Improvement in Forecasting Accuracy Using the Hybrid Model of ARFIMA and Feed Forward Neural Network

Abstract

In recent years, autoregressive fractionally integrated moving average (ARFIMA) models have been used for forecasting of long memory time series in the literature. Major limitation of ARFIMA models is the pre-assumed linear form of the model. Since many time series in real-world have non-linear structure, ARFIMA models are not always satisfactory. Both theoretical and empirical findings in literature show that combining linear and non-linear models such as ARIMA and artificial neural networks (ANN) can be an effective and efficient way to improve forecasts. However, to model long memory time series, any hybrid approach has not been proposed in the literature. In this study, a new hybrid approach combining ARFIMA and feedforward neural networks (FNN) is proposed to analyze long memory time series. The proposed hybrid method is applied to tourism data of Turkey whose structure shows dominantly the characteristic of long term. Then, this hybrid method is compared with other methods and it is found that the proposed hybrid approach has the best forecasting accuracy.