商业银行小微企业违约风险管控及违约概率估计模型研究

发布时间：2018-07-29 07:13

【摘要】：作为国民经济中最具活力和创新力的组成部分,小微企业在拉动经济增长、保持经济稳定并扩大就业中扮演了重要的角色。企业发展离不开金融的支持,然而,与小微企业重要地位相矛盾的是作为其最重要外部融资来源的商业银行信贷却未能有效地满足其融资需求。尽管在大型客户金融服务市场竞争激烈,利率市场化和金融脱媒等多重压力下,小微信贷成为了银行业未来的“蓝海”业务,但现实中,由于小微信贷违约风险大,单笔成本高,造成其实际风险不可控,盈利能力弱,使得这片“蓝海业务”既难以简单地与银行盈利划上等号,又难以实现监管达标。因此,各家银行在真正投放小微信贷时往往慎之又慎。小微信贷之所以面临如今的困境,其根本原因在于银行缺乏相应的风险管理能力。银行的本质是经营风险的企业,通过对风险进行有效地识别、计量、缓释、对冲和定价赚取风险调整后的利润。在传统的信贷业务中,银行基于对贷款的违约风险进行管理和估计并确定相应的风险溢价,从而赚取存贷利差。然而,小微企业由于生产经营规模小、财务报表不健全,缺乏有效的抵质押物,导致与银行之间的信息不对称严重,银行无法有效识别其真实的违约概率,缓释违约风险造成的损失并进行合理的定价。结果造成市场实践中的两个极端:或是通过非价格手段抑制金融需求(如信贷配给);或是放任小微信贷风险失控,最终难以实现可持续发展。所以,小微信贷难破局的关键在于银行如何建立适用于小微企业的违约风险管理方法和技术,实现对小微信贷违约风险的有效识别和计量,进而支撑合理风险定价的实现。其中,对违约概率的准确估计是提升违约风险管控能力的核心。尽管许多学者对中小微信贷违约风险的成因,管控策略,及道德风险(Moral Hazard),逆向选择(Adverse Selection)和信贷配给(Credit Rationing)行为等相关议题进行了广泛研究。但目前还少有研究专注于对小微企业违约风险的特征及相应的违约管控机理进行系统分析与总结,更缺乏基于这类客户风险管控机理的违约概率建模研究。因此,本文基于相关理论、文献的梳理与借鉴及对小微信贷运营与管理实践经验的提炼、总结,研究了适用于小微企业的违约风险管理机制,对小微企业的违约管控机理进行了深入刻画与量化,分别构建了基于理想数据条件下,适用于小微企业的不完全信息违约估计模型、统计学与机器学习类违约估计模型和基于现实数据缺失条件下的违约估计模型。本文的研究成果将有助于加深对小微企业违约风险特征、管控机理的了解与认识;改变传统的依赖于资产和抵质押物这类不适合于小微企业的风险管控模式,提升对这类客户违约概率估计和预测的准确性、有效性和可行性;帮助建立并完善适用于小微企业的内部评级体系,并在客户准入,授信审批,风险监测与经济资本配置等方面发挥重要作用。全文共分为七章,包含三部分主要内容:第一部分,在文献综述、相关理论回顾和成功实践梳理的基础上对小微企业的违约风险特征和违约管控机制进行研究、提炼和总结。首先分析了小微企业的四大风险特征:缺乏有效的抵质押物;信息不透明程度较为严重;单笔规模小、资产池规模大以及对外部环境变化更为敏感。然后结合业界实践和前人研究成果提出了适用于小微企业的三大违约管控机理:基于现金流的违约触发机制;基于关系信贷减少信息不对称和分池管控,定量与定性相结合。并据此提炼了适用于这类客户的违约风险模型所需具备的特征。本部分的研究成果,为后文模型的建立提供了理论基础和建模思路。第二,基于小微企业违约风险管控机理,分别构建了理想数据条件下,适用于小微企业的不完全信息类模型与统计学和机器学习类模型。应用不完全信息模型的框架,通过对基于现金流的违约触发机制中包含的核心要素:违约边界和真实现金流分布进行提炼、抽象与刻画,构建了适用于对信息不对称程度不断变化条件下小微企业违约概率进行有效估计的理论模型,并通过逐步放松:银行可以完全观测到客户的初始信息和客户新发生的借贷金额对违约概率估计没有影响这两个假设,构建了具有实际应用价值,可以有效刻画小微企业违约风险的理论模型。基于真实的客户违约数据,构建了适用于小微企业的统计学和机器学习类模型。首先,建立了系统的小微企业违约风险评价指标体系。其次,利用真实数据,对适用于小微企业的违约风险评价指标进行拟合检验,获得了以客户现金流类指标和关系信贷类指标为核心的最具有预测效力的违约预测指标,验证了前文的理论分析结论。第三,通过对不同模型的预测效力进行实证分析,发现整合Logistic回归模型和支持向量机方法的混合违约概率预测模型是最适用于本文小微企业数据样本的统计学和机器学习类模型,这一模型不仅具有最高的预测精度而且综合误差成本最低,预测稳定性最好。第三,针对现实中存在的小微企业信贷违约数据缺失,模型估计有效性难以保证的问题,通过运用贝叶斯估计,整合专家先验信息和数据信息,获得更为有效的后验估计结果。结论表明,后验估计结果既可以弥补由于历史数据信息不足带来的传统估计结果不可信问题,又可以平滑极端历史数据对真实违约概率估计的冲击,从而有效提升违约概率估计的准确性和有效性,鲁棒性检验的结果也证明了上述结论。在此基础上,进一步加入基于现金流触发机制的单因素违约相关性模型,t-copula违约相关性模型和多期条件以提升估计有效性,并设计了基于Proper Scoring Rules的评分规则,以对专家经验的有效性进行评价,对专家的权重进行动态设计。
[Abstract]:As the most dynamic and innovative component of the national economy, small and micro enterprises have played an important role in stimulating economic growth, maintaining economic stability and expanding employment. The development of enterprises can not be separated from financial support. However, it is in contradiction with the important position of small and micro enterprises to be the commercial bank credit for its most important source of external financing. However, small credit has become the "blue sea" business of the banking industry in the future, despite the fierce competition in the large customer's financial services market, interest rate marketization and financial disintermediation. But in reality, because of the large risk of default and high cost, the actual risk is uncontrollable and the profit is uncontrollable. Because of its weak ability, this "blue sea business" is difficult to be equated with the bank's profit simply, but it is difficult to realize the regulation. Therefore, each bank is often cautious and cautious when it is really put in small and micro credit. The reason why the small micro credit is facing today's dilemma is that the bank lacks the corresponding risk management ability. Quality is an enterprise that manages risk by effectively identifying, measuring, sustained-release, hedging, and