A1254
Title: Tight-difference-based centrosymmetric kernel estimators for long-run variance
Authors: Kin Wai Chan - The Chinese University of Hong Kong (Hong Kong) [presenting]
Ya Xuan Wang - The Chinese University of Hong Kong (Hong Kong)
Abstract: Long-run variance estimation is important for many statistical inference procedures. Existing methods give disappointing results when time series data exhibit both time-varying mean trends and significant serial dependence. Differencing and the kernel method are mainstream methods for resolving these problems and achieving mean robustness and consistency, respectively. Nevertheless, a large differencing lag is required to make kernel variance estimators work for time series, but the de-trending effect is substantially affected. It makes mean-robustness and consistency hard to be achieved simultaneously. This problem is tackled by constructing a novel centrosymmetric kernel to apply a tight differencing operation. The centrosymmetrization principle is simple and applicable to a large class of kernels. Optimal tight differencing sequences for handling serially dependent data are proven to be data-independent and universal. Therefore, they can be implemented directly without fitting into practice. The proposed principlealso appliese to spectral density estimation, where the robustness property is well inherited, showing great discrimination between the spectral pattern within noises and signals. This research was partially supported by General Research Fund 14304420, 14306421, and 14307922 provided by the Research Grants Council of HKSAR.
