基于余弦相似度的企业违约预测模型及实证  

A cosine similarity-based corporate default prediction model and empirical evidence

作　　者：沈隆 周颖[1] 赵轩铎 SHEN Long;ZHOU Ying;ZHAO Xuanduo(School of Economics and Management,Dalian University of Technology,Dalian 116024,China)

机构地区：[1]大连理工大学经济管理学院,大连116024

出　　处：《系统工程理论与实践》2022年第7期1826-1842,共17页Systems Engineering-Theory & Practice

基　　金：辽宁省社会科学规划基金(L21BGL011)。

摘　　要：企业违约预测,是通过挖掘指标数据与违约状态之间的函数关系,来预测企业未来的财务风险,其对银行贷款、股票投资、公司债券投资等具有重要的参考意义.本研究的贡献主要有三个方面:一是通过将企业的夹角余弦值而非欧氏距离作为违约状态相似性的度量,构建企业违约预测的Cos-k-means模型,避免了通过欧氏距离聚类时,忽略企业和聚类中心的角度关系以及对噪声样本和极端值敏感的弊端.二是通过构建兼顾模型违约鉴别力最大和指标组合冗余最小的目标函数,采用二进制粒子群算法(binary particle swarm optimization,BPSO)遴选指标组合,保证了构建的信用评价指标体系简洁、合理.三是通过采用企业t-d(d=1,2,3,4,5)期的指标数据和第t期的违约状态构建违约预测模型,实现了使用企业第t期的指标数据预测t+d期违约状态的目的.Enterprise default prediction is to predict the future financial risk of the enterprise by mining the functional relationship between indicator data and default status,which has important reference significance for bank loans,stock investment,and corporate bond investments.The contributions of this study are mainly in three aspects.The first is to measure the similarity of default states by using the angle cosine value of the enterprise rather than the Euclidean distance.The Cos-k-means model of enterprise default prediction is constructed.Avoiding the disadvantages of ignoring the angle relationship between enterprise and cluster center and being sensitive to noise samples and extreme values when clustering by Euclidean distance.Second,by constructing an objective function that maximizes the model’s default discrimination ability and minimizes the redundancy of the indicator combination,the binary particle swarm optimization(BPSO)is used to select the indicator combination to ensure that the constructed credit evaluation indicator system is simple and reasonable.Third,the default prediction model is constructed by using the indicator data of the enterprise in period t-d(d=1,2,3,4,5)and the default status in period t.The purpose of using the indicator data of the enterprise in period t to predict the default status in period t+d is achieved.

关 键 词：违约预测 指标选择 Cos-k-means 余弦相似度 

分 类 号：F830.56[经济管理—金融学]