Time Series Prediction Using Hybridization of AR, SETAR and ARM Models
Dursun AYDIN Öznur ISÇI GÜNERI

Abstract
It is known that parametric and nonparametric methods are used for nonlinear time series. In recently, hybrid models are also considered in time series forecasting. In this paper we present the hybrid models whose components are parametric and nonparametric models. Of the parametric methods, autoregressive (AR) model and self-threshold value (SETAR) model and, of the nonparametric methods, additive regression model (ARM) and hybrid AR&AAR, AR&SETAR, AAR&SETAR, SETAR&AR, SETAR&AAR and AAR&AR models are used in this study. In this context, back fitting algorithm based on smoothing spline method in the existing literature is discussed. A comparison has been made for the performance of the models obtained for the export volume index numbers and domestic producer price index data for Turkey. These results showed that AAR&SETAR hybrid model has denoted the best performance among the all models in time series forecasting.

Full Text: PDF