Markov level shifts and the unit-root hypothesis

This paper examines the properties of tests for the presence of an autoregressive unit root in time series that are subject to multiple level shifts. The latter are assumed to be governed by a time-homogeneous finite Markov chain, thus allowing for an arbitrary number of stochastic breaks. It is demonstrated that standard tests of the unit-root hypothesis against stationary or single-break alternatives experience serious difficulties in the presence of Markov level shifts.