Forecasting the Movement of International exchange rate with Support Vector Machines

Forecasting financial time series is an important and complex problem in machine learning and statistics. This paper examines and analyzes the general ability of Support Vector Machine (SVM) models to overcome weak-form market efficiency by predicting and trading daily RMB/USD exchange rate return directions. For this purpose, ClusterSVM models with Gaussian kernels along with one conventional SVM are compared to investigate the separability of Granger-caused input data in high dimensional feature space. Experiment results indicate that ClusterSVM method outperform conventional SVM method for in predicting RMB/USD return directions.