Constraint bagging for stock price prediction using neural networks

This paper proposes a novel constraint bagging forecasting method for stock price prediction. In the proposed approach, each of predictors is firstly constructed by training on a set of samples produced by bootstrapping using neural networks. Secondly, to improve the forecast capability of the bagging method, the constraint conditions are set to select competitive predictors. Finally, with the appropriate predictors, ensemble paradigm is applied to forecast stock prices. During the process, the network architectures of neural networks are discussed, and the optimal network architecture and proper parameters of the proposed model are determined by grid search method. To validate the proposed method, Dow Jones Index (DJI) is used for evaluation. The results show that constraint bagging outperforms traditional bagging and neural network predictor. These findings imply that constraint bagging is a promising approach for stock price prediction.