An Evaluation of Alternative VAR Models for Forecasting Inflation

This paper provides a comparative analysis of several alternative vector autoregression (VAR) models for forecasting inflation. These include classical and Bayesian framework. Although traditional BVAR models can improve UVAR models forecasts by using extra information as priors, they can not be used to resolve a mixed drift case, which is common in most of the macroeconomic forecasting models. Hence I apply the Bewley transformation for the re-parameterization of the VAR to estimate drift parameters directly by using instrumental variables. As an alternative the g-prior is considered. I assess the performance of the different specifications of BVAR models by one-step ahead forecasts using real data and Monte Carlo experiments.