Research On Financial Early Warning Model Of Listed Companies On Science And Technology Innovation Board

The Science and Technology Innovation Board is a new sector in the Chinese stock market,and its listed companies are mainly engaged in the research and development and production of high-tech and strategic emerging industries.These companies have good development potential and prospects,but they will also face higher risks,especially the financial risk deserves attention,so it is necessary to carry out early warning and prevention of their financial risks.In addition,the information disclosure system under the registration system is more perfect,and various text information is increasingly rich,so the introduction of text characteristic information can build a comprehensive financial early warning index system and optimize the performance of the model.In this paper,82 new ST companies and 246 corresponding non-ST companies on the Science and Technology Innovation Board during 2021～2022 are taken as research objects,and the financial data from 2019～2020 are selected as data samples.Text mining technology is used to extract text characteristic indicators and incorporate them into the financial risk early warning model,aiming to study the influence and function of text characteristic indicators on the financial risk early warning model of listed companies on the Science and Technology Innovation Board,and compare the prediction accuracy between XGBoost,Bayesian network and LSTM model,Logistic and ANN classical models.To explore the financial early warning model suitable for the listed companies on the Science and Technology Innovation board.This paper analyzes the empirical results and draws the following conclusions:First,Kolmogorov-Smirnov test and Mann-Whitney U test are used to test the significance of the financial early warning index system.It is found that in the financial early warning of the Science and Technology Innovation board,the most significant contribution is the profitability index,followed by the growth ability index.Second,when the pure financial index and text characteristic index are introduced respectively to build the model,it is found by comparing the AUC value and F1-score value of Logistic model,ANN model,XGBoost model,Bayesian network model and LSTM model.The AUC value and F1-score value of the five models have been greatly improved when the text feature index is added.Meanwhile,the feature importance analysis of the financial early warning index system is carried out.The results show that the constructed text feature index stock bar investor comment(F3)has the greatest impact on the financial early warning of the listed companies on the Science and Technology Innovation Board.Among the five models,the prediction results of XGBoost model and Bayesian network model are more stable and accurate.