This paper examines the effectiveness of forecasting methods using multiple information variables in forecasting the rate of changes in the consumer price index (CPI) and real GDP in Japan, and investigates the background of forecast performance improvement and its limitations. We first examine the performance of forecasts that use individual information variables as well as forecasts that use multiple information variables. The results show that no single variable improves forecasts in all periods for either CPI or GDP, but combining the information from individual forecasts can lead to a stable forecast performance. Next, to explore the backdrop to these improvements in forecast performance, we decompose and analyze the forecast error of forecast combinations using a simple mean. We discover that the irregular movements of forecast errors generally cancel each other out, which in turn leads to a reduction in errors. At the same time, the effect of reducing forecast errors rapidly diminishes with the addition of variables, and we verify that forecast performance stops improving after two to four variables are added. For this reason, it is necessary to consider both the performance of original forecast series that comprise the combination, and the combination of variables that best reduces the correlation among forecast error series to obtain the optimal combination of series.
Keywords: Information variable; Multivariate forecast; Out-of-sample forecast; Forecast combination
Views expressed in the paper are those of the authors and do not necessarily reflect those of the Bank of Japan or Institute for Monetary and Economic Studies.