GLS Detrending-Based Unit Root Tests in Nonlinear STAR and SETAR Frameworks.

This paper consider the GLS detrending procedure advanced by Elliott et al. (1996) for unit root tests against alternative hypotheses where the time series data under investigation follow either globally stationary SETAR or STAR processes with deterministic components being present. It is found that the proposed testing procedures have considerable power gains against both the standard Dickey-Fuller unit root tests and existing nonlinear unit root tests recently proposed by Kapetanios and Shin (2002) and Kapetanios et al. (2003). The empirical application to DM and Yen bilateral real exchange rates against a number of other currencies also confirms that nonlinear unit root tests based on GLS detrending will be more powerful than linear ones. Interestingly, we find that the DM dataset seems to produce more rejections of the null using the GLS detrending-based SETAR tests than using the GLS detrending-based STAR tests, whereas the number of rejections of both tests are similar for the Yen dataset. The different results may arise from the respective liquidity of the DM and Yen Forex markets.