【作者】 张传新；

【导师】 王光伟；

【作者基本信息】 苏州大学 ， 金融学， 2012， 博士

【摘要】 信用风险是商业银行面临的最主要的一种风险。准确地度量信用风险既是商业银行经营管理的内在要求，也是应对《新巴塞尔资本协议》的现实需要。从目前我国的实际情况来看，信用风险度量仍是商业银行信用风险管理的薄弱环节，制约了内部评级法的实施和商业银行的健康发展。为了加强我国商业银行信用风险管理，论文以商业银行信用风险为主要研究对象，以构建信用风险度量模型为核心，在学习和借鉴国内外先进信用风险度量技术的基础上，利用国内样本数据构建了不同的信用风险度量模型，以期为我国商业银行的信用风险管理提供技术支持。论文从商业银行信用风险的特点和成因出发，根据国际和国内最新的信用风险监管要求，介绍了信用风险的内部评级和外部评级，并结合我国实际提出二者协同发展的建议。在比较和分析主要的信用风险度量方法及模型的基础上，以“Z-score”模型为例，对我国直接应用国外的成熟模型的效果进行了实证检验，证实了必须依据国情才能建立适用的模型。经过选择，确定基于会计数据和市场价值的多元判别分析和Logistic回归分析作为构建我国商业银行信用风险度量模型的主要方法。为了建立模型，文章详细介绍了商业银行信用风险度量的财务因素，并分析了信用风险度量要素的选择和降维处理方法，确立了采用逐步选择法筛选变量和主成分分析进行数据降维的思路。然后根据对财务困境的界定和国内的研究习惯，从我国证券市场选取了78家财务正常公司样本和78家财务困境公司样本以及七大类共33个财务比率指标，运用2010年的数据和SPSS软件，构建了一个六变量的逐步判别分析模型和一个四变量的Logit模型。考虑到财务数据的高维性和高相关性特点，在利用主成分分析从18个具有组间显著性差异的指标中提取5个主成分的基础上，构建了一个主成分分析下的判别分析模型和一个主成分分析下的Logit模型。经检验，这四个模型都是有效的模型，且模型都具有一定的超前预测能力。通过比较发现，Logit模型比判别分析模型的准确性高，判别分析模型的稳定性要比Logit模型好。还发现主成分分析能够提高模型的超前预测能力和稳定性，且对Logit模型的作用更为显著。此外，还发现四个模型都具有跟随宏观经济的波动而同向变化的“顺周期性”。在经济上行时，模型均表现宽松，在经济下行期间模型则表现严格。建议通过适当调整临界点改变两类错误率的比例，从而减弱模型的“顺周期性”。研究表明，多元判别分析和Logistic回归分析两种方法仍是有效的信用风险度量方法。主成分分析是有效的数据降维方法。文中所建模型可以用于度量我国商业银行信用风险。更多还原

【Abstract】 Credit risk is the main risk for commercial banks. Accurate measurement of the creditrisk is not only the inherent requirement of the operation of commercial banks, but also thereality needs of dealing with the New Basel Capital Accord. From the current reality of ourcountry, the credit risk measurement is the weak link of commercial bank credit riskmanagement, which restricts the implementation of the IRB and the healthy developmentof commercial banks. In order to strengthen the credit risk management of commercialBanks in China, this paper takes commercial bank credit risk as the main object of study,takes construct the credit risk evaluation model as the core, and on the basis of learningand learn from the domestic and foreign advanced credit risk measurement techniques,using domestic sample data to build different credit risk measurement model, so as toprovide technical support for China’s commercial banks credit risk management.This paper starting from commercial banks credit risk characteristics and causes,according to the latest international and domestic credit risk regulatory requirements,introduced internal ratings and external ratings of credit risk, and according to the reality ofour country, it is recommended that the two should be joint development. On the basis ofcomparison and analysis of credit risk measurement methods and models, this paper usingthe Z-score model as an example, empirically examined the effects of direct application ofthe foreign mature model in China, confirmed that the applicable model must be based ondomestic conditions. After selected, the method of multivariate discriminant analysis andLogistic regression analysis that based on accounting data and market value was chosen forthe main method of build commercial banks credit risk measurement models in ourcountry.In order to establish the model, the article details the financial factors of commercialbanks credit risk measurement, analyzed the selection approach and dimension reduction techniques of the credit risk factors, established the idea that using stepwise selectionfiltering variables and using principal component analysis for data dimension reduction.Then, according to the definition of financial distress and domestic habits, selected78financial normal company samples and78financial distress company samples, and a totalof33financial ratios of the seven categories from our stock market, using2010data andSPSS software to build a six-variable stepwise discriminant analysis model and afour-variable Logit model. Taking into account the financial data with high dimension andhigh correlation, using the method of principal component analysis extracted5principalcomponents from18indicators which have significant group differences, and constructed adiscriminant analysis model and a Logit model which based on principal componentanalysis.Upon examination, four models are effective models, and have certain predictivepower in advance. By comparison, the accuracy of the Logit model is higher than thediscriminant analysis model, but the stability of the discriminant analysis model is betterthan the Logit model. Also found that the principal component analysis can improve thepredictive power in advance and the stability of the model, and that the Logit model ismore significant. In addition, the four models have the feature of follow macroeconomicfluctuations in the same direction change. We called it “Procyclicality”. During aneconomic boom, each model is loose, and vice versa. By appropriate adjustments to thecritical point, you can change the ratio of the two types of error rates and thus weaken the“Procyclicality”.The results of the study show that multivariate discriminant analysis and Logisticregression analysis are still effective credit risk measurement methods. The principalcomponent analysis is effective data dimensionality reduction method. The models thatbuilt in this paper can be used to measure the credit risk of commercial banks in China.更多还原