Research Of Stock Quantitative Investment Based On Deep Learning

At present,the domestic quantitative investment field is developing steadily,and its application in the securities market has a wide range of needs.With the rapid development of deep learning,stock quantitative investment strategies based on deep learning are gradually becoming an interdisciplinary research hotspot.Compared with the method of formulating investment portfolios through technical analysis in the early stages of quantitative investment,the method based on deep learning can obtain more accurate stock price trends and help investors formulate more reasonable quantitative investment portfolios.However,the price of stocks is affected by many factors such as macroeconomics,exchange rates,news media,etc.,making the application of quantitative investment in the securities market quite challenging.This paper is based on deep learning to study quantitative investment,and the main work is as follows:(1)Aiming at the problem of short-term stock price prediction,this paper uses dilated convolution and the stock modeling method designed in this paper to help improve the model’s feature learning ability of stock time series data,and helps the model to focus on important features by adding a CBAM attention module.Borrowing from DenseNet,adding channel connections in the network to help improve feature propagation and feature reuse,finally builds the ADCNN model.Experimental test results show that the short-term prediction ability of ADCNN is significantly better than other comparison models.It has good generalization ability,certain profitability and risk resistance ability.(2)In view of the insufficient long-term predictive ability of many current models,this paper uses the PF-LSTM proposed in recent years as a basic unit to construct an encoding-decoding structure to improve feature learning capabilities,and at the same time,combine the Luong global attention mechanism to improve the long-term prediction effect.The PF-Seq2Seq model is used,and the experimental results show that PFSeq2Seq has outstanding long-term forecasting ability,good short-term forecasting ability,profitability and anti-risk ability are outstanding.