| Quantitative investment has been introduced to China for nearly 30 years.Among them,multi-factor stock picking is the most frequently applied stock picking strategy.In the context of the proliferation of data,how to tap into the appropriate factors is an urgent problem to be solved.With the development of artificial intelligence technology and the continuous improvement of financial theory,numerous machine learning models are widely used in the field of factor mining,but the existing models are not efficient in factor mining research and do not achieve the integration of mining and validation.Therefore,the deep research on factors is of certain practical significance.To this end,this thesis constructed a multi-factor intelligent mining system for stocks.The data module of the system realized the functions of data acquisition and storage management for stock factors;The management module could manage the logged-in users and their operation logs;The financial information management module realized the functions of analyzing and managing the company’s finances,allowing investors to better understand the financial situation and liabilities of individual stocks;The data display module realized the data visualization in the form of graphs for stock K-line charts and earnings line charts,facilitating investors to observe stock trends and key capital movements;The genetic trading module realized the functions of managing various factors and simulating trading verification for strategies;The data display module visualized stock K-line charts and return line charts in the form of charts to facilitate investors’ observation of stock trends and key capital movements;The genetic trading module realized the management of various factors,simulated trading verification of strategies,and stored the backtest results of each strategy for users’ reference and browsing.In this thesis,a GA-TGCN factor combination and prediction validation model was constructed based on deep learning theory.Firstly,the GA algorithm was used to encode the factors to randomly construct a multi-factor strategy,and a quantitative backtest model was used to validate the analysis.Then,the T-GCN time-graph convolutional neural network factor selection model was introduced.In order to make the model fully take into account the complex variation of factors and thus capture the global feature relationships,the concept of mathematical integration was introduced to integrate the state values of adjacent time slices in the model and replace the previous state values.A factor filtering model was also used to predict the best factor values within the strategy based on the existing strategy and to backtest it.Finally,the model was compared with other models in the same operational framework,and the results verified the effectiveness of the proposed model and its better performance in improving the prediction accuracy and the quality of the multi-factor strategy.The system used PHP,Python,My SQL and other development tools,based on B/S architecture to achieve full process development.The system combined with GA stochastic stock selection algorithm and improved T-GCN factor prediction model can better analyze and predict the impact of factors on strategy returns,which helps investors better select and determine investment strategies to achieve excess returns in the stock market. |
