Long-run money demand in Taiwan revisited: evidence from a cointegrating STR approach

Abstract

Previous studies on the nonlinearity of the money demand function focus only on short-run dynamics and assume that the long-run cointegrated relationship is linear, which, according to economic theory, need not be the case. Thus, this article focuses on the variables that make up the long-run money demand function in Taiwan and their determinants by using the cointegrating smoothing transition regression (CSTR) test developed by Choi and Saikkonen (2004). This model is more general than the previous STR model in that it may contain several transition functions and has more than a single transition variable. We demonstrate our empirical results for robustness and adopt two models to ascertain whether or not the stock price should be added as a determinant of the demand for money. Our empirical results show the existence of a stationary long-run relationship among real money demand functions for the period from 1976Q1 to 2005Q1 and the robustness test covers the 1980Q1–2005Q1 period in Taiwan. By comparing the two types of model, we find more evidences of a long-run nonlinear cointegrated relationship for the money demand function when we consider stock prices as one of its variables. Finally, the empirical results give rise to some critical policy implications.

Notes

¹A large number of articles have applied the cointegration method to test the nature of the long-run money demand function in developing countries. There are several studies that have discussed the money demand function for developing country, for example: Frenkel and Taylor (1993) in Yugoslavia; Phylaktis and Taylor (1993) in Argentina, Bolivia, Brazil, Chile and Peru. However, to date very few studies have investigated the stability of the money demand function in the developing countries (Chaisrisawatsuk et al., 2004).

²Additionally, based on the general-to-specific methodology, Cuthbertson and Taylor (1992) have created the model to assess the relative performance of the forward-looking models and the error correction models using the demand for M1 in the United Kingdom.

³Choi and Saikkonen (2004) indicate that money demand is more responsive to the interest rate when the rate is higher. The public will then be more sensitive to the interest rate when they make decisions regarding holding money.

⁴In the STECM, the error term is coming from the OLS estimation. However, the conventional money demand function views the error term as a linear process. Therefore, the whole cointegrating relationship in the STECM is still characterized by linearity.

⁵Taylor (1994) pointed out that we should pay attention to the phenomenon such as the lagged real-money balances smaller with the size of the deviation from equilibrium in estimated STR model, because the lagged dependent variables have large estimated coefficients but will cause one that hard to explain at a theoretical level.

⁶Using the Cagan model of money demand as well as the narrow definition of the money supply over the period from 1945 to 1949, when Taiwan being placed in very high rate of price inflation, Phylaktis and Taylor (1992) rejected the hypothesis that the authorities expanded the money supply in order to maximize the advantaged inflation tax revenue from money creation.

⁷It is also deflated by the consumer price index (2000=100) and transformed into a real variable.

⁸This section draws heavily on the work of Choi and Saikkonen (2004).

⁹There are five requisite econometric restrictions for the transition function g(z), assumed by Choi and Saikkonen (2004).

¹⁰Assumptions 3–6 in Choi and Saikkonen (2004) need to be satisfied.

¹¹These findings are similar to Choudhry (1996) who reported that real stock prices play an important role in the money demand function both in Canada and the US.