Research for construction and application of PCA-SVM for exchange rate forecasting

Abstract

The traditional SVM method has the problem of kernel function's parameters and dynamic optimization of penalty coefficient C. This paper constructs a hybrid model by extending the SVM method with PCA method to solve the problem. Finally we use the daily date of the exchange rate to test the high prediction accuracy of PCA-SVM model. In order to achieve better prediction accuracy, four kernel functions are used to construct different SVM. The empirical results show that SVR based on RBF kernel has the highest prediction accuracy. This result illustrates that the relevant government can take use of the model to monitor the smooth fluctuations in the exchange rate market.