| Traditional approaches to time series forecasting such as Box-Jenkins (ARIMA) models assume that the time series under study are generated from linear processes. The limitation of the linear models lies in that most real world systems are often nonlinear and the approximation of linear models to complex nonlinear relationships will not be satisfactory. Although some nonlinear models were proposed during the last decade, the pre-specification of the model form restricts the usefulness of these models. Recently, there is an increasing interest in forecasting using artificial neural networks (ANNs). ANNs, with their unique features of adaptability, nonlinearity, and arbitrary function mapping ability, provide a promising alternative tool for forecasting.; Interest in using ANNs for forecasting has led to a tremendous surge in research activities in the past decade. Yet, researchers to date are still not certain about the effect of key modeling factors on performance. The lack of systematic approaches to ANN model building is probably the primary cause of inconsistencies in reported findings.; The overall research objective of this dissertation is to have a thorough and systematic investigation of the application of ANNs for forecasting. The study first integrates previous research and identifies key issues in modeling ANNs for time series forecasting. Then the effects of several major factors on the modeling and forecasting performance of neural networks are examined through an extensive simulation study. These factors are the number of input nodes, the number of hidden nodes, and the training sample size. Sixteen types of linear and nonlinear time series with 30 replications for each type are generated. The results from ANNs are also compared to those with Box-Jenkins models. Finally, an application of neural networks to foreign exchange rate forecasting is employed to illustrate the use of ANNs to real-world problems. The robustness of neural networks with regard to sampling variation is investigated by using cross-validation methods. |
