我国商业银行信用风险违约概率判别及预测模型研究

发布时间：2018-06-23 21:55

本文选题：商业银行 + 信用风险　；参考：《安徽大学》2014年硕士论文

【摘要】：随着金融市场的迅速发展,各国的金融监管机构对风险的管理和监控也逐步加强,信用风险因其重要性及影响深远,成为风险管理的重中之重。特别是对于商业银行来说,能否有效管理和控制信用风险会对其盈利水平和稳健能力产生决定性的影响。因此,从2004年新资本协议出台到2010年巴塞尔协议III的问世,负责监控发达国家银行业风险的巴塞尔委员会在不断总结风险管理经验的基础上,为国际银行的信用风险管理提出意见的同时,也加强了对商业银行的风险控制的要求,设定了更高的标准。在巴塞尔新资本协议中涉及了关于商业银行信用风险管理的关键内容即内部评级法,其主要内容就是测算借款人的违约概率。近年来,我国商业银行也加强了对信用风险的管理和监控,大力推进内部评级体系建设,信用风险管理水平明显提高,但国内大部分商业银行,特别是中小商业银行距新资本协议以及巴塞尔协议III的要求和国际银行业的先进水平仍有一段距离。国外的先进银行对信用风险管理的研究已经有很长一段时间的历史,并积累了研究数据,因此信用风险管理的模型能够不断推陈出新并得以应用。我国商业银行应当结合我国的经济环境与金融发展程度,借鉴与吸收国外银行先进的信用风险管理经验与思想,研究与开发能够适用于我国的优秀的信用风险违约概率判别及预测模型。本文从描述信用风险的特征和违约的概念入手,对信用风险度量方法进行综述。为了清晰地展现信用风险度量方法发展的历程以及各种方法的特点,将信用风险度量方法分为古典方法、传统方法和现代方法。本文的实证分析,是以在我国证券市场2011年和2012年被披露发生ST的公司作为违约公司的样本,以随机抽取非ST的上市公司作为非违约公司的样本,并将样本分为训练集和测试集,并通过财务指标组间的均值检验和相关性检验的筛选,选取8个财务指标(资产负债率、资本积累率、总资产增长率、财务杠杆系数、营业收入现金比率、营业利润率、应收账款周转率、固定资产周转率)作为建立信用风险度量模型的自变量,然后分别建立Bayes判别模型、Logisitic模型以及BP神经网络模型,对样本公司的信用违约进行判别和预测。从实证结果来看,Bayes判别模型在对上市公司信用违约的判别和预测时效果相对不太理想,Logisitic回归模型和神经网络模型更胜一筹,其中,Logisitic回归模型,对上市公司信用违约的预测性比神经网络模型更高,从而认为Logisitic回归模型是这三者之中最优的信用风险违约概率计量模型。最后,针对我国商业银行信用风险管理存在的信用风险管理技术相对落后、不具备信用数据库、缺乏成熟的信用评级机构等问题,提出建立先进的适合我国商业银行的信用风险模型、建立和完善商业银行信用数据库、以及建立具有专业素质的第三方信用评级机构等建议。
[Abstract]:With the rapid development of the financial market, the management and monitoring of risk is gradually strengthened by financial regulators in various countries. Credit risk has become the most important factor in risk management because of its importance and far-reaching impact. Especially for commercial banks, the effective management and control of the risk of credit will produce a decision on its profitability and stability. Therefore, from the introduction of the new capital agreement in 2004 to the advent of the Basel agreement III in 2010, the Basel Committee, which is responsible for monitoring the risk of the banking industry in the developed countries, on the basis of constantly summarizing the experience of risk management, puts forward some suggestions for the management of the credit risk of the international bank, and also strengthens the risk control of the commercial banks. In the new Basel capital agreement, the key content of the credit risk management of commercial banks is the internal rating method, which is the main content of calculating the default probability of the borrowers.
In recent years, China's commercial banks have also strengthened the management and monitoring of credit risk, vigorously promoted the construction of the internal rating system and improved the level of credit risk management, but most of the domestic commercial banks, especially the small and medium-sized commercial banks, still have a new capital agreement, the requirements of the Basel agreement III and the advanced level of the international banking industry. Foreign advanced banks have been studying credit risk management for a long time, and accumulated research data. Therefore, the model of credit risk management can be constantly updated and applied. China's commercial banks should combine with the economic environment and financial development of our country, draw lessons from and absorb foreign banks first. The credit risk management experience and thinking, research and development can be applied to China's excellent credit risk default probability discrimination and prediction model.
This paper, starting with the description of the characteristics of credit risk and the concept of default, summarizes the method of credit risk measurement. In order to clearly show the course of the development of the method of credit risk measurement and the characteristics of various methods, the methods of credit risk measurement are divided into classical methods, traditional methods and modern methods. The empirical analysis of this paper is based on me. In 2011 and 2012, the national securities market was disclosed as a sample of default companies, which randomly selected non ST listed companies as samples of non default companies, and divided the samples into training sets and test sets, and selected 8 financial indicators (assets and liabilities) by means of the mean test and the screening of correlation inspection among the financial indicators. Rate, capital accumulation rate, total asset growth rate, financial leverage coefficient, operating income cash ratio, operating profit rate, account receivable turnover rate, fixed assets turnover rate as independent variables for establishing credit risk measurement model, and then establish Bayes discriminant model, Logisitic model and BP neural network model respectively, to Sample Firms's credit violation. According to the empirical results, the Bayes discriminant model has a relatively poor effect on the discrimination and prediction of the listed companies' credit default, and the Logisitic regression model and the neural network model are better, and the Logisitic regression model has a higher predictability than the neural network model for the credit default of the listed companies. It is considered that the Logisitic regression model is the best credit risk default probability measurement model among the three.
Finally, in view of the relative backwardness of credit risk management technology in the credit risk management of China's commercial banks, the lack of credit database and the lack of mature credit rating agencies, it is proposed to establish an advanced credit risk model suitable for China's commercial banks, to establish and perfect the credit database of commercial banks, and to establish a professional professional bank. The quality of the third party credit rating agencies and other recommendations.
【学位授予单位】：安徽大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F224;F832.33

【参考文献】