A Study Of Financial Risk Pre-warning Model Based On RNN-GRU Neural Network For Listed Companies Under The COVID-19 Background

The financial early warning model is one of the important components of enterprise financial management,which is a research-based tool to help enterprises predict and avoid financial risks in an information-based manner.The generation of corporate financial crisis is influenced by the superposition of internal and external factors such as organizational structure,management decisions,economic events,policies and regulations,geopolitical situation,and public events.In recent years,under the background of deepening risks of economic globalization,the generation of"black swan"events such as natural disasters,public health events,wars and geopolitical changes has continuously increased the demand for corporate financial crisis early warning capabilities,especially the global public health event novel coronavirus pneumonia(COVID-19).The outbreak of COVID-19 at the end of2019 had a huge impact on listed companies,and it is particularly important to consider the"epidemic"element in financial warning analysis.In this paper,we quantify the"epidemic"related descriptions in the annual reports of the sample companies into indicators by means of the sentiment dictionary method and analyze the financial and non-financial data of 132ST and~*ST companies by using a combination of non-parametric tests and factor analysis.At the same time,considering the dynamic changes in the production and operation results of the same enterprise in different periods,this paper takes into account the time series characteristics of the data when constructing the enterprise financial early warning model,and introduces RNN and GRU methods to achieve efficient prediction of the risk of becoming an ST-type enterprise,which is important for the improvement of the financial crisis early warning management system of listed enterprises in the post-epidemic era.In this study,14 financial warning indicators,including the"epidemic"indicator,were selected from 30 initial financial warning indicators,which were incorporated into the financial warning model for training,and the RNN and GRU methods were used to incorporate the time series characteristics of the data,and the resulting RNN＿GRU,RNN and GRU models were compared.It was found that the prediction accuracy of the models introducing RNN and GRU methods was generally higher than that of the models constructed by linear regression,plain Bayesian,SVM and BP neural networks,and the RNN＿GRU model had the best effect.