian normal prior is derived to enable shrinkage estimation of the MFVAR. Monte Carlo simulations and empirical applications demonstrate that the estimators are computationally efficient, even in high-dimensional settings. The observation-driven MFVAR achieves comparable or superior out-of-sample forecasting accuracy relative to state-space MFVARs, while requiring only a fraction of the computation time.
