Exchange Rate Prediction using Support Vector Machines

Abstract

Financial forecasting in general, and exchange rate prediction in particular, is an issue of much interest to both academic and economic communities. Being able to accurately forecast exchange rate movements provides considerable benefits to both firms and investors. This research aims to propose a decision support aid to these firms and investors, enabling them to better anticipate on possible future exchange rate movements, based on one of the most promising prediction models recently developed within computational intelligence, the Support Vector Machine. The economics of supply and demand largely determine the exchange rate fluctuations. Calculating the supply and demand curves to determine the exchange rate has shown to be unfeasible. Therefore, one needs to rely on various forecasting methods. The traditional linear forecasting methods suffer from their linear nature, since empirical evidence has demonstrated the existence of nonlinearities in exchange rates. In addition, the usefulness of the parametric, and nonparametric, nonlinear models, has shown to be restricted. For these reasons, the use of computational intelligence in predicting the Euro Dollar exchange rate (EUR/USD) is investigated, in which these previously mentioned limitations may be overcome. The methods used are the Artificial Neural Network (ANN) and the Support Vector Machine (SVM). The ANN, more specifically the Multilayer Perceptron, is composed of several layers containing nodes that are interconnected, allowing the neurons to signal each other as information is processed. The basic idea of the SVM is finding a maximum margin classifier that separates a training set between positive and negative classes, based on a discriminant function that maximizes the geometric margin. The model selection for the prediction models was chosen to be based on the bias-variance dilemma, which denotes the trade-off between the amount of variation within different estimators on different values of a specific data set (variation) and the difference between the estimator’s expected value and the true value of the parameter being estimated (bias). Experiments on the Mackey-Glass dataset and on the EUR/USD dataset have yielded some appropriate parameter ranges for the ANN and SVM. On theoretical grounds, it has been shown that SVMs have a few interesting properties which may support the notion that SVMs generally perform better than ANNs. However, on empirical grounds, based on experimentation results in this research, no solid conclusion could be drawn regarding which model performed the best on the EUR/USD data set. Nevertheless, in light of providing firms and investors the necessary knowledge to act accordingly on possible future exchange rate movements, the SVM prediction model may still be used as a decision-support aid for this particular purpose. While the predictions on their own as provided by the SVM are not necessarily accurate, they may provide some added value in combination with other models. In addition, users of the model may learn to interpret the predictions in such a way, that they still signal some sort of relevant information.