What do we know about real exchange rate nonlinearities?

Nonlinear modeling of adjustments to purchasing power parity has recently gained much attention. However, a huge body of the empirical literature applies ES- TAR models and neglects the existence of other competing nonlinear models. Among these, the Markov Switching AR model has a strong substantiation in international finance. Our contribution to the literature is five-fold: First, we compare ESTAR and MSAR models from a unit root perspective. To this end, we propose a new unit root test against MSAR as the second contribution. Thirdly, we study the case of misspeci- fied alternatives in a Monte Carlo setup with real world parameter constellations. The ESTAR unit root test is not indicative, while the MSAR unit test is robust. Fourthly, we consider the case of correctly specified alternatives and observe low power of the ESTAR but not for the MSAR unit root test. Fifthly, an empirical application to real exchange rates suggests that they may indeed be explained by Markov Switching dy- namics rather than ESTAR.