【作者】 卢志红；

【导师】 郑丕谔；

【作者基本信息】 天津大学 ， 数量经济学， 2004， 硕士

【摘要】 本文以信用风险评价问题为核心，通过采用多元判别分析模型、Probit和Logit模型、概率神经网络模型以及数据包络分析对我国企业的信用风险度进行信用评分模型的建立。在研究方法上，本文采用理论研究和实证分析并重的方式。在问题研究上，着重定量分析的运用，从而得到既有理论依据同时也具有现实可操作性的解决方法。 本文首先对现有信用风险评估方法－专家评分法，信用评分模型，现代信用风险度量模型进行了综合论述。接下来讨论了各种方法的优缺点，以及各种方法在中国目前的可行性。然后介绍了信用评分模型建立的主要方法，多元判别分析和二元Probit、Logit模型，并且引入新的方法：概率神经网络和数据包络分析技术，对这两种方法在信用风险评估中应用的可行性进行论述。最后将以上四种方法应用于我国上市公司贷款信用风险度的评估，采用主成分分析法进行指标选取。 本文综合了财务管理、计量经济学和信息技术相关内容，运用多种方法建立企业信用风险度和多维财务指标之间的量化关系。通过实证结果表明，建立的信用评分模型具有较好的预测效果。更多还原

【Abstract】 The main purpose of this thesis is to find out how to quantify creditrisk which appeared in loan appraisal. Multi-Discriminate analysis、Probitand Logit models、Probabilistic Neural Networks and Data EnvelopmentAnalysis are all adopted to develop credit scoring models. To find thebest and practical method, a quantified analysis and experimental studyare emphasized.. First existing methods were introduced, namely specialist’s scoremethod, credit score model, and modern credit risk models. Thenadvantages and disadvantages of each method were discussed and theadaptability in China is presented. Some popular techniques fordeveloping credit scoring models such as multiple-discriminate analysis,probit and logit models were also introduced. In addition, probabilisticneural networks and data envelopment analysis techniques for creditscoring are presented. Lastly four techniques were applied tomeasurement of credit risk for listed companies in China, in which mainfactor method was adopted to select input variables. Experimental resultsshow credit scoring models developed here have good forecasting ability.更多还原