Research on Stock Portfolio Construction Based on Bi-LSTM Neural Networks


  • Jiawei Li



Bidirectional long and short-term memory neural network, Entropy weight TOPSIS method, Monte Carlo method


 In recent years, the rapid globalisation of China's financial market has provided more opportunities for investors, but also brought about a more complex investment environment. This paper constructs stock portfolios based on Bi-LSTM neural networks, aiming to improve the accuracy of stock price prediction and the optimisation of investment portfolios using deep learning techniques. The theoretical part introduces the portfolio theory, including the mean-variance model and the capital asset pricing model, and explores the advantages of LSTM, Bi-LSTM and ATT-LSTM in processing time series data. The constituent stocks of CSI 300 index are selected in the research design part, and the stocks are screened using entropy weighted TOPSIS method and analysed based on the data from January 2018 to April 2024. The closing price and logarithmic return are predicted by constructing and using LSTM, Bi-LSTM and ATT-LSTM models, and then the trading strategies of EMA, MACD double conditions are determined, and the investment weights are determined by Monte Carlo method for the investment portfolio. The results of the empirical study show that the Bi-LSTM model has the optimal prediction performance, and based on the prediction data of the model, the trading strategy using the dual conditions of EMA and MACD achieves a higher investment return than the strategy using only MACD. In summary, this paper demonstrates the superiority of Bi-LSTM model in stock price prediction through empirical research, and proposes an effective portfolio construction method and trading strategy, which helps investors make more effective decisions in the complex market environment.


Yan Wenxin. Application of machine learning based in stock prediction[J]. Information Systems Engineering,2024, (04):40-43.

Zhu Huainian,Chen Zhuoyang,Binning.Robust nonzero and stochastic differential portfolio games for two persons under Heston model[J/OL]. Journal of Sun Yat-sen University (Natural Science Edition) (in Chinese and English),1-12[2024-06-04]. https: // doi. org/ 10. 13471/ j. cnki. acta. snus.2022A062.

Yang Chao,Mao Junkui,Yang Yue,et al. Multi-stage turbine transition state tip clearance prediction based on long and short-term memory neural network[J/OL]. Propulsion Technology,1-12[2024-06-04] .

Yu Qianqian,Zhang Lei,Hu Xiaoxiang. Research on screening algorithm of abnormal power usage behavior based on CNN and Bi-LSTM[J]. Electronic Design Engineering, 2024, 32(09):96-100. DOI:10.14022/j.issn1674-6236.2024.09.020.

Chenxi Wang. Research on ultra-short-term wind power prediction based on IWHO-CNN-ATT-LSTM[D]. Liaoning University of Engineering and Technology, 2023. DOI: 10. 27210/ d. cnki. glnju. 2023.000039.

Q. Zhang,L. Wang,J.S. Xing. Optimization of GRU neural network based on Monte Carlo method for cogeneration load forecasting[J]. Journal of Beihua University(Natural Science Edition),2 024, 25(04):545-551.

REN Peng,LI Chong,TAO Peng,et al. Vulnerability assessment of power grid nodes based on weighted entropy TOPSIS method[J]. Journal of Electric Power Science and Technology, 2019, 34(03):143-149.








How to Cite

Li, J. (2024). Research on Stock Portfolio Construction Based on Bi-LSTM Neural Networks. Journal of Computing and Electronic Information Management, 13(2), 37-41.

Similar Articles

1-10 of 118

You may also start an advanced similarity search for this article.