pricing a risk adjusted profit. In the traditional credit business, banks manage and estimate the risk of default on loans and determine the corresponding risk premium, thus earning the loan spreads. However, small micro enterprises are produced due to production. Small business scale, incomplete financial statements and lack of effective impawns cause serious information asymmetry between banks and banks. Banks can not effectively identify their true default probability, slow down the losses caused by default risks and make reasonable pricing. The results are two extremes in market practice: or by non price means. Therefore, the key to the hard break of small micro credit is how to establish a method and technology for the management of default risk for small micro enterprises, to realize the effective identification and measurement of the risk of small and micro credit, and to support the reasonable risk. The accurate estimation of the probability of default is the core of improving the ability to control the risk of default, although many scholars have carried out a wide range of related issues such as the causes of the default risk, the control strategy, the Moral Hazard, the reverse selection (Adverse Selection) and the Credit Rationing behavior. But at present, few studies have focused on the systematic analysis and summary of the characteristics of small and micro enterprises' default risk and the corresponding mechanism of default management, and lack the research on default probability modeling based on this kind of customer risk management mechanism. Therefore, this paper is based on the relevant theories, and the literature is combed and used for reference and the operation and management of small and micro credit. The mechanism of default risk management for small and micro enterprises is studied and the mechanism of default risk management is studied. The mechanism of default management for small and micro enterprises is deeply depicted and quantified. The incomplete information default estimation model for small micro enterprises based on ideal data is constructed, and the model of statistics and machine learning default estimation model are established. The research results of this paper will help to deepen the understanding and understanding of the characteristics of default risk and control mechanism of small and micro enterprises, and change the traditional risk management model which is not suitable for the small and medium-sized enterprises, such as assets and collateral, and improve the estimation of default probability of these customers. And the accuracy, effectiveness and feasibility of the prediction, help to establish and improve the internal rating system for small micro enterprises, and play an important role in customer access, credit examination and approval, risk monitoring and economic capital allocation. The full text is divided into seven chapters, including three parts of the main content: the first part, in the literature review, the related theory review On the basis of the successful practice, this paper studies the characteristics of the risk of breach of contract and the mechanism of control and control of the small and micro enterprises. Firstly, it analyzes the four characteristics of the risk of small and micro enterprises: the lack of effective collateral; the information opaque degree is more serious; the single scale is small, the scale of the asset pool is large and the external environment changes more. Then, combining with the industry practice and the previous research results, we