Research On Credit Bond Default Risk And Contagion Effects Based On Complex Networks

Frequent defaults in the bond market have had a severe impact on the normal operation of China’s economic system.Especially since 2020,defaults have occurred in high-rated bonds,breaking investors’ faith in state-owned enterprise credit bonds.In March 2021,the State-owned Assets Supervision and Administration Commission of the State Council issued "Guiding Opinions on Strengthening the Risk Control of Local State-owned Enterprise Debt" to urge local state-owned assets supervision and administration commissions to strengthen debt risk prevention and control.Therefore,accurately identifying default risks in the bond market is of great significance for the stable operation of the financial market.Due to the complex relationships between issuing companies,this article focuses on the following core issues:whether these relationships will affect individual company risks,how to extract effective information from these relationships to enrich the existing default risk prediction index system,how to construct a highperformance default risk prediction model that better fits the data characteristics,and how to grasp the contagion effect of risk events in these relationships and their influencing factors.These issues are also important challenges that need to be overcome in the context of frequent defaults in the bond market.To answer these questions,this article uses complex network technology to study credit bond default risk prediction and risk contagion.Firstly,a complex network of issuing companies’ relationships based on supply chain cooperation is constructed,and network position features and association risk features are extracted.Secondly,the existing credit bond default risk prediction model is improved based on the concept of imbalanced concept drift.Finally,the improved SEIRDS model is used to simulate the credit bond risk contagion effect.Specifically,the main work and conclusions of this article include the following parts:Firstly,this article constructs a complex network based on time-varying supply chain cooperation relationships.Considering the supply chain cooperation relationships between issuing companies,this article extracts network position features and association risk features based on complex network technology,and studies the impact of network features on credit bond default risk.The empirical research results show that:1)From the perspective of network position features,the higher the centrality index of issuing companies in the network,the lower their default risk,indicating that the local breadth,global depth,intermediary degree,eigenvector centrality,and network status of companies in the time-varying supply chain network have a certain inhibitory effect on the formation of default risk.2)From the perspective of network association risk,the higher the association risk index of issuing companies,the higher the credit bond default risk,indicating that risk will spread along association relationships.Therefore,constructing a default risk warning model based solely on a company’s own financial index system is not enough,and the risk contagion between company association relationships should be considered.Furthermore,this article uses credit ratings as another way to measure default risk and conducts a robustness test of the results in this chapter.This part of the research creates a new perspective for credit bond default risk prediction,enriches the index system for credit risk assessment,and has high theoretical and practical value.Secondly,this article constructs an improved credit bond default risk prediction model based on imbalanced concept drift data streams.Considering that the number of default samples in the credit bond data stream is less than that of non-default samples,and there is a possibility of sample distribution changes in the time-varying credit bond data stream,this article proposes an adaptive network neighbor-dynamic ensemble updating-XGBoost default risk prediction model based on complex network technology.Furthermore,this article introduces the network position indicators and network association risk indicators proposed in Chapter 4 into the existing credit bond default risk prediction index system,and evaluates the model performance based on indicators such as AUC,G-mean,and two-class error rate.The empirical results show that:1)The performance of the proposed adaptive network neighbor imbalanced sampling technology is better than that of existing imbalanced sampling technologies;2)The proposed concept drift processing technology can effectively improve the model’s performance;3)More importantly,the network position features and association risk features proposed in Chapter 4 can effectively improve the model’s predictive performance.This part of the research improves the credit bond default risk prediction index system and method system,and has strong innovation.Thirdly,this article constructs an improved SEIRDS default risk contagion model.Based on the results of Chapters 4 and 5,this part constructs an improved SEIRDS infectious disease model to simulate and analyze the contagion effect of credit bond default risk and its influencing factors.Different from traditional infectious disease models,this article analyzes based on real supply chain networks and predicted default probabilities during the simulation process.Firstly,this article sets the enterprise latent rate related to the global risk exposure value proposed in Chapter 4.Secondly,this article uses the credit bond default risk identification model constructed in Chapter 5 for prediction,and calculates the default risk of issuing companies by weighted calculation of the prediction results.This default risk can be used as the probability of infecting risk for issuing companies.This part of the research derives the transmission threshold of the improved SEIRDS infectious disease model through theoretical analysis,explores the impact of different initial shocks,different cure rates,and different bankruptcy rates on the number of different state nodes in the SEIRDS model’s steady state,and analyzes the stable state of risk contagion under different rescue strategies.The empirical results show that:1)When the initial shock in the bond market is large,the number of infected nodes in the system increases,and the entire system faces a greater crisis;2)The increase in the cure rate greatly reduces the number of infected nodes in the system’s stable state;3)The increase in bankruptcy probability leads to a doubling of the number of bankrupt companies;4)The theoretical derivation and empirical research both confirm that the target rescue strategy is more effective