Bootstrap Tests in Bivariate VAR Process with Single Structural Change : Power versus Corrected Size and Empirical Illustration

This paper evaluates the finite-sample performance of single structural change tests based onthe asymptotic distribution and bootstrap procedures. In addition to the conventional case of stationaryregressors, we consider nonstationary regressors and others characterized by the presence of a break intheir structure. While our paper borrows the idea of assessing the performance of structural break testsfrom an other paper, ours is the first to examine and compare the power of such tests on the basis ofcorrected size by using graphical methods. We endeavour to see whether some conclusions, obtainedfor some tests, remain again valid for others based on an other type of processes. Some bootstrapprocedures quasi-perfectly correct the size distortions of their asymptotic counterparts and have thesame power performance as them on the corrected size basis; property often difficult to obtain. We finally propose a modelling strategy to study the relationship between U.S. interest rates. The results show that such relationship has been altered by a regime-shift located at the beginning of the 1980s.