Examining nonlinear dynamics of exchange rates and forecasting performance based on the exchange rate parity of four Asian economies

This paper employs smooth transition autoregressive (STAR) models to investigate the nonlinear dynamic behavior of exchange rate deviations based on the exchange rate parity (ERP) theory in four Asian economies: Japan, South Korea, Taiwan and Singapore. In this study, quarterly data from 1978 Q1 through 2007 Q4 are analyzed. The empirical results indicate that the deviations in the exchange rate of all four countries reject the null of linearity. The rate deviations in the Japanese and Korean cases exhibit a dynamic and smoothly symmetric ESTAR type process, while those in the Taiwanese and Singaporean cases match the smoothly asymmetric LSTAR type with respect to depreciating and appreciating regimes. These nonlinear characteristics can be explained by the existence of heterogeneous behavior and asymmetric information among economic agents. Furthermore, the estimation results of a nonlinear least squares (NLS) regression indicate that most of the parameter estimates are significant at the 10 percent level. The forecasted Japanese and Korean rate deviations in the ESTAR model are not superior to those from the AR model, possibly because these two countries experienced a serious fluctuation during the Asian financial crisis that occurred in 1997. However, based on the criterion of the RMSE, the forecasted Taiwanese and Singaporean rate deviations in the LSTAR model outperform those in the AR model.