B1047
Title: Structural break detection using Fourier methods
Authors: Herold Dehling - Ruhr-University Bochum (Germany)
Roland Fried - TU Dortmund University (Germany)
Max Wornowizki - TU Dortmund University (Germany) [presenting]
Abstract: A new class of tests for checking the constancy of a distributional parameter such as the variance for a sequence of random variables is presented. The corresponding test statistics are based on a Fourier type transformation of blockwise estimators of this parameter. Different weight functions result in different test statistics, which are all given by simple explicit formulae. Assuming independence and piecewise identical distributions, testing is conducted applying the permutation principle or using asymptotic results. To prove the latter the test statistics are viewed as U-statistics constructed from the blockwise parameter estimators. Since the distribution of these estimators depends on the sample size, a new LLN and a CLT are proven. The asymptotic and the permutation test are compared to other tests for constant variance in extensive Monte Carlo experiments and illustrated in an application. In comparison to their competitors the new methods offer good power particularly in the case of multiple structural breaks and estimate the positions of the structural breaks adequately.
Journal