基于某商业银行大量法人客户的信息数据，运用多元线性判别、Logit 回归、
主成分技术，在界定了正常和违约客户的基础上，本文运用 SPSS 软件对不同行
业、地区的法人客户的两类违约风险进行了实证模型研究。

基于 Merton 模型的基本原理，本文对商业银行大量的非上市公司进行信用
风险度量。通过能有效反映违约特性的历史财务信息，运用九个参数估计模式，
选择得到有较好拟合度的参数估计方程，实证的得到了违约概率与经验违约距
离之间的函数关系，并由此对模型的有效性进行了检验。

基于简约模型的基本原理，依托某商业银行内部的信用评级系统及其大量
历史数据信息，通过设计能够反映商业银行客户违约特性的统计方法，得到大
量的分欧亿·体育（中国）有限公司、分地区、分客户等级的历史违约概率数值，由此利用 Logit 模型进
行参数回归拟合，并将历史违约数据和企业财务指标进行有机整合，实现对客
户违约的简约模型实证研究。

综合统计、结构、简约模型的技术方法和参数指标，本文提出了一个两阶
段的信用风险组合模型，同时利用非线性变量技术和新边界模型技术，对组合
模型结果做进一步的优化，得到了相对已有模型适用性更好、覆盖面更广、有
效度更高的模型函数。基于此，我们对三种模型的违约特性进行了比较分析。

在对当前我国商业银行运用信用风险模型存在的制约和困难进行深入分析
的基础上，本文对如何构建规范、有效、可行的信用风险模型及系统进行了深
入思考，并提出了建设相应管理体系的初步构想和实施步骤。

最后，在汇总本文实证模型的基础上，本文对研究创新点进行了说明，并
简述了尚待解决的问题和后续研究建议。

关键词：信用风险；统计模型；结构模型；简约模型；组合模型
Abstract
Credit risk has existed almost everywhere for a long time. This paper presented
the development process and main classification of the credit risk model, analyzed the
characteristics and the essence content of the model method, and compared various
new methods, new techniques, tools, systems, and then elaborated the development
track of the structure model.
With a great deal of information data for companies of the commercial bank, and
by the multivariate discriminate analysis, Logit regression, principal component
analysis, and with the definition of the normal and default companies, this paper used
the SPSS software carried on a lot of models to do research into two types of default
risks of companies in different industries and regions.
According to the Merton model principle, the paper applied credit risk research
in a grate deal of private companies of the commercial bank. With the history finance
information to reflect the default characteristic, and with nine parameter-stimulation
models, the paper choosed the formula of the best one, and then demonstrated to
know the function of relationship of default risk and experiential default distance, and
carried on validity examination.
According to the reduced-form model principle, with the internal credit rating
system and a great deal of history data of companies of the commercial bank, the
paper set up a statistical method which could reflect the characteristic of default for
the commercial bank's customers, and got a great deal of historical default probability
of different industries, regions, customers degrees, and with the Logit model to carry
on parameter regression, combined with the historical default data and financial data
of companies, got a empirical research for default risk of companies.
Combined with the statistical model, structure model and reduced-form model's
methods and parameters, this paper put forward a two-stage model for credit risk in
form of hybrid model. With non-linear parameter and bounded technology, the paper
optimized the hybrid model and got a more widely used and more valid functions.
According to this, we carried on a comparative analysis for the three kinds of models
in default characteristic.
In carrying on thorough analysis of the restriction and difficulty of credit risk
model used by commercial banks in our country currently, the paper shows how to setup
a normative, valid, feasible credit risk model system, and put forward the
construction and implement steps for the credit risk model management system.
At last, with the foundation of gathering the empirical models, the paper
explained research innovation point, and the problem which need to be solved
recently and the suggestion of subsequent research.
Key words: credit risk, statistical model, structure model, reduced-form model,
hybrid model