1: \begin{abstract}

2: In this paper we investigate the problem of testing the assumption of

3: stationarity in locally stationary processes. The test is based on an

4: estimate of a Kolmogorov--Smirnov type distance between the true time

5: varying spectral density and its best approximation through a

6: stationary spectral density. Convergence of a time varying empirical

7: spectral process indexed by a class of certain functions is proved, and

8: furthermore the consistency of a bootstrap procedure is shown which is

9: used to approximate the limiting distribution of the test statistic.

10: Compared to other methods proposed in the literature for the problem of

11: testing for stationarity the new approach has at least two advantages:

12: On one hand, the test can detect local alternatives converging to the

13: null hypothesis at any rate $g_T \to0$ such that $g_T T^{1/2} \to

14: \infty$, where $T$ denotes the sample size. On the other hand, the

15: estimator is based on only one regularization parameter while most

16: alternative procedures require two. Finite sample properties of the

17: method are investigated by means of a simulation study, and a

18: comparison with several other tests is provided which have been

19: proposed in the literature.

20: \end{abstract}