Multi Scale Nonlinear Ensemble Model for Foreign Exchange Rate Prediction

This paper proposes a novel multi scale nonlinear ensemble methodology for analyzing and modeling the complex exchange rate behaviors. Using several techniques integrated under the proposed unified framework, it deals with data characteristics such as autocorrelation, multi scale heterogeneity and parameter instability during the modeling process. The multi scale heterogeneity property is modeled using wavelet analysis while autocorrelation property is modeled under ARMA framework. Combining independent component analysis, the proposed approach improves the model specification stability using support vector regression based nonlinear ensemble framework. Euro market is chosen as the test case for the performance evaluation of the proposed approach. Empirical studies results suggest that the proposed approach improves the forecasting accuracy and stability. It also offers valuable information as to the underlying micro market structure.