Time series forecasting with neural network ensembles: an application for exchange rate prediction
- PDF / 150,579 Bytes
- 13 Pages / 595 x 842 pts (A4) Page_size
- 70 Downloads / 188 Views
#2001 Operational Research Society Ltd. All rights reserved. 0160-5682/01 $15.00 www.palgrave-journals.com/jors
Time series forecasting with neural network ensembles: an application for exchange rate prediction GP Zhang1* and VL Berardi2 1
Georgia State University, Atlanta, GA, USA; and 2Kent State University, Kent, OH, USA
This paper investigates the use of neural network combining methods to improve time series forecasting performance of the traditional single keep-the-best (KTB) model. The ensemble methods are applied to the dif®cult problem of exchange rate forecasting. Two general approaches to combining neural networks are proposed and examined in predicting the exchange rate between the British pound and US dollar. Speci®cally, we propose to use systematic and serial partitioning methods to build neural network ensembles for time series forecasting. It is found that the basic ensemble approach created with non-varying network architectures trained using different initial random weights is not effective in improving the accuracy of prediction while ensemble models consisting of different neural network structures can consistently outperform predictions of the single `best' network. Results also show that neural ensembles based on different partitions of the data are more effective than those developed with the full training data in out-of-sample forecasting. Moreover, reducing correlation among forecasts made by the ensemble members by utilizing data partitioning techniques is the key to success for the neural ensemble models. Although our ensemble methods show considerable advantages over the traditional KTB approach, they do not have signi®cant improvement compared to the widely used random walk model in exchange rate forecasting. Keywords: neural network ensemble; exchange rate; time series; forecasting
Introduction Arti®cial neural networks have been widely used as a promising alternative approach to time series forecasting. A large number of successful applications have established the role of neural networks in time series modelling and forecasting. Neural networks are data-driven, self-adaptive nonlinear methods that do not require speci®c assumptions about the underlying model. Instead of ®tting the data with a pre-speci®ed model form, neural networks let the data itself serve as direct evidence to support the model's estimation of the underlying generation process. This nonparametric feature makes them quite ¯exible in modelling real-world phenomena where observations are generally available but the theoretical relationship is not known or testable. It also distinguishes neural network models from traditional linear models and other parametric nonlinear approaches, which are often limited in scope when handling nonlinear or nonstandard problems. The most popular neural networks used in forecasting are the single multi-layer feedforward model or multi-layer *Correspondence: GP Zhang, Department of Management, J Mack Robinson College of Business, Georgia State University, Atlanta, GA 30303, USA. E-m
Data Loading...
