B0808
Title: Bandwidth selection under random fields with short memory
Authors: Bastian Schaefer - University of Paderborn (Germany) [presenting]
Abstract: The problem of bandwidth selection in semiparametric spatial models under dependent errors is studied. We use a spatial representation of high-frequency financial data on a lattice and aim at estimation of a non-stationary regression surface and a stationary component with short memory in all dimensions, e.g. volatility or other risk measures. The non-stationary mean surface function is estimated by a nonparametric method, the demeaned residuals are modeled as a random field which is estimated by parametric methods. We propose bandwidth selectors for estimation of the regression surface under different random field representations of the error terms. Optimal bandwidths are chosen by an iterative plug-in algorithm to minimize the mean integrated squared error of the nonparametric estimator. An implementation into an R package is given and the properties of the estimator are assessed under a simulation study.
