The Research And Application Of Artificial Neural Networks In Stock Price Forecasting

Stock market forecasting is a highly nonlinear project,and neural network have advantage over traditional statistical method to deal with nonlinear problem due to its studies independently and super nonlinear fitting.This paper first analyzes the difficulties of the short-term stock prediction,and share price is influenced by many factors,therefore the model must take into account these important factors as much as possible.Many technical indicators(MACD,KDJ,BR,etc.)are the focus of attention of investors,and which should be calculated based on the daily stock data and be added to the data characteristics.Domestic scholars mainly use BP network and its improved algorithm on shortdated prediction of stock market.LSTM research is very rare in the domestic stock market forecasting,which is mostly used for natural language processing and has not been effectively applied to the stock market forecasting at present.In fact,LSTM is a recurrent neural network that is very suitable for time series analysis and prediction.Based on the forecasting model of LSTM,this paper uses Shanghai securities composite index to train and tune the network,and model uses the stock price data of the previous 14 days and its calculated technical indicators as the characteristics of the stock price to predict the closing price of the 15 th day.The final model prediction accuracy is 63.38%.The forecast value as a rise combine with forecast error value in the training process could be effective to investors who could produce a certain practical value with the data.At the same time,in order to verify the model is of generality for the A-share market index of universal,several empirical test of three A-share index : Shenzhen component index,HuShen 300 index and SSE SME composite has been conducted,the average prediction accuracy exceed 60%,which fully validated the effectiveness of the designed model.