A BVAR Model for Forecasting of Czech Inflation

Bayesian vector autoregressions (BVAR) have turned out to be useful for medium-term macroeconomic forecasting. Several features of the Czech economy strengthen the rationale for using this approach. These include in particular the short time series available and uncertainty about long-run trends. We compare forecasts based on a small-scale mean-adjusted BVAR with the official forecasts published by the Czech National Bank (CNB) over the period 2008q3–2016q4. The comparison demonstrates that the BVAR approach can provide more precise inflation forecasts over the monetary policy horizon. For other macroeconomic variables, the CNB forecasts either outperform or are comparable with the forecasts based on the BVAR model