Nonlinear Forecasting with a Hybrid Approach Combining Sarima, Arch and Ann

Abstract

Time series forecasting is a vital issue for many institutions. In the literature, many researchers from various disciplines have tried to improve forecasting models to reach more accurate forecasts. It is known that real life time series has a nonlinear structure in general. Therefore, conventional linear methods are insufficient for real life time series. Some methods such as autoregressive conditional heteroskedastiacity (ARCH) and artificial neural networks (ANN) have been employed to forecast nonlinear time series. ANN has been successfully used for forecasting nonlinear time series in many implementations since ANN can model both the linear and nonlinear parts of the time series. In this study, a novel hybrid forecasting model combining seasonal autoregressive integrated moving average (SARIMA), ARCH and ANN methods is proposed to reach high accuracy level for nonlinear time series. It is presented how the proposed hybrid method works and in the implementation, the proposed method is applied to the weekly rates of TL/USD series between the period January 3, 2005 and January 28, 2008. This time series is also forecasted by using other approaches available in the literature for comparison. Finally, it is seen that the proposed hybrid approach has better forecasts than those calculated from other methods. © 2011 by Nova Science Publishers, Inc. All rights reserved.