https://doi.org/10.52903/wp2022321
GREEK GDP FORECASTING USING BAYESIAN MULTIVARIATE MODELS
Zacharias Bragoudakis
Bank of Greece and National and Kapodistrian University of Athens
Ioannis Krompas
NBG Economic Research
ABSTRACT
Building on a proper selection of macroeconomic variables for constructing a Gross Domestic Product (GDP) forecasting multivariate model (Kazanas, 2017), this paper evaluates whether alternative Bayesian model specifications can provide greater forecasting accuracy compared to a standard Vector Error Correction model (VECM). To that end, two Bayesian Vector Autoregression models (BVARs) are estimated, a BVAR using Litterman’s prior (1979) and a BVAR with time-varying parameters (TVP-BVAR). Two forecasting evaluation exercises are then carried out, a 28-quarters ahead forecast and a recursive 4-quarters ahead forecast. The BVAR outperformed the other models in the first, whereas the TVP-VAR was the best-performing model in the second, highlighting the importance of having adjusting mechanisms, such as time-varying coefficients in a model.
Keywords: Bayesian VARs, Forecasting, GDP, TVP-VAR, VECM
JEL-Classification: C11, C51, C52, C53
Acknowledgements: We thank D. Louzis and T. Kazanas for its constructive suggestions which helped us to improve the clarity of the paper. The paper has also benefited from the comments of the participants of the 34th Panhellenic Statistics Conference, Athens, 2022. The views expressed in this paper are those of the author and do not necessarily reflect those of the Bank of Greece and National Bank of Greece. The authors are responsible for any errors or omissions in this paper.
Correspondence:
Zacharias Bragoudakis
Economic Analysis and Research Department
Bank of Greece,
21 E. Venizelos Avenue, 10250, Athens, Greece
Tel: +302103203605
E-mail: zbragoudakis@bankofgreece.gr