基于偏最小二乘法改进的CPV模型的我国商业银行信用风险压力测试研究

发布时间：2018-02-02 20:53

本文关键词： 偏最小二乘法 CPV模型蒙特卡罗模拟法压力测试违约概率　出处：《复旦大学》2014年硕士论文　论文类型：学位论文

【摘要】：现如今银行业所需要处理的各类风险中,信用风险已成为最主要的一种,如何对信用风险进行监控和对其可能引发的危机进行预警是对银行本身和监管部门而言非常重要的课题.如果需要对信用风险进行识别和预防,最急需的是构建一个科学的度量系统.传统的信用风险评估方法是信用评级为主的定性模型,直至20世纪90年代后在计算机编程技术和现代金融工程方法的辅助之下,定量的信用风险度量模型才得以产生并迅速发展.本文首先对比了其中几个常用的信用风险度量模型,即：KMV模型、CreditRisk+模型、CreditMetricsTM模型及其改进版—贷款组合观点(CreditPortfo-lio View,简称CPV)模型相互之间的优劣.虽然方法论上各有千秋,但这四类模型的主要作用都是使用可观测数据对资产组合或整个金融体系的信贷资产面临的违约风险进行拟合,以期可以预知其未来的风险水平.但事实上,利用过去或现在的数据所计算出的违约风险都是基于过去或现在的市场现状情况下的,但现实世界中的金融市场瞬息万变,例如1997年亚洲金融风暴、2007年次贷危机这类影响巨大的金融危机很难通过历史数据预知,但一旦发生,就会使得以银行首当其冲的金融体系受到巨大冲击,因此对这种异常情况下自身可能受到的损失银行业及监管机构是必须考虑提前预防的,在这种情况下压力测试这样一个通过模拟经济受到冲击(主要是负面冲击)时的情景来计算风险如何变化的工具应运而生.本文通过比较认为CPV模型这样一个与宏观因素直接挂钩的模型对于用来进行压力测试比较实用,但是前人在建立宏观因素与违约概率的关系时一般所使用的似无关回归法在自变量之间存在线性相关关系时会使参数出现错误的估计,因此需要在选择宏观因素作为自变量时需要首先手工筛选相关性比较小的因素.就此本文提出使用偏最小二乘法来克服这个缺点,即用偏最小二乘法(PLS)代替原始模型中使用的似无关回归法(SUR),在使用蒙特卡罗法进行模拟的配合下建立压力测试模型,以期对信用风险在宏观经济在未来某一时刻受到可能的打击时的变化程度进行评估,但在尝试进行实证时遇到因变量—违约概率数据难以寻找的情况,虽然国内有学者尝试直接使用不良贷款率进行替换并进行实证,但本文通过分析认为这样的做法并不合理,而是借助不良贷款率生成了违约概率的样板数据进行了详尽的实证及压力测试的模拟,并对进一步提高模型精度提出了展望.
[Abstract]:Nowadays, the credit risk has become the most important one among all kinds of risks that the banking industry needs to deal with. How to monitor the credit risk and warn the possible crisis is a very important issue for the bank itself and the supervision department. If it is necessary to identify and prevent the credit risk. The most urgent thing is to build a scientific measurement system. The traditional credit risk assessment method is a qualitative model based on credit rating. Until 1990s, aided by computer programming techniques and modern financial engineering methods. Quantitative credit risk measurement model has been produced and developed rapidly. Firstly, this paper compares several commonly used credit risk measurement models, namely: KMV model and CreditRisk model. The CreditMetricsTM model and its improved version of the loan portfolio perspective are CreditPortfo-lio View. The pros and cons of CPV models are different in methodology. But the main role of these four models is to use observable data to fit the default risk of the portfolio or the credit assets of the whole financial system in order to predict the level of future risk. Default risks calculated from past or present data are based on past or present market conditions, but financial markets in the real world are rapidly changing, such as the Asian financial turmoil in 1997. In 2007, a financial crisis such as the subprime mortgage crisis is difficult to predict through historical data, but once it occurs, it will make the financial system which bears the brunt of the banks to suffer a huge impact. So banks and regulators must consider prevention ahead of time for possible losses to themselves in such exceptional circumstances. In this case, a stress test such as a simulated economic shock (mainly a negative shock). This paper thinks that CPV model, which is directly linked to macro factors, is more practical for stress testing. However, in establishing the relationship between macro factors and default probability, the similar independent regression method used in general makes the parameters misestimate when there is a linear correlation between independent variables. Therefore, when selecting macro factors as independent variables, we need to first manually screen the factors with small correlation. In this paper, we propose the use of partial least square method to overcome this shortcoming. In other words, the partial least square method (PLS) is used to replace the similar independent regression method used in the original model, and the pressure test model is established with the cooperation of Monte Carlo simulation. In order to assess the degree of credit risk change in the future when the macroeconomic is likely to be hit, but in the attempt to carry out empirical analysis of dependent variables-default probability data difficult to find the situation. Although some domestic scholars try to directly use the non-performing loan rate to replace and carry on the demonstration, but this article through the analysis thinks that this kind of practice is not reasonable. On the other hand, the model data of default probability is generated by non-performing loan ratio to simulate the model and stress test in detail, and the prospect of further improving the precision of the model is put forward.
【学位授予单位】：复旦大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F832.33;F224

【相似文献】