Research On The Enterprise Credit Risk Based On The Multi-model Fusion Via The Genetic Algorithm Optimization

The development of modern enterprises has made outstanding contributions to the economic growth of China.However,while creating economic value,it is also accompanied by enterprise credit risks brought by uncertain factors such as political changes at home and abroad,fluctuation of economic environment and regulatory system.Typically,risk assessment is done using expert systems,scoring systems,thirdparty warranties,and long-term relationships with friendly customers.However,these management measures are often affected by human subjective factors,resulting in unnecessary losses,and then directly bring losses to economic subjects.Therefore,the control of enterprise credit risk has become an important topic in the financial field.Based on the analysis of the research status of enterprise credit risk prediction models at home and abroad,and combined with relevant theories,the financial index data of enterprises are modeled and predicted to complete the identification of credit risk enterprises.Firstly,the financial index data of 2026 listed companies in The Shanghai and Shenzhen A-share markets from 2017 to 2019 in the Guotaian Database are selected as modeling samples.Then,through the expert interview method and the operation characteristic analysis method of listed companies,the financial index system including solvency and operation ability is constructed,and it includes the risk level and cash flow two categories of indicators.Next,the irrelevant variables were eliminated by using hypothesis testing,and the final five indicators into the model were determined by principal component analysis.In constructing the model,to address the shortcomings of traditional prediction methods,such as multiple linear regression models with high requirements for data assumptions and poor interpretability of neural networks,this thesis uses five methods of logistic regression,GBDT(gradient boosting tree),XGBoost(extreme gradient boosting tree)and LightGBM(light gradient boosting tree)via stacking framework to build the fusion models,and the parameters of the individual models are optimized by using an improved genetic algorithm.Finally,it is compared with several other typical enterprise risk models such as the integration tree to illustrate the superiority of the method in this thesis.The main contribution of this study is to improve the financial index system used in the traditional enterprise credit risk early warning model,add the risk level and cash flow module indicators,so that the model can more clearly learn the enterprise portrait,and use principal component analysis,five characteristic indexes of development ability,risk level,asset-liability ratio,cash flow and receivables turnover are established.The fusion model is optimized by an improved genetic algorithm and the prediction accuracy is high.The research content of this thesis makes an effective exploration in the field of credit risk prediction,enriches the method system of enterprise credit risk research in this field,and has reliable theoretical and practical significance.