Title: Adaptive bandwidth selection for M-estimators in locally stationary time series
Authors: Stefan Richter - Heidelberg University (Germany) [presenting]
Abstract: A general class of locally stationary processes which are assumed to depend on an unknown finite-dimensional parameter curve is studied. For the estimation of these curves, we use nonparametric local M-estimators depending on a bandwidth. We investigate a local bandwidth selection procedure via contrast minimization. We prove that the corresponding estimators for the parameter curves achieve the asymptotically optimal minimax rate up to a log factor. All important conditions of our theorems only concern the (usually well-known) corresponding stationary time series model which allows for an easy verification. The performance of the method is studied in a simulation for some examples like tvAR, tvARCH and tvMA processes.