Time Series Prediction of Stock Index Return Rate Based on DBNs Model

Stock market prediction is very complex and this challenging task is a hot topic of intensive study.This paper uses Dynamic Bayesian networks(DBNs)which is the timing expansion of Bayesian network(BN)to implement the time series prediction of the stock index return rate.DBNs are able to learn interdependent probability relations of variables and the laws changed with time.The latent information curtained in time series is captured with the hidden variable set in DBNs.The prediction method developed in this paper is based on the stock psychological analysis technology and makes use of the real world data of the Chinese stock index.The historical data of daily return rate of Shanghai Stock Exchange composite index are used for training.The successfully trained model can reach a high hit rate of 80.12% for the discrete prediction.When Gaussian mixture model(GMM) distribution is employed for continuous forecast,the mean absolute percentage error(MAPE) of proposed model is lower than 1% compared to BP neural networks and GARCH-BP neural networks,and cumulative errors are steady.The experimental results show that the proposed model has good predictability and stability in the market circumstances of high noises.