Stock market index prediction using deep neural network ensemble

In this paper, we put forward deep neural network ensemble to model and predict Chinese stock market index (including Shanghai composite index and SZSE component index), based on the input indices of recent days. A set of component networks are trained by historical data for this task, where Backpropagation and Adam algorithm are used to train each network efficiently. Bagging approach combines these component networks to generate ensemble, which reduces the generalization error. Indices of test examples are predicted with the model, and the trend predictions is calculated based on the predicted indices. Finally, relative errors between actual indices and predicted indices, as well as accuracy of trend predictions are computed to measure the performance of predictions. It turns out that the model proposed in this paper can partially model and predict the Chinese stock market. The accuracy of trend predictions of the daily barycenter, high, low are 71.34%, 74.15%, 74.15% respectively for Shanghai composite index and 75.95%, 73.95%, 72.34% respectively for SZSE component index. But the predictions on close are unsatisfactory.