On the empirical size of Nielsen's multivariate likelihood ratio test of fractional integration

Abstract

Nielsen (2004) provides multivariate maximum likelihood procedures to test whether several series are fractionally integrated. This note examines the finite sample size of the likelihood ratio tests applied to a bivariate system experiencing breaks in means. The results suggest that tests of a common unit root are somewhat undersized in small samples.

Acknowledgement

Author thanks Morten Nielsen and Mark Jensen for sharing their Ox code and feedback, and an anonymous referee and Mark Taylor for very helpful suggestions.