A Comparison of Ensemble Methods in Financial Market Prediction

Abstract

Financial time series prediction is always a focus point of researchers and practitioner for its available data and profitability. As recent studies suggest that the employment of ensemble algorithms may improve the performance of a base learner, a compound experiment for comparison of ensemble methods is designed and implemented to investigated the fact that whether the ensemble methods can be employed to improve the performance of the base learner in financial time series prediction. The empirical results suggest that ensemble algorithms are powerful in improving the performances of base learners in financial time series prediction. When compared with Random Subspace and Stacking, Bagging provides a more stable and better improvement. The iteration of ensemble algorithms should be adjusted according to the situation. Higher value of iteration may not always performs well for over fitting may occur.