FORECASTING NIKKEI 225 INDEX WITH SUPPORT VECTOR MACHINE

<正>Support Vector Machine (SVM) is a very specific type of learning algorithmscharacterized by the capacity control of the decision function, the use of the kernel func-tions and the sparsity of the solution. In this paper, we investigate the predictability offinancial movement direction with SVM by forecasting the weekly movement direction ofNIKKEI 225 index. To evaluate the forecasting ability of SVM, we compare the perfor-mance with those of Linear Discriminant Analysis, Quadratic Discriminant Analysis andElman Backpropagation Neural Networks. The experiment results show that SVM out-performs other classification methods. Furthermore, we propose a combining model byintegrating SVM with other classification methods. The combining model performs thebest among the forecasting methods.