【作者】 徐志春；

【导师】 王宗军；

【作者基本信息】 华中科技大学 ， 企业管理， 2012， 博士

【摘要】 中小企业融资难是个长期以来困扰世界的难题,特别在我国,表现得尤为明显。中小企业融资难有多方面的原因,其中受信息不对称的影响,商业银行信贷从业人员对中小企业贷款违约行为的规律性无从把握,从而产生“惜贷,惧贷”现象,或片面强调抵押担保的作用,提高了中小企业融资门槛。因此研究中小企业信用风险管理中的规律性问题,提高信用风险管理的技术水平是当前的现实需要。本文从贷前、贷后和组合管理三个方面来考察中小企业信用风险问题,主要做了如下工作：首先,采用了因子分析、定性指标的模糊化处理、层次分析、自适应模糊神经网络分析等方法,构建了中小企业贷前信用风险自适应模糊神经网络评价模型。以往很多研究采用神经网络方法来评价信用风险,但大多缺乏对财务数据的不精确性和非财务因素的考察。本文针对以往研究的不足,采集了65家信贷企业样本数据,用模糊化方法来预处理非财务信息,用因子分析法预处理财务数据,利用中小企业贷前信用风险自适应模糊神经网络模型得到分类结果。通过与BP神经网络、XU G、LOGISTIC回归分类等模型比较,表明本文构建的模型有更高的分类精度。其次,通过描述性统计分析、财务指标因子分析、财务因子logistic回归分析以及加入非财务变量的logistic回归分析,构建了中小企业贷后信用风险LOGISTIC回归预警模型。目前尚没有针对中小企业贷后信用风险预警的专门研究,常见的是企业破产预测、财务困境分析、企业失败预测,通常而言这类模型主要基于会计变量,以企业的财务特征比率为解释变量。本文建立了融合财务和非财务因素的预测模型,研究结果表明：预警模型的分类正确率优于随机分类；财务变量有提前预警能力,要重视对财务状况的监测分析;非财务变量的确能提高模型的预警能力。最后,本文引入Copula函数优化传统的计算VAR的模型,构建了中小企业贷款组合经济资本配置Copula计量模型。传统对贷款组合经济资本计量的方法需要对贷款组合的分布作出假设,一般假设贷款组合服从正态分布,比如CreditMetrics模型以及BASELⅡ中所建议的方法,但众多的理论和实证研究表明这一假设不符合实际情况,实证研究表明贷款组合收益的分布有明显的“厚尾”特征,基于正态分布的假设可能会低估风险值。为此,本文提出用Copula函数的方法来改进CreditMetrics模型,这一方法不需考虑分布假设就能计算联合分布。本文通过Monte-Carlo方法进行了模拟计算,并与传统的VaR方法进行了比较。模拟结果表明中小企业贷款组合经济资本配置Copula计量模型能比较好地解决贷款组合的“厚尾”分布问题,较传统的VaR方法更能精确地计量中小企业贷款组合的经济资本需求量。更多还原

【Abstract】 Small and Medium-sized Enterprise (SMEs) financing is a challenge besetting the world, particularly in China. It is seriously unbalanced between the supports to SMEs in social funding and the contributions of SMEs to society. There are many reasons accounting for this situation. A significant one is that under the influence of information asymmetry, the credit officers in Chinese commercial banks are unable to grasp the law of loan default, thereby they always prefer to lend to large enterprises rather than SMEs, or they unilaterally emphasize the role of collaterals, which result in an increasing threshold for SME financing. As a result, it is necessary to do some researches on the rules of credit risk management to improve our levels of credit risk management.This article discusses credit risk of SMEs from three different perspectives:the timing of granting credit and post-lending as well as portfolio management of SMEs. Research findings indicated that different information should be emphasized in different scenarios.At first, through adopting factor analysis, Fuzzy analysis, AHP analysis, Fuzzy-neural adaptive analysis, the article build a Fuzzy-neural Adaptive model to deal evaluate credit risk of SMEs. There are many researches focusing on credit risk evaluation, but many of them haven’t taken the inaccuracy of financial information and the vagueness of non-financial information into considerations. In order to overcome the shortage of the present research, the article collects 65 samples of SMEs applying loans from banks. By inducing factor analysis and fuzzy analysis as pretreatment process to reduce the complexity of data, the results are inputted to Fuzzy-neural Adaptive network model. Thereby, we can get the final classifying result. By comparing the output of the model and the other models, such as BP, XU_G, Logistic, we can conclude that Fuzzy-neural Adaptive Network model can get the most accurate result.Secondly, this article constructs a post-lending credit risk warning model for SMEs based on Logistic regression method by employing some tools, such as description analysis, factor analysis, Logistic regression. There are not unique researches focusing on post-lending credit risk warning model. The similar topics are enterprise bankrupting prediction, enterprise failure and so on. Generally speaking, these models are often based on accounting variables, and take the financial ratios as explanation variables. Differed from the present researches, this article successfully combines financial and non-financial factors. The result shows that:(1) warning model has a better performance than random model; (2) financial variables have the warning ability before risk happens; (3) non-financial information can really improve warning performance.At last, this article construct economic capital allocation model for loan portfolio of SMEs by taking Copula function to optimize VaR method. An important assumption in calculating VAR through traditional method is the general normal distribution assumption of loan portfolio, but numerous theoretical and empirical studies have shown that this assumption is not realistic. Empirical studies showed that income distribution of loan portfolio has distinct "heavy-tailed" feature, and the normal distribution assumptions may underestimate the risk values. Thereby, Copula function was introduced to deal with this problem. Copula functions were used to improve the calculation of probability of joint distributions of Credit Metrics model. The Monte Carlo method was employed to simulate the model and compare the result of VAR based on traditional distribution assumption and the t-Copula methods. The result indicated that the t distribution assumption based on Copula method can simulate the heavy-tailed distributions, and close to the realities of credit risk, as a conclusion, the traditional VAR method underestimated the value of credit risk.更多还原