Forecasts Based on Official Business Indices in Japan: A Real-Time Analysis

Abstract

This paper analyses the performance of forecasts of real economic activity based on the Japanese official business indices in a real-time framework. As in other countries, preliminary numbers of the business indices are first announced, then they are revised. These revisions are often substantial, so that the business indices that are available on a real-time basis often differ from their revised form. This paper explores how this difference can affect forecasts. Results indicate that real-time forecasts based on the diffusion leading index are not significantly worse than forecasts based on the final revised form. Also, the real-time forecasts using the composite leading index are often comparable to those from final revised forms. However, real-time forecasts from the diffusion coincident index are significantly worse than forecasts that use the final revised form.Key words and phrases: business index, probit model, real-time forecast IntroductionThis work examines real-time forecasts which are based on the official business indices in Japan. These official business indices, called "Keiki Doko Shisu" in Japanese, were formerly compiled by the Economic Planning Agency (EPA), and are currently compiled by the Economic and Social Research Institute (ESRI) in the Cabinet Office. Since promptness is regarded as important, the indices are first announced in a preliminary form, then they are revised; revisions are carried out repeatedly. Is there a significant difference between forecasts based on the revised and the real-time indices? This paper compares revised and realtime forecasts and evaluates how much they differ. However good the forecasts based on the final revised form are, if forecasts based on business indices that are available in real-time are unsatisfactory, then usefulness of these indices is limited from a practical point of view.The same question over preliminary and revised numbers applies to other countries as well as Japan. Diebold and Rudebusch (1991a) show that realtime forecasts of production increase in the US based on leading business indices are hardly useful, though forecasts based on ex-post revised numbers perform satisfactorily. Diebold and Rudebusch (1991b) show similar results in regard to turning points of the business cycle in the US.There are already many studies on the performance of the official business indices in Japan, but they use business indices that are revised many times and