Asymptotics for Unit Root Tests Under Markov Regime-Switching

Abstract:     
The large sample effect of Markov switches on unit root inference is investigated. Attention is paid to semiparametric (unit root and stationarity) tests as well as to parametric (Dickey-Fuller type) tests. With respect to the existing literature, instead of relying on Monte Carlo simulation the analysis is carried out by focusing on an appropriate asymptotic theory for I(0) and I(1) processes under Markov switching. It is shown that Markov switches in the trend component can make I(0) and I(1) processes observationally equivalent and that unit root tests virtually have no power to detect stationary processes around switching trends, although autocorrelation-robust unit root tests are not affected by size distortions. Conversely, Markov switches in the mean of the transitory components do not change the usual asymptotic properties of the tests. Finally, it is shown that in large samples Markov-switching variances cause neither size distortions nor inconsistency of the tests.

Number of Pages in PDF File: 24

Accepted Paper Series