Financial Risk Warning For Listed Companies Based On Deep Learning

With China’s economy entering a high-quality development phase,the potential risks in the financial market,primarily composed of listed companies,are attracting increasing attention.Listed companies are important participants in the financial market,and their financial conditions have an undeniable impact on market stability and economic development.If some influential listed companies experience a financial crisis,not only will their own performance suffer,but it may also trigger market panic and a chain reaction,potentially leading to systemic financial risks.Therefore,accurate and effective financial risk warnings are of great significance to listed companies,investors,and regulatory agencies.This paper investigates a practical and important financial risk warning problem in the financial field.It constructs a warning model that uses the disclosed financial indicators and annual report text of listed companies to identify future financial risks,in order to assist investors and investment institutions in making decisions.There are many financial indicator data for listed companies,but some traditional warning models have a relatively simple structure,which still poses certain challenges in discovering important features that indicate financial risks.Secondly,how to more effectively mine valuable information from the annual report text of listed companies is also an urgent issue to be addressed.Finally,in past research that combined annual report text with financial indicators for financial risk warning,the models were often simply integrated,and failed to effectively integrate important information from financial indicator data with gain information from annual report text,which also affected the efficiency of the warning model.In response to the above problems,this paper proposes the CNN-LSTM+SW model and the CL-Bi LSTM model for financial risk warning.In the financial risk warning based on financial indicator data,this paper conducted experiments by constructing a CNN-LSTM+SW model and comparing it with traditional machine learning models.While using single-period financial data for financial risk warning,this paper also attempted to mine historical multi-period financial data using deep learning models.Secondly,in order to fully utilize the annual report text,this paper extracted text from the nine main chapters of the annual report and represented them,and constructed LSTM,Bi GRU,and Bi LSTM models to explore valuable information in the annual report text.Finally,in order to fully utilize the important information in financial indicators and annual report text,this paper designed the CL-Bi LSTM deep fusion model to combine them for financial risk warning.The experimental results of this study demonstrate that the annual report text contains information that is helpful for financial risk warning.However,the important information contained in financial indicators is more abundant,and using only annual report text for financial risk warning is not significantly effective.At the same time,in the financial risk warning based on financial indicators,the CNN-LSTM+SW model outperforms traditional models,indicating that using deep learning models to extract risk information from multi-period financial data has good results.When conducting financial risk warning based on financial indicators and annual report text,the deep fusion model comprehensively utilizes the two different input data,integrates the information extracted by different models,and the AUC value of the CL-Bi LSTM model can reach 0.94,achieving a better prediction effect than models based solely on financial indicators.This study proposes a financial risk warning fusion model,verifies the effectiveness of combining financial indicators and annual report text to predict financial risks,and enriches the research methods of financial risk warning for listed companies.