de Conference on Forecasting and Monetary Policy

We compare the Bank of England’s Inflation Report (IR) quarterly forecasts for growth and inflation to real-time forecasts using a variety of statistical forecasting models that have previously been found useful as forecasting benchmarks. These include linear and non-linear univariate models, and VARs. The results reveal the well-known difficulty of forecasting in a stable macroeconomic environment, and the IR forecasts for GDP growth are inferior to forecasts from linear and non-linear univariate models. However, for the inflation forecast the IR strongly dominates. The explanation for these inflation results appears to be that the official forecasts incorporate judgements about monetary regime changes, leading to inflation mean shifts, which improve forecast performance markedly. The joint use of rolling windows and simple unweighted forecast combination helps improve the statistical models’ performance, although where they outperform the IR, the improvement is not significant.