出版社：Suntory Toyota International Centres for Economics and Related Disciplines
摘要：Threshold autoregressive (TAR) models have been studied intensively last decade. The asymptotic properties of the least squares estimation are established by Chan (1993) and Chan and Tsay (1998). See also Hansen (1997) for a di¤erent asymptotic approach. Various testing methods have been developed for the presence of a threshold e¤ect, see Chan and Tong (1990) and Hansen (1996) among others. On the other hand, testing the unit root null hypothesis in a TAR model is a rather recent area of investigation. For this purpose, the models are distinguished depending on whether the threshold variable is a level or a di¤erence of the time series (we call them a level-based TAR model and a di¤erence-based TAR model respectively). Unlike the di¤erences, the level is nonstationary under the unit root hypothesis and the relevant distribution theory is completely di¤erent. Caner and Hansen (2001) provide a rigorous treatment of statistical tests for the di¤erence-based model. In the absence of an appropriate distribution theory, however, the standard unit root test such as the ADF test has been commonly used for the level-based TAR models. The purpose of this paper is to develop such a distribution theory and a bootstrap for nite sample inferences
