Finite-sample properties of modified unit root tests in the presence structural change

摘要

Using Monte Carlo experimentation the finite-sample properties of modified unit root tests are examined in the presence of structural breaks. It is found that previously derived results concerning the size robustness of weighted symmetric and recursively mean-adjusted unit root tests in the presence of structural breaks are achieved at the expense of dramatic losses in power, thereby reducing the appeal of the tests to the practitioner. The properties of the tests are also seen to be highly dependent upon the position or timing of the observed break. Finally, bias in the estimation of the autoregressive parameter is analysed. The results derived demonstrate that while recursive mean adjustment reduces negative bias in the absence of a break, it leads to increased positive bias when a break occurs.