A0790
Title: Detection of relevant changes in the frequency domain
Authors: Yan Liu - Waseda University (Japan) [presenting]
Abstract: The problem of detecting relevant changes in the frequency domain is considered. The relevant changes are formulated in the framework of nonparametric functionals of the spectral density of the time series in consideration. We propose a consistent test statistic for detecting relevant changes in the frequency domain. Specifically, we construct a new CUSUM statistic of the nonparametric estimator for the spectral density. We also elucidate the consistency of the CUSUM statistic with the relevant change. The CUSUM statistic consisting only of periodograms is not available here because it is not consistent. The proposed statistic has good features such as asymptotic convergence to the Brown bridge, and can be applied to the detection of relevant changes in hidden structures of integer-valued time series. We will also show some numerical examples and applications of this method to the real data based on the above theoretical results.
