图嵌入在上市公司信用风险预测中的应用

Application of Graph Embedding in Credit Risk Prediction of Listed Companies

当今经济全球化背景下,几乎没有一家公司可以独善其身,在对公司进行信用风险评估时引入外部关联方分析显得尤为重要.受图嵌入方法的启发,从时间和空间双维度整合上市公司的持股关系,借助复杂网络理论,建立多年份持股网络图,并将改良的图嵌入算法引入模型,用于持股网络结构的学习.同时,结合公司自身财务数据和违约记录,运用KMV模型和Z-score模型评估公司的信用风险等级,从结构和内容两方面学习关联网络的节点信息,最后依据多年份持股网络图进行针对性的随机游走,对上市公司的信用风险进行分析和传染预测.还与两种经典算法Node2vec和Deepwalk进行对比分析,并预测新冠疫情对上市公司信用风险的影响,以验证本文方法的良好效果.

In the context of economic globalization, almost no company can stand alone. It is particularly important to introduce external related party analysis when assessing the company’s credit risk. Inspired by the graph embedding method,this paper integrates the shareholding relationship of listed companies from the two dimensions of time and space, builds a multi-year shareholding network graph with the complex network theory,and introduces an improved graph embedding algorithm into the model for the learning of shareholding network structure. Meanwhile,with the company’s own financial data and default records,the KMV model and Z-score model are used to evaluate the company’s credit risk level,and the node information of the associated network is learned from both the structure and content. Finally,based on the multi-year shareholding network diagram,a targeted random walk is performed to analyze and predict the credit risk of listed companies. In addition,the proposed method in this paper is compared with two classic algorithms,Node2vec and Deepwalk,and is used to predict the impact of COVID-19 on the credit risk of listed companies for the verification of the good effect of this method.