Research On Stock Price Prediction Based On RCI And GC-GAN

As a financial product with high risk and high yield,the rise and fall of the stock has been concerned by the majority of investors.Because it is easily affected by many factors such as political factors,economic policies,investor sentiment and public opinion,stock time series data has the characteristics of nonlinear and non-stationary.Research on stock price prediction has always been a hot and difficult topic in the field of finance and computer.In view of the problem that the stock prices of related companies in the same industry may rise and fall with each other,considering the relationship between the changes of stock prices predicted by the target and the whole stock market,the stock prices of related companies in the same industry may cause the changes of the stock prices of target companies,this thesis proposes to introduce related company information(RCI)as the input feature of the model.In order to eliminate the differentiation of prediction results between industries and models,multiple industries and models were used for empirical analysis to verify the impact of fusion RCI on the stock price prediction of target companies.In stock prediction research,technical indicators are usually used as input features,and there may be information overlap and linear correlation among a large number of technical indicator data.In this thesis,PCA is proposed to reduce the dimension of technical indicators before model prediction.SPSS software is used to select the most important principal components.Explore the impact of PCA data dimension reduction on model prediction;GAN model is widely used in the field of image generation,and can be used to generate realistic images by the antagonism mechanism of its generator and discriminator.In this thesis,the core principle of GAN is applied to stock price prediction.In the benchmark GAN model,the generator usually uses LSTM model,and the above experimental results show that the GRU model with RCI is better.Therefore,this thesis uses GRU model as generator and proposes a generative adversarial neural network model based on GRU model as generator and CNN model as discriminator,which is referred to as GC-GAN model.In this thesis,from the three perspectives of feature expansion,data dimension reduction and GAN model improvement,based on the theoretical method of deep learning to forecast the stock timing data.The experimental results show that integrating the information of related companies in the same industry(RCI)as the input feature of the model is helpful to reduce the error of stock prediction,and provides a new idea for the selection of stock price prediction feature.PCA dimension reduction is helpful to improve the predictive performance of the GRU model.The prediction performance of GC-GAN model based on the fusion of RCI features based on GRU model as generator and CNN model as discriminator is superior to statistical model,machine learning and traditional GAN model.