Examining the robustness of cointegration analysis under weighted symmetric estimation

Abstract

The research of Leybourne and Newbold (2003) is extended to examine the finite-sample size of the weighted symmetric cointegration test when applied to independent unit root processes subject to structural change. The results obtained show the weighted symmetric cointegration test to be more robust to structural change than previously examined cointegration tests. Combined with its previously noted higher power, the findings of the present analysis suggest the recently proposed weighted symmetric cointegration test to be of use to the practitioner.

Notes

¹ The analysis of Leybourne and Newbold (2003) also considers the less popularly employed test of Banerjee et al . (1986).

² The oversizing exhibited by the Johansen test for early breaks in either series (dependent variable or regressor) results from the symmetrical treatment of time series under the Johansen procedure. It should be that although LN report results for the trace test only, the authors state that very similar results were derived for the maximal eigenvalue test.

³ For further information on weighted symmetric estimation in the context of testing the unit root hypothesis, see Park and Fuller (1995). For more general information on weighted symmetric estimation, see Fuller (1996.

⁴Leybourne and Newbold (2000) provide theoretical justification for the markedly different values of b considered under level and trend breaks.

⁵ The use of a single representative sample size follows the approach of LN. However, further near identical results for alternative sample sizes are available from the authors upon request.