【作者】 梁凌；

【导师】 彭建刚；

【作者基本信息】 湖南大学 ， 金融学， 2009， 博士

【摘要】 管理信用风险是商业银行在21世纪的一个重要挑战,银行信用风险管理的重要工作是对信用风险的度量和对风险的定价。本文研究国际现代信用风险度量模型在我国商业银行信用风险管理和贷款风险定价的理论和实践问题,旨在建立一套符合中国实际的、可操作的商业银行贷款资产信用风险测度方法和风险定价方法,并解决相应的理论问题。通过对信用风险测度模型相关概念的梳理和界定,从观测期的选择、信用损失的衡量方法、信用损失概率密度函数、条件模型与无条件模型、组合信用风险、违约的相关性等方面,本文对现代信用风险模型进行了比较分析。在风险测度的指标选取上,作者研究了VaR与CVaR的特性,认为信用风险的传统度量指标VaR在很多情况下不能体现Markowitz资产组合理论的风险分散化效应,在数学上不满足次可加性,而CVaR作为风险度量领域的最新进展,正是具备这些特性的风险度量指标商业银行贷款信用风险的测度和风险定价包含交易级与产品级两个层面。从交易层面来看,依据五级分类次级贷款的核心定义和划入原则以及新资本协议关于债务人违约的定义,作者认为,银行可以通过贷款迁徙率矩阵来计算各级贷款的违约概率,据此采用贷款的五级分类形态来标示违约概率,从而解决现阶段信用等级违约概率数据质量差的问题。此外,在我国信贷文化、司法环境的基础上,本文还构建了一种抵押品池的综合违约损失率计算模型,可以符合实际地解决违约损失率的计算问题。依此所构建的违约损失率模型和五级分类的违约概率体系,在IRB模型的框架下对某商业银行信用风险测度的实证表明,运用本文提出的修正方法计算的资本要求,在一般情况下要低于银监会“商业银行资本充足率管理办法”的资本要求。在核算贷款交易的经济资本成本时,因为实施新资本协议内部评级法前,一般准备金可以包含在二级资本中,贷款风险所占用的股权包括预期损失和未预期损失两个部分,通过对经济资本度量模型的研究,作者认为采用新资本协议CP3的版本确定贷款的资本消耗更符合股东的心理。在确定资金成本时,可以采用资金转移定价(FTP)的方式,将贷款业务难以分析和管理的利率风险分离出来,通过转移价格的方式将其转移到银行内部其他的专业部门进行管理。通过对贷款交易风险定价机制的理论探讨和实证研究,作者发现,由于不同的银行具有不同的运营成本与资金成本,覆盖了资本成本后的贷款风险定价在信贷市场上会表现出一种“翘板效应”,即经营成本低的银行在对信用等级高的债务人放贷时有价格优势;相反,经营成本高的银行在对信用等级低的债务人放贷时有价格优势。在贷款交易风险定价的理论基础上,作者还提出了银行授信边界的概念,在既定的贷款成本下,贷款利率、信用等级、违约损失率、贷款期限等授信条件相互制约,共同构成银行的授信边界。通过实证研究,利用利率等高线在RAROC三维授信曲面上标示出了授信边界的形状。从产品级的层面来看,本文研究了商业银行零售业务贷款的聚合信用风险,系统性地建立了基于特征函数的聚合信用风险模型,解决了传统的CreditRisk模型在违约率可变的条件下聚合信用风险无解的缺陷。在运用特征函数聚合信用风险模型的基础上,使用VaR和CVaR两种风险测度指标,作者对商业银行零售贷款的一种产品进行了风险定价的实证,实证说明,如果需经济资本保持不变,CVaR使用的置信度τ要小于VaR使用的置信度α,两者的关系τ=2α-1是偏于保守的估计。更多还原

【Abstract】 Managing credit risk is an important challenge for commercial banks in 21st century, and the essential work of bank credit risk management is to measure credit risk and to price risk. The paper studies the theoretical and practical issues on applying international modern credit risk measure models in China’s commercial banks for credit risk management and loan risk pricing, with the aim to provide an excisable loan asset credit risk measurement and credit pricing methods those fix Chinese commercial banks and therefore solve the related theoretical problems.Through the combing and definition of the credit risk measure model related concepts, from the selection of observation period, credit loss measure method, credit loss probability density function, conditional model and unconditional model, portfolio credit risk, default relativity and so on, the paper does comparative analysis on modern credit risk models. In terms of the selection of credit measure index, the paper studies the characteristics of VaR and CvaR, and points out that the traditional measurement for credit risk, i.e. VaR can not reflect the Markowitz portfolio risk diversification effect in a lot of situations and do not satisfy the subadditivity, while CaR as the new development of risk measurement field is the right risk measurement processing the above characteristics.The commercial banks’loan credit risk measurement and risk pricing include two levels, i.e. transaction level and product level. From the perspective of transaction and according to the core definition and principles of five-grade loan and new Basel Accord definition of default, the paper suggests that banks can use loan transition matrix to calculate the default probability of each grade, accordingly, the loan’s five grades can reflect the default probability, therefore the problem of unsatisfied data of default probability at current stage can be solved. Besides, based on China’s credit culture and judicature environment, the paper builds a comprehensive default loss rate model on mortgage pool that practically solves the default loss rate calculation problem. The default loss rate model and five-grade default probability system built accordingly and the empirical analysis on a commercial bank’s credit risk measurement under IRB model framework indicate that the capital requirement calculated with the modified method proposed by this paper is less than that according to Measures for the Management of Capital Adequacy Ratios of Commercial Banks promulgated and implemented by China Banking Regulatory Commission.As to calculating the economic capital cost of loan transaction, before the implement of new Basel Accord internal rating-based approach, general reserve is included in Tier 2 capital, credit risk contained equity consists of expected loss and unexpected loss. Through the study of economic capital measure model, this paper believes that adopting new capital accord CP3 version to calculate loan’s capital consume is more proper for shareholders’sake. Regarding to calculate capital cost, fund transfer pricing method can be used to separate interest risk that is difficult to analyze and manage from loan business and transfer it to other specific department to manage.With theoretical and empirical study on loan transaction risk pricing, the paper finds out that due to different operation cost and funding cost, the loan risk pricing covered capital cost in credit market shows a“seesaw effect”, that is to say, those banks with low operation cost have the price advantage regarding to debtors with high credit rating. On the contrary, those banks with high operation cost have the advantage as to debtors with low credit rating. On the basis of loan transaction risk pricing theory, the paper proposes the concept of bank credit frontier, with the mutual constrains of specified loan cost, loan interest, credit rating, default loss rate, loan period and other conditions, bank credit frontier is built. With empirical study, it describes the frontier shape on three-dimension surface with interest contour line.Regarding to the product level, the paper studies the aggregation credit risk of commercial banks’retail loan and systematically build eigenfunction-based aggregation credit risk model, solving the drawback that traditional CreditRisk+ model under default probability changing condition can not provide analytic expression. Based on adopting eigenfunction-based aggregation credit risk model, with VaR and CvaR, the paper does risk pricing empirical study on one product of commercial bank’s retail loans and the test shows that if economic capital keeps unchanged, the confidence levelτof CVaR is less than the confidence levelαof VaR, the relationship ofτ=2α-1 is the conservative estimate of their relationship.更多还原