The size of the nonstationary component and its effect on tests for unit roots
We consider a nonstationary time series that is composed of a stationary and nonstationary component. Monte Carlo experiments show that common unit root tests have probabilities of committing a type I error that significantly exceed the level of significance. We find that the probabilities vary according to the relative size of the nonstationary component.
KeywordsUnit Root Variance Factor Unit Root Test Data Generation Process Monte Carlo Experiment
Unable to display preview. Download preview PDF.