Do BVAR Models Forecast Turkish GDP Better Than UVAR Models?
Irem Sacakli-Sacildi *
Department of Econometrics, Marmara University, Faculty of Economics, Goztepe Campus, 34722, Kadikoy, Istanbul, Turkey
*Author to whom correspondence should be addressed.
Abstract
Forecasting gross domestic product (GDP) is crucial for developing macroeconomic policies and managements. Vector autoregression models are one of the commonly used multivariate time series models for forecasting. The Bayesian vector autoregression models are used to avoid problems of multicollinearity and over parameterization that occur in general with the use of vector autoregression models. The aim of this paper is to forecast Turkish GDP using Bayesian vector autoregression models with quarterly data from 2005q4 to 2013q3 and compare the results with unrestricted vector auto regression models. The out-of-sample forecasting accuracy of these models are compared with unrestricted vector autoregression models. The results confirm the accuracy of Bayesian vector auto regression models for forecasting GDP. On the other hand unrestricted vector autoregressionmodels are most accurate for exchange rates except the first quarter and for interest rate in the first two quarter forecasts.
Keywords: Vector autoregression, forecasting, bayesian methods, forecast accuracy