A COINTEGRATION ANALYSIS OF SINGAPORE INTEREST RATES

Abstract

The development of the idea of cointegration has offered an exciting era for the testing of economic theories and economic modelling. The recent sprout of cointegration literature on its application to the analysis of interest rates has prompted this study, which employs the Engle and Granger 2-step estimation procedure for investigating the long-run relationships of the local interest rate series. In particular, the interbank money market rates and the fixed deposit rates are used for our analysis. This study finds that the interbank money market rates and the fixed deposit rates are well modelled as I( I ) series. Moreover, the interbank money market rates across different maturities are cointegrated. This implies that there exists a long-run equilibrium relationship between the interbank rates across different maturities. The common stochastic trend that drives the term structure of the interbank rates could be the expected inflation rate, the risk premium, foreign interest rates, some common institutional factors that have had an impact on the interbank money market in general or a linear combination of these factors. Our results also suggest the absence of a "rational bubble" in the term structure. Lastly, the results seem to give some empirical support for the expectations hypothesis. However, the study shows that the fixed deposit rates are not cointegrated. This could have resulted because the fixed deposit market is largely controlled by the "Big Four" local banks, namely the Development Bank of Singapore (DBS), the Overseas Union Bank (OUB ), the United Overseas Bank (UOB) and the Overseas Chinese Banking Corporations (OCBC). The fixed deposit rates may not share a common trend because the institutional forces could have acted independently upon each series across different maturities. Since cointegration also implies causality in the system, it may be possible to generate good forecasts for the interbank rates using the error-correction models. Our results show that the forecasting performance of the error-correction models is better than the naive no-change predictions for the 1-month interbank rates. However, these good results are not replicated for the error-correction models of the interbank rates of higher maturity. This may indicate that the long-term rates are driving the term structure so that they would yield significantly more information for the forecast of the short-term rates than vice versa. The result is also consistent with other studies that have shown that the simple random walk models are good approximations for the interest rates with longer maturities. Although the cointegration techniques are useful for economic research, they have often been misused. For example, it has been a common practice for investigators to split up their sample sizes and test for structural change in the domestic term structure relationship within each sub-sample. However, this procedure has been criticised in this study because of the low power of unit root tests for the small sub-samples. Moreover, the efficiency implications of the cointegration tests and the forecasting results of the error-correction models should also be carefully interpreted.