A0542
Title: An observation-driven mixed-frequency VAR model with closed-form solution
Authors: Maurizio Daniele - ETH Zürich, KOF Swiss Economic Institute (Switzerland)
Heiner Mikosch - ETH Zurich (Switzerland) [presenting]
Stefan Neuwirth - ETH Zurich - KOF Swiss Economic Institute (Switzerland)
Abstract: A mixed-frequency VAR model is developed using a stacked vector approach. It is then shown how to transform the model from its stacked vector form into a form in which the model can be estimated analytically by multivariate least squares. Also, a Bayesian normal prior is developed for analytical shrinkage estimation of the model. The mixed-frequency VAR does not involve the modelling of latent variables and falls in the class of observation-driven models. It contrasts with previous literature that proposed paramater-driven mixed-frequency VARs which rely on a latent variable state space framework. Monte Carlo simulations yield that both the multivariate least squares and the Bayesian estimator of mixed-frequency VAR are consistent and fast. In an empirical out-of-sample forecasting exercise with quarterly, monthly, and weekly macroeconomic and financial data, the mixed-frequency VAR is found to outperform a standard quarterly-frequency VAR.
