Abstract
Building on work by Phillips (Econ. Theory 7:450463, 1991), we derive LAD based unit root tests in a first-order ESTAR model with strong mixing innovations. Further theoretical results are derived and LAD based unit root tests in general nonlinear first-order dynamic models admitting a Taylor-series approximation are thereby easily obtained. Finite sample properties of the tests are explored using Monte Carlo experiments. The results show that the size properties of the tests are satisfactory, and the power against stationary ESTAR alternatives with innovational outliers is significantly higher than the power of the LS based unit root tests by Kapetanios et al. (J. Econ. 112:359379, 2003) and Rothe and Sibbertsen (Allg. Stat. Arch. 90:439456, 2006). In contrast, the LS based tests are more powerful than our tests in the case of stationary ESTAR models with Gaussian errors (no outliers). In an empirical application to eight real effective exchange rates (for major economies), evidence of the PPP hypothesis is supported for six of the countries using our tests. If LS based tests are instead used, the PPP hypothesis is supported for three countries only (countries for which the PPP hypothesis is also supported by our tests).