The Research Of Stock Composite Index Prediction Based On Optimized BP Neural Network And Granular Computing

According to effective market hypothesis, we know stock price is random and unpredictable. If this is true, we cannot benefit from prediction with technical skills. But, lots of evidence proves that the market is not effective. First, rational investor hypothesis is suspected. In this theory, we assume that all investors respond to public market information available alike at the same rate which is not true in reality. Second, according to lots of studies, there are correlations between economical figures and important events and existences of strong price trends. So, stock price is predictable on some level. As we don’t know how many factors work on stock price, it is difficult for us to predict stock price with traditional method. Neural network is a good method to fit non-linear function. So it is good to predict stock price.This paper predict stock price with most popular BP neural network. According to the result, author finds some weakness. To solve the problem, author optimizes BP neural network with GA. Comparing the results of two experiments, we find the optimization is effective. In this paper author brings in granular computation to solve long-term prediction problem.The conclusions of this paper are(1) The result of prediction is good demonstrates that Chinese market is predicable, and Chinese market is not efficient market on some level. (2) The fact that optimizing BP neural network with GA makes the result better demonstrates that the optimization is effective.(3) Combining granular computation with optimized BP neural network makes the result good demonstrate that granular computation is effective in financial forecasting.