【作者】 季峰；

【导师】 方兆本；

【作者基本信息】 中国科学技术大学 ， 管理科学与工程， 2009， 博士

【摘要】 随着金融全球化和自由化进程的加快,金融市场的风险呈现出高关联、高频发和高损失的特点。拉美银行危机、亚洲金融风暴、美国次贷危机等一系列事件反映出金融市场的脆弱性,也引发了业界和学术界对商业银行信用风险管理的深入思考。巴塞尔银行监管委员会作为国际清算银行的正式机构,其所制定的巴塞尔协议是全球公认的商业银行监管标准。该委员会于2006年颁布了新巴塞尔协议,新协议将信用风险、市场风险、流动性风险和操作风险纳入风险计量范畴,构建了资本监管的“三大支柱”,即资本充足率、监督检查和市场约束。新协议对商业银行信贷风险管理提出了更为严格的条件,它将违约概率的测度和评估列为内部评级法的核心内容,要求各成员国银行使用内部评级法来确定风险权重和计提风险资本。同时,理论界对违约概率模型的研究也做了大量的研究,主要集中于影响违约率的关键因素的选择,并基于分类算法,以历史数据为驱动、以数学模型和统计方法为基础来建立违约概率模型。我国现代商业银行体制刚刚建立,自身的风险管理水平有限以及历史数据积累不够,尚不能满足商业银行对各种形式贷款安全性的准确测量。但是随着我国商业银行国际化程度的加深,中国银监会以新巴塞尔协议为基准,要求我国各大商业银行提升消费信贷风险管理水平,在加强违约损失率历史数据库建立的同时,着重研究贷款的违约概率。本文在现有消费信贷违约概率度量研究的相关文献进行系统综述的基础上,构建了SenV-RBF-SA和时变相依的量化模型来度量消费信贷的违约风险,采用Copula方法建立了商业银行消费信贷整体风险的度量模型,并利用商业银行消费信贷的实际数据进行实证。本文的主要工作及成果如下:首先,考虑到SenV-RBF神经网络对数据无分布要求且在处理非线性问题是表征出的特性,以及半参数Cox比例危险模型可进行违约概率的动态预测,因而构建了基于SenV-RBF-SA融合的违约概率动态模型,对借款人未来某个时点的违约概率风险进行度量;并通过商业银行消费信贷的实际数据,实证研究发现,本文构建的混合模型在判别精度和稳健性方面与传统模型相比有一定的竞争性。其次,考虑到GDP、利率、CPI、上证综合指数等宏观经济的波动给借款人带来的系统性风险的影响,本文在上述混合模型基础上,采用时变相依Cox比例危险方法构建了一类消费信贷违约概率度量模型,客观度量了宏观经济因素对借款人平均违约水平的影响,它克服了以Logistic回归模型为代表的传统模型在度量消费信贷违约概率时仅考虑个体非系统性风险的局限。最后,通过实证分析,本文提出的时变模型相比于传统违约概率模型有较高的准确率和稳健性,这是对第三章研究的一个拓展。最后,本文通过研究提前还款与实质性违约之间的相依关系,基于Copula方法建立了二者间的整体违约风险的度量模型。我们依据非参数核密度方法估计出两组信贷产品的生存时间的边际分布;然后对每一个生成的Copula相依结构采用QQ图和Kolmogorov-Smirnov检验挑选出最优的Copula;再利用之前已获得的违约边际分布,基于Copula相依测度思想,从而构建构建了一种新的相依违约度量模型;最后给出基于Copula相依性违约测度的Kendall的秩相关系数,并进行了实证研究。更多还原

【Abstract】 In the accelerated development of globalization and liberalization in financial market,the risk of finance has appeared to be highly inter-related,frequently occurred and caused heavily losses.A series of events such as bank crisis in Latin America,the Asian financial crisis,the subprime mortgage crisis in the US reflected fragility of finance market,and it also arose the further thinking of the commercial banks’ credit risk management in banking section and academic world.Basel capital accord,which is enacted by Basel Committee on Banking Supervision(as the formal institution for international settlement) is the standards of regulation of commercial banks accepted globally.The Committee issued the new Basel accord in 2006.It included the credit risk,market risk,liquidity risk and operational risk into the research field of the measuring risk;hence he argued the three pillars of capital regulation,namely capital adequacy ratio,supervision and inspection,market discipline.The new accord has set up more tightened requirements in the credit risk management of commercial bank.The new approach ranked the measurement and evaluation of the default probability as the core of Internal Rating Based(IRB) approach;it proposed that IRB should be adopted in practice of risk weight as well as the counting and drawing of capital at risk.Meanwhile,theoretical circles have done a lot of works on the research of Probability of Default Model.It mainly focused on the option of influence on the key factor of default rate and established Probability of Default Model to historical data as driven,on the basis of mathematical models and statistical methods based on classification algorithm.China’s modern commercial banking system had just been established,the limited level of their risk management,as well as the inadequate accumulation of historical data,all of those were not possible to meet all forms of commercial banks loans accurate security measurement.However,as the China’s commercial banks deepening degree of internationalization,based on the new Basel Accord,the China Banking Regulatory Commission required the major commercial banks to enhance their level of consumer credit risk management,and focused on studying the loan default probability,at the same time of enhancing building default loss rate history databases.Based on the systematic reviews of relevant literature on researching consumer credit loan default probability measurement,this paper constructed SenV-RBF-SA and time-varying covariates quantitative model to measure consumer credit loan default risk,and established the commercial bank Overall consumer credit risk measurement model using Copula method,and using actual commercial bank consumer credit loan data to research.The main work and the results of this paper are as follows:First of all,considering the SenV-RBF neural network had no requirements for data distribution,and the characterization in dealing with non-linear problems,as well as the semi-parametric Cox proportional hazard model can dynamically predict default probability,this paper constructed SenV-RBF-SA dynamic models of default probability to measure borrower’s default probability in a future point.At the same time,the empirical research on actual data of commercial banks consumer credit, found that this hybrid model had a certain degree of competitive in the identification accuracy and robustness compared with traditional model.Secondly,taking into account the systemic risks impact on borrowers,caused by GDP,interest rates,CPI,the Shanghai Composite Index and Etc.macroeconomic fluctuations.Base on the above hybrid model,this paper used the time-varying Cox proportional hazard model to build a class of consumer credit default probability measure model,and objectively measure the impact of the macro-economic factors on the borrowers’ average default level.The model overcomed the limitations of traditional model which took the Logistic regression model for representative,which only considered individual non-systemic risk when measured consumer credit default probability.Finally,through empirical analysis,the time-varying model had higher accuracy and robustness compared with the traditional default probability model,this is a development study of the third chapter. Finally,this chapter mainly investigated the dependent relationship of prepayment and real default,and constructed a model which can measure the whole default risk between prepayment and real default based on the Copula method.We gained survival time maginal distribution of the two groups of credit products through non-parameter kernel density estimation method.Then optimal Copula would be chosen by QQ diagram and Kolmogorov-Smirnov test on every dependent structure of generated Copula.And then a new PD model of dependent risk was established based on measurement ideas of Copula and default marginal distribution.At last gave the kendallτbased on Copula dependent default measure,and conducted empirical research.更多还原