I introduce a factor structure on the parameters of a Bayesian TVP-VAR to reduce the dimension of the model's state space. To further limit the scope of over-fitting the estimation of the factor loadings uses a new generation of shrinkage priors. A Monte Carlo study illustrates the ability of the proposed sampler to well distinguish between time-varying and constant parameters. In an application with Swiss data the model proves useful to capture changes in the economy's dynamics due to the lower bound on nominal interest rates.
Bayesian VAR time-varying parameter dimension reduction