Research On Enterprise Credit Risk Early Warning Under Complex System Framework

In recent years,the non-performing loan ratio of China's commercial banks has been at a high level,making many commercial banks and even the entire financial market face higher credit risks.In order to fundamentally reduce the credit risk faced by commercial banks and reduce the non-performing loan ratio,it is necessary to give early warning of the possibility of credit risk arising from the company's future repayment process before granting loans.From the viewpoint of system theory,many companies are closely linked with each other and constitute a complex system and each individual entity is a subsystem of this complex system.Therefore,it is necessary to study the company credit risk from the perspective of a complex system.Complex networks and neural networks are important tools for studying complex systems.Although existing research uses complex networks to do a lot of work on credit risk contagion,it is basically done from the network topology itself,without considering the mutual influence of the infectivity of the credit risk between related companies.In addition,traditional neural networks tend to fall into saddle points during the convergence process.How to determine whether the neural network falls into the saddle point and how to jump out of the saddle point is a research difficulty.This dissertation takes the listed companies in our country as the research object,and uses the evolutionary neural network theory and complex network theory to study the enterprise credit risk early warning model from the perspective of the complex system.In the course of the study,a comprehensive consideration of the internal and external factors of the company and the relationship between individuals.The internal factors,the external environmental factors,and the relationships between individuals are complex nonlinear relationships.Therefore,they must be studied with nonlinear tools.The internal factors of an enterprise are mainly its various financial indicators.It is a classification problem in machine learning whether the enterprise is alerted to credit risk through numerous financial indicators.The evolutionary neural network integrates the advantages of genetic algorithms and neural networks and is suitable for classification prediction,but there is a problem that can easily converge on the saddle point.To solve this problem,this dissertation presents a new evolutionary neural network algorithm-Resume Evolutionary Neural Network(RENN)algorithm.The algorithm builds “performance volatility” indicators on the basis of on-line performance and off-line performance to determine whether the neural network converges to the saddle point.In addition,under the premise of taking into account the optimization tendency and robustness,this algorithm eliminates the current father,selects the new father needed for evolution from the best historical individuals,and jumps out of the saddle point for subsequent optimization.The external environmental factors of the enterprise and the relationship between individual enterprises have a direct impact on the spread of credit risk among enterprises,the theory of communication in complex networks can provide a good description of the mechanism of transmission of credit risks among enterprises.This dissertation proposes to use the mutual information entropy coefficient to measure the nonlinear and non-stationary relationship between different stock time series,and uses this coefficient to construct a complex network of enterprise credit risk.It builds a complex system to analyze the impact of external environmental factors on enterprise credit risk.On this basis,the use of complex network communication theory to study the path dependence of enterprise credit risk and snowball effect in the process of credit risk transmission,analysis of the impact of dissemination ability on enterprise credit risk,to get the enterprise credit risk infection ability and its ability to spread credit risk is equivalent to this conclusion.On the basis of community detection,comprehensively consider the impact of community and weight on enterprise credit risk,redefine the proportion of credit risk in different propagation paths of the node,and solve the problem that the original PageRank algorithm does not consider weights and does not apply to undirected networks.The CommunityRank algorithm was proposed,which measures the enterprise's ability to infect credit risk.Based on the study of internal factors,external environmental factors,and the effects of inter-enterprise relationships on enterprise credit risk,this dissertation uses evolutionary neural network theory and complex network theory to comprehensively analyze the influence of internal and external factors on its credit risk from the perspective of complex system theory.The analysis results are used as the enterprise credit risk index,the CI index.The empirical results show that the company's credit risk index can effectively give an early warning of enterprise credit risk.