Stock Prediction Based On Convolutional Neural Network

The economy and finance are fast growing and integrating around the world,especially in the financial industry.With the increase in financial activities,the uncertainty is changing rapidly.To explore and master the laws of financial activities and predict their various possibilities in the future have become the main research contents of financial practitioners and researchers.This paper firstly sorts out and summarizes the current research methods on time series,and introduces the current situation and characteristics of China's stock market.Then the structure and theory of the CNN(Convolutional Neural Network)are introduced.After that,the time-series data of the stock market is standard aligned and graphically converted.The time-series data is converted into image.The labels are classified according to the average price in the next half year.Finally,the CNN is established to predict the time-series data.This specific research work includes the following two aspects:(1)The alignment of financial time-series data standardized and converted into images.The CNN is improved,and a CNN prediction model suitable for stock time-series is established.Therefore,two combination modes are constructed and applied to stock forecasting.(2)Further the method of image generation is improved so that the image contains more valid data.The relative value forecast is made for the constituent stocks of the CSI 300 and the CSI 500 Index.The combinations of two methods is used to simulativelycompare,which determines a feasible relative value prediction model.Through research and experiment,this paper finally obtains a stock market relative value decision model based on CNN algorithm.Based on the first half of 2018,The model achieved better than index gains in simulation experiments.That is,when the number of convolution layers is 3,the size of the convolution kernel is 23,the number of convolution kernels at each layer is 7,14,28,and the daily update image is used for analysis and adjustment,a relatively stable relative gain is obtained in the market of the CSI 300 and CSI 500.