Title: Bayesian nonparametric methods for analysing macroeconomic time series
Authors: Maria Kalli - University of Kent (United Kingdom) [presenting]
Abstract: The analysis of macroeconomic time series often involves the use of a vector autoregressive (VAR) model. VAR models provide a framework for the analysis of the complex joint dynamics present between macroeconomic series, but they have been criticised for their unrealistic assumptions (linearity, homoscedasticity, Gaussianity). We are going to describe how Bayesian non-parametric methods can be used to directly model the stationary and transition densities of such a multivariate system. This approach allows for nonlinearity in the conditional mean, heteroscedasticity in the conditional variance, and non-Gaussian innovations. It can also allow for non-stationary. Our empirical applications lie within the study of monetary policy and macro financial linkages within the aggregate economy. We find that the Bayesian nonparametric VAR (BayesNP-VAR) model predictively outperforms competing models.