Research On Data-driven Stock Trend Prediction And Dynamic Portfolio

With the continuous development of China’s economy and the improvement of its stock market system,an increasing number of individuals are showing interest in stock market investments.Predicting the future trends of stocks not only yields lucrative returns for investors,but also serves as a risk management tool for enterprises.The abundance of multi-source heterogeneous data containing the reasons behind the market’s unpredictable nature,along with the development of artificial intelligence and big data,offers the possibility of predicting stock market trends.Portfolio management is a crucial research direction in the field of financial management,and this paper aims to explore how to apply the prediction results of stock trends to portfolio strategies,ultimately improving portfolio returns while effectively controlling risk.In this study,we have first extracted features from multi-source heterogeneous data via feature engineering,integrating technical analysis indicators of the stock market for technical feature construction and enhancement,thereby obtaining feature sets for A-share stock market cycle classification,stock volatility,and trend classification.We have then employed a data-driven approach to iteratively train the market cycle classification model,individual stock volatility classification model,and stock trend prediction model.Based on the prediction results of market cycle,stock volatility,and stock trend,we have designed a dynamic portfolio strategy,updating the underlying and positions of the portfolio on a rolling basis,and utilizing all A-share market stocks from January 2017 to December 2021(excluding some delisted stocks and those with shorter trading time).Our results have demonstrated the efficacy and necessity of feature engineering based on multi-source heterogeneous data for predicting stock market patterns.Additionally,we have discovered that iterative training with time-rolling sampling can improve the generalization ability of the tree model,better matching the prediction order of the time series.The dynamic portfolio strategy,considering market cycle classification,individual stock volatility,and trend classification,has been shown to effectively control portfolio risk while simultaneously improving portfolio yield.Predicting stock trends can help companies develop better investment strategies,enhancing the efficiency and accuracy of their financial management and providing insights for risk management.