Predict Stock Trends With Self-supervised Learning Based On Technical Data

Financial stock market is an important economic activity that all kinds of investors are keen to participate in,which has an important impact on economic development.In-vestors expect to invest in the financial stock market to obtain profits.Therefore,how to predict the future trend of stocks more effectively has become a task concerned by in-vestors.The traditional market trend prediction model is usually based on manual factors or features,which seriously depends on expensive professional knowledge.In addition,it is difficult to find the hidden features contained in the stock time series data,which will help to predict the stock market trend.This paper proposes a stock market trend prediction framework SMART,which takes the self supervised learning sequence coding model S3 E as the pre training model.Specifically,the model encodes and represents the stock tech-nology data sequence,and the representation is further jointly trained through multiple self supervised learning auxiliary tasks.The encoder in S3 E model is trained through multi task joint learning to encode and characterize the stock sequence data,and then the stock market trend is predicted based on LSTM and feed forward neural networks.Extensive experiments on both China A-Shares and NASDAQ stock datasets demonstrate that the features learned from stocks' daily data sequences are effective for stock market trend pre-diction.According to the prediction results of SMART framework,the investment return based on the model output is obviously better than other methods.After the research on the stock market prediction algorithm model based on self su-pervised learning is completed,how to efficiently manage and maintain the algorithm model and make it actually land on the industrial scene,that is,the landing process of the algorithm,is also a problem considered in this paper.For the scenario of algorithm landing,this paper designs a quantitative model management platform to automatically carry out the rolling training and prediction tasks of the model.At the same time,it makes a visual display of the evaluation and prediction of each model,so as to facilitate users to have an overall control over the business evaluation effect of the model.In conclusion,the core contributions of this paper are as follows:· This paper introduces self supervised learning technology into the field of finan-cial quantitative investment,proposes a set of self supervised learning framework smart based on technical data,designs three business-related self supervised learn-ing auxiliary tasks,and models the prediction task of stock market rise and fall in the financial market through multi task joint training.· This paper has conducted sufficient experiments on multiple data sets,compared various existing mainstream models,evaluated the accuracy and F1-score effects of SMART framework and its variants and other mainstream models in the task of stock rise and fall classification,and evaluated the cumulative rate of return,information coefficient and sharp rate that SMART framework can achieve in terms of business indicators,Sufficient and effective experiments show the effectiveness of the proposed method framework.· In order to effectively manage each quantitative model,this paper designs a set of quantitative model management platform to give investors an intuitive and three-dimensional model effect display,including the daily prediction of the model and the cumulative income of the quantitative strategy,so as to facilitate investors to compare the relevant differences between various models.The SMART financial stock market prediction framework proposed in this paper has achieved leading results in multiple data sets,and can achieve high annualized rate of return in the financial market,which is ahead of other methods,indicating that the self supervised learning technology has a good effect in the data encoding of stock financial series,It lays a foundation for the application of self supervised learning technology in the field of financial quantitative investment.