A BVAR model for forecasting Paraguay’s inflation rate in turbulent macroeconomic enviroments

Abstract

In this research I explore the methodology of Bayesian autoregressive methods to forecast inflation and other macroeconomic time series of interest. I estimate a Bayesian vector of autoregressive model to forecast inflation, GDP and the interest rate of Paraguay taking as main approach the Minnesota prior methodology developed by R.B. Litterman (1984). The main out of sample accuracy statistics, the RMSFE and U-Theil statistic results show that in the 75% of the subsamples of forecast characterized as turbulent macro environments, Bayesian specifications outperform traditional VAR models in terms of accuracy. When using quarterly data Bayesian techniques deliver also more accurate forecasts than VAR models ones.