I 摘要 自2003年我国把房地产业作为支柱产业，房地产迎来了近二十年的快速发 展时期，世界500强榜单中也不乏房地产企业的身影，期间房价也屡创新高， 但近年来政策导向发生变化，国家层面一直强调房地产业的“去杠杆”，2020 年8月出台了房地产融资监管新规，为控制房地产企业有息债务增长，设置了 “三道红线”，以防范房地产业引发系统性金融风险。随着 “房子是用来住的， 不是用来炒的”房地产业发展根本原则的确立，房地产行业的资产负债结构和信 用管理面临着比以往更加严格的监管。从宏观经济形势看，研究房地产企业的 信用风险评级问题，具有重要的现实意义。 房地产企业天然具有资产负债率高，融资结构及融资来源较为单一，资金 需求大的特性，因此潜在的信用风险巨大。今年受新冠肺炎疫情影响，房地产 企业的销售普遍不理想，未来三年房地产企业普遍又迎来偿债高峰，房地产企 业的信用违约风险压力不断加大。在此背景下，充分揭示和预警行业的风险情 况，提前对房地产企业和房地产企业债券投资者、房地产企业的其他债权人进 行风险预警具有重要意义。在行业尚未大规模爆发实质违约前，充分揭示和预 警行业的风险情况，一方面有助于房地产企业计量自身的信用风险水平，做好 信用风险管理；另一方面，有助于投资者和债权人提前合理判断风险，有效控 制风险敞口。本文基于沪深交易所房地产上市公司、全国中小企业股份转让系 统房地产上市公司和银行间市场、交易市场公开发行债券发行人等披露的数据， 通过数据获取、数据清洗、违约标记、指标筛选、模型建立和模型验证等流程， 基于5种不同的指标筛选方法分别构建了5个基于逻辑回归方法的信用评分模 型，并使用ROC曲线和KS曲线分别对5个模型进行有效性评估，得出5种不 同指标筛选方法下较优的筛选方法为广义交叉验证法、逐步回归法和Boruta法， 基于这三种方法构建得出的评分卡以及行业的批量评级结果分布特征来看，收 入规模较小、盈利能力较弱、资产负债水平较高、现金流回收较慢的房地产公 司面临更为显著的违约可能性。从这三种方法所构建的模型得出的风险排序结 果中分别选取违约风险较大的30家主体进行风险提示，经过进一步比较，重复 摘要 II 出现在风险提示名单的主体有12家，这也说明了所推荐的三种方法筛选出的指 标所构建的模型具有一定的有效性、可靠性与实用性。最后，本文基于研究结 果，提出了系统化的信用风险管理建议，如借助量化结果构建各项业务的黑白 名单作为业务准入标准，并对已准入的主体进行稳健的授信等。 关键词：房地产企业；信用风险；信用评级模型；逻辑回归方法； Abstract III Abstract Since China took the real estate industry as the pillar industry in 2003, the real estate industry has ushered in a rapid development period for nearly 20 years. There are also many real estate enterprises in the list of the world's top 500. During this period, the house prices have also reached new highs. However, in recent years, the policy orientation has changed, and on the national level has always stressed the "deleveraging" of the real estate industry, In August 2020, new regulations on real estate financing supervision were introduced. In order to control the growth of interest-bearing debts of real estate enterprises, "three red lines" have been set to prevent the real estate industry from causing systemic financial risks. With the establishment of the fundamental principle of real estate development that "the house is for living, not for speculation", the asset liability structure and credit management of the real estate industry are facing more stringent supervision than ever before. From the macro-economic situation, it has great practical significance to study the credit risk rating of real estate enterprises. Real estate enterprises have the characteristics of high asset-liability ratio, single financing structure and financing source, and large capital demand, so the potential credit risk is huge. Due to the coVID-19 epidemic, the sales of real estate enterprises are generally not ideal this year, in the next three years, real estate enterprises will generally meet the peak of debt repayment, and the pressure of credit default risk of real estate enterprises is increasing. In this context, it has great significance to fully reveal and warn the risk situation to real estate enterprises, bond investors of real estate enterprises, and other creditors of real estate enterprises in advance. Give early warning before the real default of real estate enterprises breaks out on a large scale is helpful, on the one hand, it is helpful for real estate enterprises to measure their own credit risk level and do a good job in credit risk management; on the other hand, it is helpful for investors and creditors to judge risks in advance and effectively control risk exposure. Based on the data disclosed by the real estate listed companies of Shanghai and Shenzhen Stock Exchange, the real estate listed companies of the National Small and Medium-sized Enterprise Share Transfer System, and the issuers of bonds issued in the inter-bank market and trading market, through the process of data acquisition, data cleaning, default marking, indicator screening, model Abstract IV establishment and model validation, five credit scoring models based on logistic regression method are constructed based on five different index screening methods, and using ROC curve and KS curve to evaluate the effectiveness of the five models, it is concluded that the better model are generalized cross validation method, stepwise regression method and boruta method. Based on the score card constructed by these three methods and the distribution characteristics of the batch rating results of the real estate industry, the income regulation is as follows the real estate companies with small model, weak profitability, high level of assets and liabilities and slow cash flow recovery are more likely to default. From the risk ranking results of the models constructed by these three methods, 30 entities with higher default risks are selected for risk warning. After further comparison,12 subjects repeatedly appeared in the risk prompt list. This also shows that the model constructed by the three methods is effective, reliable and practical. Finally, based on the research results, this article proposes systematic credit risk management recommendations, such as constructing a black-and-white list of various businesses with the help of quantitative results as business access standards, and granting stable credit to the entities that have been admitted. Keywords： real estate enterprise；credit risk；credit rating model； logistic regression method 目 录 I 目 录 摘要 ............... I Abstract ....... III 目 录 ............. I 第一章 绪论 . 1 第一节 研究背景和研究意义 .................... 1 一、研究背景 ............................ 1 （一）房地产行业债务现状分析 ....................... 1 （二）房地产行业债务现状成因分析 ............... 6 二、研究意义 ............................ 6 （一）理论意义 .................... 6 （二）现实意义 .................... 7 第二节 国内外相关研究动态及文献综述 7 一、国外相关研究及文献 ........ 7 二、国内相关研究及文献 ........ 8 三、文献综述小结 .................... 8 第三节 研究方法和研究内容 .................... 9 一、研究方法 ............................ 9 （一）文献研究法 ................ 9 （二）比较分析法 ................ 9 二、研究思路和主要研究内容 9 三、数据来源、统计方法、统计工具选择背景 .. 11 第四节 本文的创新与不足 ...................... 12 一、创新之处 .......................... 12 二、不足之处 .......................... 12 第二章 相关概念及理论 .......... 14 第一节 信用风险相关概念 ...................... 14 一、信用风险度量及其重要性 ............................. 14 二、信用风险度量模型比较 .. 14 第二节 本文理论方法概述 ...................... 17 目 录 II 一、模型开发指标筛选方法 .. 18 二、模型构建方法 .................. 19 三、模型验证方法 .................. 21 四、统计分析工具说明 .......... 22 第三章 模型所需数据准备 ...... 23 第一节 数据的获取 ... 23 一、建模样本数据获取 .......... 23 二、违约样本数据获取 .......... 24 第二节 数据的预处理和初步探索 .......... 25 第三节 技术性违约认定和违约样本标记 ............................. 27 第四章 变量筛选结果及比较 .. 29 第一节 使用随机森林法筛选入模指标 .. 29 第二节 使用相对重要性比较法筛选入模指标 ..................... 30 第三节 使用广义交叉验证法筛选入模指标 ......................... 30 第四节 使