Research On Financial Early Warning Of Listed Companies Based On Neural Network

As a typical representative of the real economy,if a listed company has a financial crisis,it will not only cause great harm to itself,but also seriously damage the interests of investors,and may even have an impact on the entire economic environment.The formation of financial crisis is a gradual process,so it is necessary to adopt scientific and effective methods to early warning and prevent the financial crisis of listed companies in my country.Among many financial early warning models,the neural network model has the unique advantages of high accuracy and strong adaptability.Therefore,this paper uses the neural network method to establish the enterprise financial crisis early warning model,and combines the optimization of the financial early warning index system,and then obtains a better forecast effect.In terms of the selection of financial early warning indicators,after reviewing relevant research at home and abroad,this paper finds that most of the research on whether an enterprise is in financial crisis is based on financial indicators(generally including 10-20 indicators),while non-financial indicators focus on Few,especially related indicators related to technology research and development related to the sustainable development of enterprises.This paper fully considers the factors that affect the financial crisis of listed companies,selects more financial indicators,expands the indicator system of the early warning model,and especially introduces three indicators related to enterprise technology research and development(this study is regarded as non-financial indicators),established a financial early warning indicator system consisting of forty-three financial indicators and five non-financial indicators.This paper collects the index data of more than 2,000 A-share listed companies from 2015 to2020,uses the Python3.7 development platform,applies the artificial neural network method,compiles programs to process the relevant data,and then conducts repeated training,adjusts the parameters according to the feedback,and finally A 3-layer neural network model is established,which includes input layer,hidden layer and output layer,and the specific node structure of each layer is 48-16-2.In the process of evaluating the performance of the neural network early warning model,most studies only use the accuracy rate as an evaluation index.This paper establishes a comprehensive evaluation system of model performance including recall,precision and accuracy.According to this evaluation system,the financial early warning model established in this paper is used to predict 100 test samples,including 50 normal enterprises and 50 financial crisis enterprises.The test results show that the overall prediction accuracy of the model reaches 86%,and the F1 score of the comprehensive evaluation index of model performance reaches more than 93%.A longitudinal comparison of the prediction results of enterprises that have experienced financial crisis shows that the closer the data to the year of financial crisis occurred,the higher the accuracy of the early warning results.Through comparative experiments,it is found that the performance of the model with non-financial indicators related to R&D is better than that of the model without such non-financial indicators.This paper selects the W company that was specially treated(ST)by the China Securities Regulatory Commission in April 2021 as the case company in this paper.Although the W company may have modified its financial statements before,especially the 2020 report data performed well.However,after inputting the 48 indicator data of W company in 2019 and 2020 into the financial early warning model constructed in this paper for simulation experiments,the results are all shown as financial crisis.From the analysis results,the financial early warning model based on neural network can achieve good results.