The Mid-term Stock Index Forecasting Based On Neural Network Ensembles

This paper mainly studies intelligent mid-term forecasting of stock market. Three parts are included in the paper, using Principal Component Analysis(PCA) for data process, modeling with Artificial Neural Networks(ANN),proper size of the ANN is determined by an iterative pruning algorithm, the results of three ANN are ensembled.Some new results and presented, and to be used to solve some problems in stock forecasting and neural network ensembles. These new methods are very useful to the field of financial applications as follows:How to select proper input factors is very important .Through improving the input factors, eliminating the correlation among the inputs and simplifying the structure, the PCA-based ANN could accelerate the learning process and increase the convergence accuracy.Using one-hidden-layer ANN to forecast the mid-term trend of stock index. A pruning algorithm is employed to eliminate units and adjust the remaining weights in such a way that the network performance does not worsen over the entire training set.Using Genetic Algorithm to neural network ensemble. This method is simply to realize and easy to explain.Three linear combination of neural networks, optimal ensemble, average ensemble and weight ensemble are compared.Shanghai and Shenzhen Stock Index are taken as an example. The ANN ensemble output results are satisfied.