Testing the Power of a Generalization of the KPSS-Tests against Fractionally Integrated Hypotheses

We analyse in this article the size and the power properties of differentgeneralizations of the KPSS-tests proposed by Hobjin et al. (1998) for testingthe null hypothesis of stationarity in univariate time series when thealternatives are of a fractional form. We show that the test based on the useof the Quadratic Spectral kernel along with an automatic bandwidth selectionprocedure produces the best results and thus, it might be employed for testingI(0) against I(d>0) stationary or nonstationary processes. An empiricalapplication, showing the performance of the tests in finite samples is alsocarried out at the end of the article. Copyright Kluwer Academic Publishers 2003