The dynamic relationship between real exchange rates, real interest rates and foreign exchange reserves: empirical evidence from China

Abstract

This article examines the long-run and short-run relationship between China's real exchange rate, foreign exchange reserves and the real interest rate differential between China and the United States using monthly data from 1980 to 2002. Extensive testing for unit roots allowing for up to two structural breaks in the trend indicates that the variables are not integrated of the same order. Thus, the bounds testing approach to cointegration is used, which finds that there is a single long-run relationship between the three variables. In the long run the real exchange rate has a statistically significant positive effect on foreign exchange reserves. The coefficient on the real interest rate differential is also positive, but is statistically insignificant. In the short-run it is found that the relationship between the real exchange rate, real interest rate differential and foreign exchange reserves is non-monotonic.

Acknowledgements

We thank Zhongxia Jin for generously providing us with his data set, Brett Inder for assistance with the GAUSS codes used to run the unit root tests with structural breaks and Dietrich Fausten for comments on an earlier version of this article. We alone are responsible for the views expressed.

Notes

¹ We also experimented using an unrestricted intercept and no time trend in the cointegrated model. We found that this weakened the diagnostics of the model. Therefore we choose a model with an unrestricted intercept and unrestricted time trend, which is case V in Pesaran et al. (2001).

² The dummy variables are relevant to the short-run estimations. Note that the asymptotic theory in Pesaran et al. (2001) is not affected by the inclusion of ‘one off’ dummy variables provided that the fraction of periods in which the dummy variables are non-zero does not tend to zero with the sample size T, which is the case here. In the empirical example given in Pesaran et al. (2001, p. 307) the fraction of non-zero observations were 7.6% and 19.2% of sample size respectively and these were not considered sufficient to alter the critical values.

³ To select the lag length, we adopted the approach suggested in Pesaran et al. (2001, Table I).The optimal lag lengths were selected based on the Schwarz Information Criterion.