The Yield Spread as a Predictor of Japanese Recessions

Economists have used various financial economic indicators to forecast economic activity. For example, the information contained in the term structure of interest rates has long been regarded as useful for predicting future economic activity. And recently, this topic has received renewed attention as a result of the increasing importance of bond markets behavior for monetary policy and real economic activity in developed economies. Estrella and Mishkin (1996a, b) have found that the yield spread the difference between long-term and short-term interest rates contains useful information for predicting recessions in the U.S. economy by employing the Probit type estimation. They have also found that the yield spread outperforms the stock prices and other leading economic indicators in predicting recessions over forecasting horizons longer than a quarter. In this article, we apply the Estrella-Mishkin analysis to the Japanese economy. The results are somewhat tentative, given the relatively short duration of the period of free interest rate movements, that is, the period since the late 1970s. Nonetheless we find that (1) The yield spread does predict Japanese recessions, based on the Probit model. (2) Other financial variables also predict recessions. The monetary aggregates predict recessions well, but they are also close to a coincident indicator. The stock price is a leading indicator of real economic activity, but is a noisy indicator. Among the three, the yield spread seems to be the best leading indicator of recessions. On the other hand, (3) The predictive power of the yield spread in the Japanese financial markets, however, is not so strong as that in the U. S. markets. For example, in simple OLS regressions of indicators of economic activity on the yield spread, the yield spread is insignificant at a quarterly frequency, though it is significant at a monthly frequency. We suspect that the result reflect the short duration of the sample period and that they may well change as more data become available in the future. In this sense, the results must be interpreted with caution.