摘要
不良贷款的回收率与其处置方式有密切的关系,因此寻找最优处置方式是实际工作中很自然的诉求.但是因为不良贷款的特征与其最优处置方式之间的关系过于复杂,以往的文献中对处置方式只有定性的研究.本文不去关注不良贷款的特征与最优处置方式决定关系的具体形式,而是从处置方式的样本个数入手对处置方式进行定量研究,根据历史处置时的"投票"数越多则表明处置方式越优秀的思想,建立了处置方式的判别模型.文章首先通过列联表检验发现不良贷款的处置方式与贷款本金余额、本金占比、贷款银行、贷款担保方式、贷款企业的工商登记状态和经营现状等因素相关,而与债务企业是否上市公司和注册资本不相关.基于这些影响因素,本文首次将部分线性决策树(PLTR)的方法应用于建立不良贷款处置方式的判别模型,同时表达了影响因素与处置方式之间的线性关系和非线性关系,获得了相对较好的判别效果.由此,本文也发掘出不良贷款的历史处置模式,因而可以从提高期望回收率的角度提出合理建议.
The recovery rates of Non-Performing Loans （NPLs） are close related to their disposal methods, so to find the most effective disposal method is naturally demanded in practical work. But only qualitative results exist in literature because of the complicate relationship between characters of NPLs and their corresponding best disposal methods. The complicated relationship was avoided here, and we conducted our qualitative research from the viewpoint of sample size of specific disposal method. According to the basic idea that the best disposal method should have the largest number of ＂votes＂ in the disposal history, this paper obtained a diseriminant model of disposal methods. The determinants of disposal methods were first determined. According to contingency table testing results, the disposal methods of NPLs are related to their loan principal, principal ratio, loan bank, loan guarantee way, commercial registration status of the borrowing company, and business status of the borrowing company, but not their listed or unlisted status, or registered capital. After this, we applied the Partial Linear Tree-based Regression （PLTR） method in NPLs＇ modeling for the first time, and included both linear and nonlinear relationship in the diseriminant model, and got better model accuracy compared to Decision Tree （DT） method. Finally, some historical disposal patterns were discovered and some useful advices were given from the viewpoint of expected recovery rate accordingly.
出处
《数理统计与管理》
CSSCI
北大核心
2013年第6期963-973,共11页
Journal of Applied Statistics and Management
基金
国家973计划(No.2007CB814902)
国家自然科学基金(No.70933003)
国家自然科学基金(No.71203247)
教育部人文社科青年基金(11YJC790015)
中央财经大学青年创新团队的资助
关键词
不良贷款
处置方式
列联表
决策树
部分线性决策树模型
non-performing loan
disposal methods
contingency table
decision tree
partial linear tree-based regression model
