In this chapter, we have analyzed money demand in Indonesia in a period in which major instabilities in basic economic relations due to the Asian crises may be expected. In addition to an econometric approach we have applied fuzzy clustering in order to analyze the robustness of the econometric results. Both the econometric and the fuzzy clustering approach divide the period from 1990 to 2002 into separate sub-periods. In the econometric approach this is accomplished by inclusion of dummy variables in the regression model, and in the fuzzy clustering approach different clusters are identified in which local regression models are valid.
Both approaches reveal that there have been structural changes in Indonesian money demand during the late 1990s. A common result is that the income elasticity of money demand is quite stable before and after the crisis, the econometric estimation of the income elasticity after the crisis is about 0.93 and the fuzzy estimate is 0.87. The interest rate elasticity differs in both approaches: the econometric model indicates a absolutely smaller but still significant negative interest rate elasticity after the crisis, while the fuzzy approach yields an insignificant interest rate elasticity after the crises. A further difference is that the fuzzy approach suggests a higher number of sub-periods, namely four clusters, while the econometric model is based on only two sub-periods. However, it might well be that the results of the two approaches become even more similar when the fit of the fuzzy model is improved.
Our main conclusions are: Firstly, Indonesian money demand has been surprisingly stable in a troubled and difficult time. Secondly, the fuzzy clustering approach provides a framework for the robustness analysis of economic relationships. This framework can especially be useful if the number and location of sub-periods exhibiting structural differences in the economic relationships is not known ex-ante. Thirdly, our analysis does also reveal why the previous studies of Indonesian money demand delivered unstable results. Theses studies applied cointegration techniques. However, we show that the relevant Indonesian time series are trend-stationary such that the cointegration framework is not appropriate.