63rd European Meeting of the Econometric Society, Milan, İtalya, 27 - 31 Ağustos 2008, ss.1
While a number of unit root testing
procedures have been developed to account for nonlinearity under the
alternative hypothesis of stationarity, almost all available tests assume a
linear DGP under the unit root null hypothesis. This paper establishes some
theoretical results relating to the inclusion of nonlinear terms in an ADF
regression and proposes two new unit root tests that allow the process to be
nonlinear under the null hypothesis. More specifically, block and model-based
bootstrap procedures are developed for smooth transition threshold models.
Simulations show that the latter is preferred and the model-based bootstrap
test delivers a good size performance across all specifications, including
linear and effectively abrupt transition models. The model-based test also
dominates the other tests in terms of power and an application to the US
unemployment rate shows that it can overturn conclusions based on an erroneous
linearity assumption.