this paper investigates the EH by estimating a bivariate vector autoregression (VAR) for the longterm and short-term interest rates and testing the restrictions implied by the EH. This test was first suggested by Campbell and Shiller (1987); however, the procedure used here was developed by . 1 Second, this paper deals directly with the issue of stationarity. While stationarity is frequently considered in testing the EH, its implications for the EH are seldom discussed. This paper attempts to fill this void. As a matter of theory, many economists and financial specialists appear to believe that interest rates are stationary. If they are not, the role played by the EH in monetary policy may be diminished.As a practical matter, interest rates tend to exhibit considerable persistence. Indeed, the null hypothesis of nonstationarity is frequently not rejected even in relatively large, finite samples. This is particularly important for Japanese interest rates because they exhibit considerable persistence at the monthly frequency. It is well known, however, that tests of unit roots may have low power. Moreover, many financial market economists argue that interest rates are stationary on theoretical grounds. Given the differences of opinion about whether interest rates are stationary in general, the VAR test is applied under the assumption that interest rates are either stationary or nonstationary. The issue of stationarity is important only if the conclusions concerning the EH differ markedly depending on the assumption made.1 Several of the tests of the EH that are frequently used have low power and, more importantly, tend to generate results that can give a misleading impression of the strength or weakness of the EH. For a more detailed discussion of this problem, see Thornton