We show that tests for a break in the persistence of a time series in the classical I(0) - I(1) framework have serious size distortions when the actual data generating process exhibits long-range dependencies. We prove that the limiting distribution of a CUSUM of squares based test depends on...