Testing structural stability with endogenous breakpoint A size comparison of analytic and bootstrap procedures

We compare the performance of two alternative approximations to the finite-sample distributions of test statistics for structural change, one based on asymptotics and one based on the bootstrap. We focus on tests acknowledging that the breakpoint is selected endogenously — in particular, the ‘supremum’ tests of Andrews (1993). We explore a variety of issues of interest in applied work, focusing particularly on smaller samples and persistent dynamics. The bootstrap approximation to the finite-sample distribution appears consistently accurate, in contrast to the asymptotic approximation. The results are of interest not only from the perspective of testing for structural change, but also from the broader perspective of compiling evidence on the adequacy of bootstrap approximations to finite-sample distributions in econometrics.