Prediction of exchange rates using averaging intrinsic mode function and multiclass support vector regression

Prediction of nonlinear and nonstationary time series datasets can be achieved by using support vector regression. To improve the accuracy, we propose a new model ‘averaging intrinsic mode function’ which is a derivative of empirical mode decomposition to filter datasets of an exchange rate, followed by using a new algorithm of multiclass Support Vector Regression (SVR) for prediction. Simulation results show that the proposed model significantly improves prediction yields of the exchange rates, compared to simulation of SVR model without filtering and multiclass.