Abstract
The finite-sample properties of recently proposed range unit root tests are examined in the presence of serial correlation and drift. The results obtained show that both tests suffer from severe size distortion when applied to unit root process which either possess serially correlated disturbances or exhibit drift. Consequently, the noted robustness of the tests and the appropriateness of the previously provided critical values are both questioned.
Notes
1It should be noted that while the Dickey–Fuller test includes deterministic terms explicitly in the testing equation, the other tests mentioned remove deterministic components prior to estimating the appropriate testing equation.
2As noted by Hylleberg and Mizon (Citation1989), for a given sample size, the strength of the resulting drift (μt) is dependent on the ratio of μ to the variance of the error υ t .