put forward the three major mechanism of default management for small micro enterprises: the trigger mechanism of default based on cash flow; based on the relationship credit reduction information asymmetry and the pool management, the quantitative and qualitative combination. The research results of this part provide theoretical basis and modeling ideas for the establishment of the later model. Second, based on the mechanism of small and micro enterprises' default risk management and control, the incomplete information model and statistics and machine learning model for small micro enterprises are constructed under the ideal data conditions. By refining, abstracting and depicting the core elements of default triggering mechanism based on cash flow: default boundary and real cash flow distribution, the framework builds a theoretical model for effective estimation of default probability for small and micro enterprises under the condition of continuous change of information asymmetry. The total observation of the customer's initial information and the customer's new loan amount has no effect on the probability of default, and constructs a theoretical model that has practical application value and can effectively depict the default risk of small and micro enterprises. Based on the real customer default data, the statistics and machinery for small micro enterprises are constructed. First, the system is established to evaluate the default risk evaluation index system of small and micro enterprises. Secondly, using real data, the evaluation index of default risk applicable to small micro enterprises is fitted and tested, and the most predictive effective default prediction index, which is the core of customer cash flow index and relation credit index, is obtained. Third. Third, through the empirical analysis of the predictive effectiveness of different models, it is found that the mixed default probability prediction model of the integrated Logistic regression model and the support vector machine method is the most suitable statistical and machine learning model which is most suitable for the data samples of small and micro enterprises. This model is not only the highest. The prediction accuracy and the comprehensive error cost are the lowest and the prediction stability is the best. Third, in view of the lack of credit default data in the small micro enterprises and the problem that the validity of the model estimation is difficult to guarantee, the expert prior information and data information are integrated by using Bayesian estimation, and the results of the more effective posterior estimation are obtained. The conclusion shows that The results of the posterior estimation can not only compensate for the unbelievable problem of the traditional estimation results due to the insufficient historical data information, but also smooth the impact of the extreme historical data on the real default probability estimation, thus effectively improving the accuracy and effectiveness of the default probability estimation. The results of the robust test also prove the above conclusion. On the other hand, we further add the single factor default correlation model based on the cash flow trigger mechanism, the t-copula default correlation model and the multi term conditions to improve the effectiveness of the estimation, and design the scoring rules based on the Proper Scoring Rules to evaluate the effectiveness of the expert experience, and to dynamically design the weight of the experts.
【学位授予单位】：南京大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：F832.4

【参考文献】