Prediction Of Credit Risk Contagion Of Listed Companies Based On Holding Network

Under the background of economic globalization,enterprises have gradually established an inseparable relationship.Companies are not only directly affected by their own development,but also limited by their partner support and competitors.In the course of a company's globalization,a company will often develop into multiple cooperations and cross-holdings to reduce risk.The assessment of a company must not only rest on the company's own balance sheet and profit statement,but also conduct a risk assessment on the company associated with the company.For example,in 2016,LeTV experienced credit risk,and the sudden decline of a large company affected a number of related companies and individuals in a credit risk crisis.In order to prevent the outbreak of such a chain of credit risk crises,the prediction of credit risk needs to be transferred from the company itself to its related companies.Therefore,in-depth understanding of a listed company from the aspects of the company's connected transactions and information about shareholders has certain needs and practical value.Complex networks are a relatively common method for studying connected transactions of companies.This method uses the relevant theoretical basis of graph theory and has a good effect in the field of credit risk transmission.However,the method of graph theory is mainly aimed at the characteristics of the network structure,and does not consider the specific neighbor information of a certain node.In view of this pain point,this paper uses the graph embedding method to take the top ten shareholders of listed companies as the research object,and uses the credit risk of its associated shareholders as the research content to predict the credit risk of listed companies.Predict the credit risk level of listed companies under the influence of their affiliates from more dimensions.This article takes the ten largest shareholders of listed companies as the research content,and takes the credit risk of its associated shareholders as the research target.First,the top ten shareholders of listed companies are sorted out,and a network of shareholding relationships is established to represent the shareholding relationship of a certain year.Then integrate the holding relationship networks of multiple years to build a holding relationship network of multiple years.Use popular graph embedding methods in recent years.On the basis of this network,its corresponding walking path and objective function are set.Finally,the risk prediction of listed companies is realized.The specific research work is as follows:(1)First of all,based on the information of the top ten shareholders of listed companies in each year,a shareholding network map is established.Then set the weight of the edge in the network according to its shareholding ratio.The integration of the holding network graph of multiple consecutive years is called a holding network graph of multiple years.The nodes of the same listed company in multiple years are connected,and the edge between the nodes of the same company is regarded as the change of the year,that is,the jump of the network.The result is a multi-year holding network diagram.(2)Use KMV model and Z-score model to calculate the risk of listed companies.The KMV model uses the default distance DD as the evaluation criterion,and the Z-score uses a four-variable formula for calculation.Because the indicators calculated by the two methods are not discrete data,the data of the two indicators are discretized into seven credit risk levels according to related literature.In addition,this article also introduces the default information table of listed companies to directly downgrade the credit risk level formed by the two methods.Finally,the three indicators are combined to form the credit risk assessment standard in this paper.(3)According to the multi-year shareholding network diagram proposed in this paper,targeted random walks are formed to form a unique walk sequence.A global label variable is proposed,and the network is targeted to learn based on the original negative sampling objective function,so that the resulting vector has better meaning in space.Finally,two classic graph embedding methods are compared to verify the effectiveness of the proposed method.(4)Using the method proposed in this article,the infection risk of epidemic situation is simulated,and it is found that the credit risk level of 30% of listed companies will be reduced.The industries most affected by the epidemic may be industry.The financial and real estate industries will also be affected.In contrast,China's telecommunications industry is more resistant to risks.