Acta Univ. Agric. Silvic. Mendelianae Brun. 2015, 63(4), 1269-1276 | DOI: 10.11118/actaun201563041269

A Predictive Likelihood Approach to Bayesian Averaging

Tomáš Jeřábek¹, Radka Šperková²

¹ Faculty of Economic Studies, University of Finance and Administration, Estonská 500, 101 00 Praha 10, Czech Republic

² Department of Economy and Management, College of Business and Hotel Management, Bosonožská 9, 625 00 Brno, Czech Republic

Multivariate time series forecasting is applied in a wide range of economic activities related to regional competitiveness and is the basis of almost all macroeconomic analysis. In this paper we combine multivariate density forecasts of GDP growth, inflation and real interest rates from four various models, two type of Bayesian vector autoregression (BVAR) models, a New Keynesian dynamic stochastic general equilibrium (DSGE) model of small open economy and DSGE-VAR model. The performance of models is identified using historical dates including domestic economy and foreign economy, which is represented by countries of the Eurozone. Because forecast accuracy of observed models are different, the weighting scheme based on the predictive likelihood, the trace of past MSE matrix, model ranks are used to combine the models. The equal-weight scheme is used as a simple combination scheme. The results show that optimally combined densities are comparable to the best individual models.

Keywords: predictive likelihood, density forecasts, Bayesian averaging, Bayesian VAR model, GDP growth, inflation, real interest rates

Grants and funding:

This work is supported by funding of specific research at University of Finance and Administration, Faculty of Economic Studies.

Prepublished online: September 2, 2015; Published: September 1, 2015  Show citation