On Forecast Strength of Some Linear and Non Linear Time Series Models for Stationary Data Structure
Year: 2015
Type: article
Abstract: The motivation for this study is that the multi-step forecast performance of nonlinear time series models is often not reported in literature. This is perhaps surprising since one of the main uses of time-series models is for prediction. Therefore, this paper aimed at comparing a number of models that have been proposed in the literature for obtaining h-step ahead forecasts for different features of linear and nonlinear models. These forecasts are compared to those from linear autoregressive model. The comparison of forecasting methods is made using simulations in R software. The relative efficiency of the models was assessed using MSE and AIC criteria. The best model to forecast linear autoregressive function is AR. The performance of LSTAR model supersedes other models as number of steps ahead increases in nonlinear time series data except in polynomial function where SETAR model performs better than others. The predictive ability of the four fitted models increases as sample size and number of steps ahead increase.
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: diamond
OpenAlex ID: W1410177184