SEMIFAR models

Abstract

Recent results on so-called SEMIFAR models introduced by Beran (1997) are discussed. The nonparametric deterministic trend is estimated by a kernel method. The differencing- and fractional differencing parameters as well as the autoregressive coefficients are estimated by an approximate maximum likelihood approach. A data-driven algorithm for estimating the whole model is proposed based on the iterative plug-in idea for selecting bandwidth in nonparametric regression with long-memory. Prediction for SEMIFAR models is also discussed briefly. Two examples illustrate the potential usefulness of these models in practice.