Recent developments in non- and semiparametric regression with fractional time series errors

Abstract

This paper summarizes recent developments in non- and semiparametric regres- sion with stationary fractional time series errors, where the error process may be short-range, long-range dependent or antipersistent. The trend function in this model is estimated nonparametrically, while the dependence structure of the error process is estimated by approximate maximum likelihood. Asymptotic properties of these estimators are described briey. The focus is on describing the developments of bandwidth selection in this context based on the iterative plug-in idea (Gasser et al., 1991) and some detailed computational aspects. Applications in the framework of the SEMIFAR (semiparametric fractional autoregressive) model (Beran, 1999) illustrate the practical usefulness of the methods described here.