Empirical Study Of Quantitative Stock Investment Based On Graph Representing Learning

With the development of artificial intelligence and other cutting-edge technologies,quantitative investment has also ushered in a new round of changes in this wave of technology.In recent years,we can see that the application of many deep learning models,especially the time series model,has brought continuous breakthroughs and progress in the field of quantitative investment.Its ability to use stock information indicators and powerful processing capabilities for sequence data brings a powerful forecast of stock trends,and also plays a key role in maximizing the profits of related investments.However,it must also be recognized that professional investors with considerable experience have accumulated considerable experience in understanding the market and their understanding of the value of stock investment.Such professional investors’ understanding of the intrinsic nature of stocks has an incalculable effect on the prediction of stock market trends..The current deep learning in the field of quantitative investment often only uses many stocks’ own technical indicators and rarely uses professional investors’ understanding of the intrinsic nature of stocks.This is very likely to be an important improvement point in the application of deep learning to quantitative investment.This paper mainly explores the change of stock prediction ability of deep learning model and the influence of this change on quantitative investment after adding professional investors’ grasp and understanding information of stock market.In the research,the LSTM model based on the common technical quantitative factors in the past research is used as the benchmark model,and the stock holding data regularly disclosed by many funds are used as the original information for professional investors to grasp the stock market.Based on the fund stock holding data,the corresponding bipartite graph network is constructed,and different static embedded representations of stock nodes are obtained by using deep walk,node2 vec and other graph representation learning methods.Finally,the dynamic embedded representation of the stock node is obtained through the comprehensive analysis of the embedded representation of the node and the return of the stock market,and the static embedded representation and dynamic embedded representation are added to the benchmark model respectively to compare and analyze whether there is embedded representation information and whether the embedded representation information is dynamic or not.Finally,according to the prediction of each model,the corresponding back test is carried out to further reflect the differences of various indicators of the investment returns with different embedded representations.The final experimental results show that the dynamic embedded stock node representation can improve the prediction ability of the model for the future trend of the stock to a certain extent,and can show a better portfolio in the back testing stage.