Bayesian Vector Autoregressive Procedure for Forecasting Swiss Economy

This book adopts a methodology for forecasting real GDP and inflation growth in Switzerland. Introduced by Litterman (1986), this study builds forecast models for the Swiss economy. Firstly, autodistributed lagged models (ARDL) are computed, followed by the framework of Bayesian models. Bayesian vector autoregressive models (BVARs) strongly rely on the VAR framework, however they allow a better exploitation of all the information available. Using the data from 1980, out-of-sample forecasts have been computed from 2000 to 2014. Suggesting four categories that variables are grouped into, this study finds that Bayesian VAR models improve forecast errors, principally for inflation. An extension of the model is performed using foreign data, which further reduces forecast errors. Asset prices are found to contain valuable information in forecasting real GDP and, particularly in predicting inflation growth. However, BVARs cannot substitute for a complete structural method for economic policies analysis. Nevertheless, these models tend to produce good forecasts performance and thus, should be used as complementary benchmark forecasting models for the Swiss National Bank.