【作者】 过新伟；

【导师】 梁琪；

【作者基本信息】 南开大学 ， 金融学， 2012， 博士

【摘要】 中小企业（Small and Medium Enterprises，SMEs）在我国经济社会发展中发挥着日益重要的作用。但是长期以来由于各种原因，中小企业一直面临融资难的问题，引起了政府部门、商业银行等社会各界的广泛关注。影响中小企业融资难有多方面的原因，但是其中最根本的原因是商业银行与中小企业之间存在信息不对称问题。本文认为，解决银企间信息不对称的重要方法是发展一套适合评估中小企业信用风险的度量方法，从而在对中小企业信用风险进行准确衡量的基础上，开展授信工作，能够极大地缓解中小企业融资难问题。基于上述认识，本文对我国上市中小企业和非上市中小企业的信用违约风险进行实证研究，采用离散选择模型实证分析了影响中小企业信用风险的财务因素和非财务因素，并建立了信用风险预警模型。具体来说，本文的主要发现和结论有如下两点：第一，对上市中小企业而言，我们以财务失败事件（ST事件）作为信用风险的代理指标，使用改进后的logistic模型构建信用违约模型。研究发现，一方面，有必要按企业规模分层次对中小企业单独建模，以获得更好的风险度量效果。另一方面，在财务指标基础上加入公司治理指标能显著提高上市中小企业信用违约风险的度量准确性，而且这种风险度量准确性的提升在排除了含有财务指标极端值样本（outliers）的情况下，并采用截面数据随机效应logistic回归时表现得更加显著。第二，对非上市中小企业而言，我们以企业的信用评级作为信用风险的衡量指标，使用多元有序logistic响应模型进行建模。研究发现，非上市中小企业的财务信息、非财务信息以及外部机制等因素对信用风险都具有显著影响。在财务信息的基础上，加入非财务信息和外部机制等信息能显著提高模型的度量效果。研究进一步发现，非上市中小企业的信用风险在企业规模、所属行业、年龄等方面存在异质性。按资产规模、所属行业、年龄对中小企业进行细分并建立模型，其风险度量的准确性高于未进行样本细分的模型。因此，在构建非上市中小企业信用风险模型时有必要考虑企业的异质性特征。本论文的研究价值和可能的创新之处主要体现在以下四个方面。第一，国内目前对信用风险的研究大部分是以大型上市企业为对象，而专门对中小企业信用违约风险的研究还非常有限，本文的研究则丰富了国内有关中小企业信用风险度量的研究成果。第二，在中小企业样本选择上，本文不仅对我国上市公司中的中小企业信用违约风险进行实证研究，而且还通过调查问卷的方式，获取了非上市中小企业的第一手数据，并在此基础上建立了非上市中小企业信用风险的度量模型。因此，我们的研究结论对商业银行开展中小企业信贷风险评估具有一定的参考意义。第三，在中小企业信用风险度量模型的指标选取上，已有的国内研究大多只考虑以财务指标建立模型，而对非财务指标较少关注。本文对上市中小企业样本和非上市中小企业样本建立模型时，不仅考虑了财务指标的作用，而且还进一步考察了公司治理等非财务指标在度量中小企业信用风险中的作用。第四，在信用风险度量模型的选择上，对上市中小企业而言，我们以目前主流的logistic模型为基础，采用新近发展的稳健logistic估计方法和基于截面数据的随机效应logistic回归对logistic模型进行改进。研究结果显示，改进后的logistic模型相比普通logistic模型具有更高的风险度量效果和较好的稳健性。对于非上市中小企业，本文则采用多元有序logistic响应模型对多状态信用风险问题进行处理，多元有序logistic模型相比二元logistic模型能对中小企业的信用风险状态进行更细致的刻画和衡量。更多还原

【Abstract】 Small and Medium Enterprises (SMEs) play an increasing role in nationaleconomy and social development. However, due to all kinds of reasons, there existsfinancing difficulty for SMEs for a long time which has caused extensive concernamong government, commercial banks and so on. There are various reasons forfinancing difficulty for SMEs, one of which is the problem of asymmetricinformation between commercial banks and SMEs. This paper argues that one of theeffective ways to solve the problem of asymmetric information is developing asuitable approach to measure the credit risk of SMEs, and then, based on the accuratemeasurement for credit risk of SMEs, the loan available for SMEs will increase andfinancing difficulty for SMEs will be relieved to a great extent.This paper attempts to empirically research the credit default risk for Chineselisted SMEs and non-listed SMEs, employing the discrete choice models. Weinvestigate the financial factors and non-financial factors in modeling credit risk andestablish risk measure model. Specifically, our main conclusion are following twopoints.First, for listed SMEs, on the one hand, we take the event of financial distress(ST event) as the proxy indicator of credit risk, using the logistic model to establishcredit default model. We find that, it is necessary to separate SMEs with asset sizeand to model credit risk separately in order to obtain a better effectiveness of riskmeasurement. On the other hand, the risk measurement effectiveness is highlyimproved when the model specification takes corporate government indicators intoaccount, and will be furtherly improved when we remove outliers from sample aswell as introduce random effect logistic regression based on cross-sectional data.Secondly, for non-listed SMEs, we take the credit rating as the proxy indicator ofcredit risk, using the multivariate logistic ordered response approach to model creditrisk. We find that, the financial information and non-financial information andexternal support mechanism have important influence on credit risk of SMEs. On thebasis of financial information, the risk measurement effectiveness is also largely improved when the logistic ordered response model specification takes non-financialinformation and external support mechanism into account. Our finding also showsthat there exists heterogeneity in size, industry, age and etc. among the non-listedSMEs. The risk measurement effectiveness based on the sample separated by assetsize, industry and age is better than the risk measurement effectiveness based on thecomplete sample. Therefore, it is also necessary to take the heterogeneitycharacteristics into accout when modeling credit risk for non-listed SMEs.The research value and potential contribution of this paper are following fouraspects.First, the sample are usually selected with large listed companies in mostexisting literatures of credit risk measurement, but literatures based on SMEs samplespecially are rare, this paper enriches the research about SMEs’ credit riskmeasurement.Secondly, on SMEs sample, besides the listed SMEs, we also model the creditrisk measurement model for non-listed companies employ non-listed SMEs from aquestionnaire about SMEs financing. Therefore, our finding have credit pricingimplications for the commercial banks.Thirdly, on credit risk measurement indicators of SMEs, most of the existingliteratures only focused on the financial indicators but ignored the non-financialindicators. We employ not only financial indicators but also governance indicatorsand other non-financial indicators to model credit risk for both listed SMEs andnon-listed SMEs.Forthly, on model of credit risk measurement, we consider to employ the currentpopular model——logistic model, and improve it with the newly-developing robustlogistic regression as well as random effect logistic regression based oncross-sectional data. The empirical result shows that the risk measurementeffectiveness of improved logistic model is better than logistic model, and theestimators of improved logistic model is also more robust than the logistic model. Fornon-listed SMEs, we use multivariate logistic ordered response model to treat themultistate credit risk, which can model the credit risk more accurate than binarylogistic model.更多还原