THE EVALUATION OF BAYESIAN DENSITY FORECASTS FOR THE SINGAPORE ECONOMY

Abstract

Since its debut in the early 80's, the Bayesian Vector Autoregression (BVAR) model has proven to be a very useful tool for econometric forecasting. It utilizes Bayesian techniques to restrict a Vector Autoregression's (VAR) coefficients to a prior belief that the behavior of an economic variable is a random walk around an unknown deterministic component. This paper considers the justification for the Bayesian approach, its implementation and the performance of the model. A simple BVAR model is built within local specifications and its density forecasts of Singapore's quarterly Gross Domestic Product (GDP) are evaluated. The analysis reveals interesting features of the density forecasts in relation to realized GDP. On most occasions, the BVAR model underestimated the GDP forecasts. This is contrary to the BVAR's otherwise impressive track record elsewhere. Although the model is able to capture the volatility dynamics operative in the process, it is still inferior to a random walk process. This suggests that the success of previous BVAR models have been exaggerated .