期刊名称：Discussion Paper Series / Economics Department, University of Swansea
出版年度：2004
卷号：1
出版社：University of Swansea
摘要：We confirm that unit root tests can exhibit substantial size distortion when breaks in mean are
generated by a first-order Markov chain, but unlike previous literature, we find augmentation largely
remedies this situation. However, considerable heterogeneity is evident in the size properties of the
tests when faced with breaks in mean varying in duration, in number, and in position within the
sample. This heterogeneity will be hidden when a Markov chain is employed. For instance, when the
transition probabilities generate single period outliers, rejection frequencies rise substantially with
the number of outliers, but augmentation results in approximately nominal rejection frequencies.
Qualitatively similar results hold when a number of structural breaks are allocated randomly in
the central section of the sample. Interestingly, very different behaviour is exposed by a design
exploring the impact on the tests of two breaks imposed at a range of fixed intervals, rejection
frequencies rising when breaks occur in the extremities of the sample, a situation unaffected by
augmentation.
关键词：Unit roots; Markov chains; Monte Carlo Simulation; Structural breaks; Outliers.
Word count: 6473 words.
Loading...
