【作者】 丁东洋；

【导师】 刘乐平；

【作者基本信息】 天津财经大学 ， 统计学， 2009， 博士

【摘要】 信用风险遍及所有的金融交易,贷款违约风险是银行等金融机构进行信用风险分析时关注的焦点。贷款组合是银行在有限贷款总额的约束下,将款项贷给两个以上债务人,以分散信用风险的方法。由于行业特征和商业周期等宏观因素,以及企业间商业活动关联性等微观因素的影响,使得贷款组合中的违约依赖表现为周期相关性和风险蔓延性。违约依赖性越高,贷款组合的潜在风险损失越大。如何在贷款组合信用风险度量中充分准确的反映违约依赖性,是当前学术研究和实践应用中的重要问题之一。我国正在逐步实施新巴塞尔协议,将信用风险度量技术从单个债务人扩展到贷款组合的角度进行研究,有利于商业银行更加准确的计算协议中要求的风险资本。同时新协议虽然对许多信用风险度量模型进行了完善,但是这些模型并不适于宏观层面评估整体经济的信用风险,这也是银行监管部门评价整个银行系统稳定性所面临的主要难题。从微观和宏观视角分别研究贷款组合的违约概率与信用损失分布,不仅有助于银行有效分散风险,完善信用风险度量技术,而且对于监管机构评估金融环境稳定性,加强风险管理都具有重要的现实意义。贷款组合信用风险度量的显著特征是缺少实际违约数据。贝叶斯统计方法可以用来考察概率模型中与参数相关的不确定性,是一种科学有效使用专家意见等主观经验的研究技术。贝叶斯方法在贷款组合信用风险度量中的应用途径主要体现为两个方面,一是应用贝叶斯方法参数化模型,在风险度量时,贝叶斯方法可以作为一个技术工具估计风险模型中违约概率等重要变量;二是应用贝叶斯方法估计信用损失分布,正确描述贷款组合信用损失分布的动态变化特征,将损失分布分解为可观测变量,并诊断损失波动率。在贷款组合信用风险度量研究中,主要结论包括两点,一是基于贝叶斯方法构建的信用风险度量框架,结合MCMC模拟技术的应用,在一定程度上缓解了缺少实际违约数据问题,二是通过灵活运用贝叶斯模型中的潜在因素,能够正确反映贷款组合的违约依赖性,并可以对整体经济的信用损失分布给出动态描述。主要创新之处体现在以下三个方面:(1)拓展潜在因素的应用。在贝叶斯框架下,运用潜在因素描述贷款组合中个别债务人质量、行业特征、商业周期等影响因子,进而应用分层先验分布构建多级模型处理违约依赖性和债务人异质性等问题,不仅结果准确,而且统计推断简洁清晰。(2)从商业银行的微观视角构建贷款组合违约概率度量的贝叶斯模型。使之不仅涵盖宏观经济冲击,而且考虑债务人异质性问题,允许信用质量变化存在跨期自相关,从而更加确切的描述贷款信用等级变化过程,解释宏观系统风险对违约概率的影响,并且针对不同的数据限制,推导模型不同的特定形式。同时完善巴塞尔协议给出的样本外测试方法,不仅采用离差信息准则校验模型,强化模型预测能力,而且通过改进样本外模型比较方法,进一步说明结论的不确定性。(3)从监管机构的宏观视角构建贝叶斯信用风险损失分布评估框架。设计整体经济信用损失分布度量方法以评估金融环境稳定性,给出损失分布动态参数化方法,将损失分布分解为可观测变量,进一步解释违约蔓延性,并给出损失波动性的诊断方法。基于贝叶斯方法构建的信用风险度量模型除了计算量较大之外,某些预测结果仍有与实际违约概率存在偏差的情况,同时模型的稳健性需进一步检验,以上不足之处有待深入研究。更多还原

【Abstract】 Credit risk expand all over the financial business,default riks of loan is the focus of credit risk analysis in financial institution,especially in bank.Loan portfolio is the method that bank lend the money to two or more debtors to diversify credit risk under the constraint of total loans. As the effects of macroeconomic factors such as industry characteristics and the business cycle, as well as micro-factors such as related commercial activity between enterprises,default dependence of loan portfolio exhibit in cyclical correlation and default contagion.The higher the dependency of default,the the greater the potential risk of loan portfolio loss.How to fully accurate reflect default dependence in loan portfolio credit risk measurement,is the the focal point of current academic research and practical applications.China is gradually implementing the new Basel Accord,extending the study of risk measurement techniques from the perspective of single debtor to the loan portfolio,can get a more accurate calculation of the accord request risk capital.At the same time,although the new accord improve many credit risk measurement model,but these models are not suitable for the macro level to assess the credit risk of the whole economy,which is the main problems that banking supervision departments will be faced in the evaluation of the stability of the whole banking system.From the micro and macro perspective to study the default probability of loan portfolio and the distribution of credit losses,will not only help banks to diversify the risk and improve the credit risk measurement techniques,but also have important reality significance for regulatory agencies to assess the stability of the financial environment and reinforce risk management.Loan portfolio credit risk measurement is significantly characterized by lack of empirical default data.Bayesian statistical methods can be used to study the uncertainty associated with parameters in the probability model,is a research technique that can use subjective experience such as expert opinion scientifically and effectively.The applications of Bayeisan methods in loan portfolio credit risk measurement are mainly reflected in two ways.The first,Bayesian methods can be used to parameterize models,which are technical support tools to estimate the probability of default and other key variables in the process of risk management decision analysis.The second,Bayesian methods can be used to estimate the distribution of credit losses, the Bayesian method can effectively describe dynamic changes of the loan portfolio credit losses, and divide loss distribution into observation variable,and diagnosis the fluctuation rate of of loss.In the loan portfolio credit risk measurement study,including the two main conclusions,one is based on Bayesian methods to build the credit risk measurement framework,combined with the application of MCMC simulation techniques,can alleviate the problem of lack of empirical default data to a certain extent.And the other is through neatly use of of the latent factors in Bayesian model,can accurately reflect the default dependence in loan portfolio,and can give the dynamic describe of credit loss distribution of the whole economy.The major innovations are in the following areas:Firstly,expand the application of the latent factors,especially in the loan portfolio credit risk measurement.The impact of factors such as the quality of the individual debtor,industry characteristics and the business cycle can all be described through the latent factors.And then use hierarchical prior distribution to build multi-level model deal with default dependency and heterogeneity of debtor,thereby not only made the results more accurate to avoid underestimating the risks,but also have more concise definition of statistical inference.Secondly,construct Bayesian model from the micro perspective of commercial bank to measure default probability of loan portfolio,which not only cover the macroeconomic impact, given the heterogeneity of the debtor,and allow the existence of cross-phase-related of changes in credit quality,so that have a more precise description of credit rating changes in the loan process,can explain the risk of macro-system influence on the probability of default.And for different data constraints,the model is derived the specific forms.At the same time improve the sample testing methods in BaselⅡ,not only use Divance Information Criterion to validate model, and strengthen the capacity of models to predict,but also through improved out-of sample model comparison methods to further explain the uncertainty of conclusions.Thirdly,construct the framwork of Bayesian assessment of loss distribution from the macro perspective of regulatory institutions.Propose the approach of dynamic parameterizing of the loss distribution,and decompose the loss distribution into observed variables,further explain the contagion of default,and give the diagnostic methods of loss volatility.In addition to the mass calculation in the credit risk measurement model based on Bayesian methods,there still have some deviations between prediction and the actual probability of default, at the same time the robustness of the model need for further testing,which are to be studied in-depth.更多还原