Abstract
This article analyses the historical statistics of the Shanghai Composite Index over the last 25 years, namely the characteristics and influencing factors. The article also predicts the trend of a 60-day time sequence of Shanghai composite index after 30th, Nov, 2018, using BP neural network model. The BP model fits the historical statistics well after consecutive training. The prediction result of the BP neural model indicates three characteristics of the trend in next 60 days: (1) the price will go up; (2) fluctuation frequency will reduce; (3) fluctuation range of the price will remain low.By analysing the relevant daily data of Shanghai Composite Index from 1993 to 2018, it is found that the forecast for Stock Market is very challengeable due to the intrinsic indeterminacy of the stock market and the complexity of the variables. To train and test the BP neural network model, this research adopts the trading record of 4465 transaction records as the training group, the next 935 trading days as the testing group and 935 trading days as verification group. The input variable in this BP neural network is Shanghai Composite Index (SZCindex), and the expected Shanghai Composite Index (EXSZCindex) is output variable. The BP neural network model is constructed as a three-layer neural network model.In the analysis of the daily closing price of the Shanghai Stock Exchange Index, this paper uses the goodness of fit to evaluate the BP neural network model and to continuously adjust the weight of the parameters. The results show that the training goodness of fit is 0.99898, and the correction goodness is 0.99897. The test goodness is 0.99904, and the overall goodness of fit is 0.99899, which indicates that the BP neural network has a rather good fitting effect. It is predicted that the price of the next 60 days from November 30, 2018 will rise to the highest point of 2623.874.In the end, the rationality of the predicted results is analysed in this paper. Via the analysis of the obtained data, it is found that there is no over-fit. The images of predicted values and real values are drawn by using the data from subjective and objective analysis.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have