Affiliations: Department of Industrial Engineering, Isfahan University of Technology, Isfahan, Iran
Note: [] Corresponding author. Mehdi Khashei, Department of Industrial Engineering, Isfahan University of Technology, Isfahan 84156-83111, Iran. Tel.: +98 311 3912550 1; Fax: +98 311 3915526; E-mail: [email protected]
Abstract: Improving forecasting especially time series forecasting accuracy is an important yet often difficult task facing forecasters. Both theoretical and empirical findings have indicated that integration of different models can be an effective way of improving upon their predictive performance, especially when the models in the hybridization are quite different. In the literature, several hybrid techniques have been proposed by combining linear and nonlinear models, in order to overcome the deficiencies of single models and yield results that are more accurate. However, recent research activities in hybrid linear and nonlinear models indicate that these models have two basic limitations that have decreased their popularity for time series forecasting. These two basic limitations are: (a) the hybrid linear and nonlinear models have some assumptions that will degenerate their performance if the opposite situations occur, and (b) the hybrid linear and nonlinear models require a large amount of historical data in order to produce accurate results. In this paper, a novel hybrid model is proposed for time series forecasting by combining linear autoregressive integrated moving average (ARIMA), nonlinear artificial neural networks (ANNs), and fuzzy models. In the proposed model, no prior assumption of traditional hybrid linear and nonlinear models is considered for the relationship between the linear and nonlinear components. In the proposed model the data limitation of traditional hybrid linear and nonlinear models is also lifted through investing on the advantages of the fuzzy models. Empirical results of financial markets, especially exchange rate market, forecasting indicate that proposed model performs significantly better than its components used separately, traditional hybrid linear and nonlinear, and other fuzzy and nonfuzzy models in incomplete data situations.
Keywords: Artificial neural network (p, d, q), fuzzy logic and models, time series forecasting, incomplete data, financial markets, exchange rate
