Is the Consensus Forecast Just the Average? -- Re-examining an Evaluation of the ESP Forecast -- (in Japanese)

This paper examines forecast data of the ESP Forecast (hereafter ESPF), the monthly survey of forecasts for the Japanese economy, for 4 variable -- three quarterly figures of real GDP growth rates, year-on-year CPI growth rate and unemployment rate, and fiscal year real GDP growth rate -- in FY 2004 and 2005. The consensus forecast turns out to perform well, which is a robust result, regardless of changes in evaluation methods. This is because even though realized values often "betray forecasts", that is, they could be regarded as outliers from forecasters' viewpoints, the consensus forecast, by definition, does not deviate so much from realised values. Hence, each forecaster should find it easy to beat the consensus a few times, but difficult to continue. Frequent large deviations of realised values could result from relatively large uncertainty attached to the forecasts. The ESPF keeps the following two compatible: producing the consensus valuable to users; making it possible for participating forecasters to beat the consensus, thereby encouraging the forecasters to continue to participate in the ESPF. Forecast combination is shown to reduce RMSE of forecast errors the most for quarterly real GDP growth rate (s.a.a.r.), some 1.5 percentage points.