Forecasting Russian Macroindicators with Frequentist and Bayesian Econometrics

This work is devoted to forecasting Russian macroeconomic indicators with Bayesian methods. In order to evaluate the forecast quality of Bayesian vector autoregressions (BVARs) we compare prediction accuracy of random walk with drift (RW), unrestricted vector autoregression (VAR) and BVAR with Minnesota and conjugate normal-inverse Wishart priors with 3, 6 and 13 variables. Forecasting is performed for consumer price index (CPI), industrial production index (IPI) and interbank rate. As a result, we find that BVARs forecast much better than unrestricted VARs for all predicted indicators and all forecasting horizons (1, 3, 6 and 12 months). However, comparing BVAR to RW model we get non-uniform results. BVAR with 13 variables has more accurate forecast for CPI and partly for IPI, but interbank rate is forecasted slightly worse.