than the balanced rescue strategy.Therefore,accurately understanding the relationship between companies is of great significance for formulating corresponding risk rescue order and strategies.This article mainly innovates in the following four aspects:Firstly,this article proposes a credit bond default risk assessment method based on the inter-enterprise correlation network.Traditional credit bond default risk models mostly predict default risk based on individual financial indicators of enterprises,which have limitations such as neglecting correlation risk and insufficient correlation feature mining.As enterprises are a group with correlation relationships,risk will spread and replicate along the correlation relationships.Therefore,predicting default risk based on individual risk indicator systems has significant limitations.To address these limitations,this article considers the interenterprise supply chain cooperation relationship and constructs an inter-enterprise supply chain correlation network based on complex network technology,extracting network position features and network association risk features.In the extraction of complex network association risk features,this article improves the restart random walk algorithm and proposes a global risk exposure indicator.Firstly,the starting point of the algorithm is set as the node where the risk event occurs,and secondly,the starting point vector is weighted adjusted based on degree centrality.The improved restart random walk algorithm can measure the correlation between nodes and high-risk nodes,and this article uses it as a node’s global risk exposure indicator.The extracted network position features and network association risk features are included in the credit bond default risk indicator system,enriching the credit bond default risk assessment theory and improving the credit bond default risk indicator system.Secondly,this article establishes a more comprehensive network of interenterprise supply chain relationships.Most existing research on supply chains is based on the top five suppliers and customers of listed companies.However,nonlisted companies account for a relatively large proportion in China,and studying only listed companies may result in sample bias for the entire economic system.The research object of this article is all suppliers and customers of credit bond-issuing enterprises.The dataset of this article is more comprehensive,reflected in two aspects.First,the research dataset of this article includes all bond-issuing enterprises,including both listed and non-listed companies.Second,the information collected on enterprise suppliers and customers in this article is more comprehensive and complete.This article collected the top five suppliers and customers of bond-issuing enterprises,and manually sorted out all bidding information published on public websites such as the China Government Procurement Network,local government procurement networks,local public resource trading centers,and company official websites.Therefore,the complex network constructed in this article can better comprehensively depict the supply chain relationships between economic entities,enriching the data scope of existing research on inter-enterprise relationships.Thirdly,this article establishes a higher-performance credit bond default risk prediction model.In constructing the credit bond default risk prediction model,considering that the number of default samples is less than that of non-default samples,this article proposes an imbalanced sampling method that considers the density of default samples and network position.This method can correctly identify noise in default samples and effectively utilize the information of default samples.At the same time,considering the distribution changes that may occur in time series data streams,the classifier weights are assigned based on the Hellinger distance between the old and new data streams of credit bonds,and the sample weights and classifier weights are updated based on the classification results of the new data stream samples.Furthermore,the network features extracted based on complex network technology are added to the existing credit bond default risk indicator system,and an adaptive network neighborhood-dynamic ensemble update-XGBoost credit bond default risk prediction model is proposed.Empirical research results show that the proposed model has good default risk prediction performance,and network features have a certain improvement effect on the prediction performance of the model.Therefore,this research enriches the theory of credit bond default risk assessment,improves the construction of credit bond default risk indicator system,and provides new perspectives and methods for credit bond default risk warning.Fourthly,this article provides a new idea for the study of risk contagion effects under complex network technology.Traditional infectious disease models simulate networks as randomly generated small-world or scale-free networks,and the initial infected samples are randomly selected sample sets,with the infection rate set to a random initial value.This research is different from traditional infectious disease models,specifically,the simulation process of this article depends on the real supply chain network,and the initial shock nodes are real credit bond risk events,and the node infection rate is the weighted default probability obtained from the default risk prediction model.This makes the simulation results of the infectious disease model more in line with the transmission process of real risk shocks.This article depicts the stable distribution of nodes in different states after risk shocks in simulation,and explores the changes in the distribution of stable state nodes under different risk shocks,cure rates,and bankruptcy rates,analyzing the impact of different rescue strategies on the stable state of risk contagion.This provides new ideas and methods for the study of risk contagion under the perspective of complex networks and has strong innovation.This article conducts a systematic analysis and empirical research based on the integrated research idea of "complex network feature extraction-default risk identification model construction-risk contagion effect analysis" to study credit bond default risk prediction and risk contagion issues under complex network technology.It constructs a high-performance risk prediction model and a risk contagion model that is more in line with reality.This research contributes to improving the ability to identify credit bond default risks,providing important references for regulatory authorities to formulate relevant risk prevention policies,and has strong innovation and practical significance.