摘要
2007年我国开始发行企业债券，借着经济发展的东风，公司债券的发展
十分快速。直到 2014年 3月，国内公司债券出现了第一例违约事件。可以说
这并非偶然，在随后的几年里，债券违约事件越发频繁，尤其是在 2018年后，
债券违约的数量及规模如“火箭”般直线上升，究其原因，是国内企业制度、
市场发展、市场监督、投资者素质等各方面的因素综合起来造成。通过对近
些年来我国金融市场的发展进行洞悉发现，诸如信贷违约、信托违约等越来
越多的违约事件出现，给市场发展带来了很多不确定性。基于高投资回报的
诱惑，人们往往经不住诱惑，在诱惑面前缺失理性，抱有侥幸心理挺而走险，
导致各种违规违法事件出现，市场风险加剧。
这篇文章通过进一步研究分析这些上市企业的对外财务报表，基于盈利、
经营、偿债能力等指标分析公司的财务状况，并深入判断财务风险，从财务
风险的角度去分析公司债券可能面临的信用风险等等，进一步帮助企业做出
投资决策。即使财务比率、财务报告在很多时候都会被质疑：财务造假、财
务被粉饰这些情况出现，考虑到当前市场和学术界的情况，使用财务信息是
辨别信贷风险的最简单、最有效的方法。在这篇文章中，我们避免使用由许
多条件和模型的实用性所限制的复杂模型。由于通常需要基于财务信息的优
点来建立和使用模式，所以其适用性很容易被投资者接受。据专家介绍，考
虑到时间、成本和有效性，首先需要选择财务信息。而且，从实际应用效果
来看，上市公司发行的债券有可能发挥非常好的风险判断效果。即便在判定
风险是无法作出一个比较清晰明了的量化性定义，但通过判断，投资人能够
以自身的实际情况来决定自己自己是否需要对该企业做投资。
本文中，笔者在机器学习的理论基础和数学原理以及数据挖掘的实际操
作方法进行了具体的研究，证明了数据清洗和数据降维在使用机器学习训练
算法的过程中的举足轻重的地位，使用了 SPSS这一强大的数据处理软件和
Python这个新晋的编程界“顶流”软件，通过实证分析验证了模型的可信度。
用 SPSS进行了数据处理和独立性 T检验等工作，利用 Python编程软件分别
验证了决策树模型和 lightGBM模型的在评估企业信用风险方面的准确性，
并对其评估结果进行对比，总结出一个最好的方式。
II

关键词：信用风险；公司债券；财务风险；机器学习
III

Abstract
In 2007, China took the lead in issuing bonds. With the help of the east wind
of China's economic development, the development of corporate bonds was very
rapid, and in March 2014,the first substantial default occurred. This is not an
accident, because in the following years, bond defaults have become more
frequent. Especially after 2018, the number and scale of bond defaults have risen
like a “rocket”. Behind it lies the system, market, regulators, and issuers. The
issues of debtors, investors, intermediaries, and so on are thought-provoking.
Looking at my country's financial market in recent years, the risks and
probabilities of various default events are increasing. For example, credit defaults
and trust defaults have frequently come into people's eyes. In the face of relatively
high returns, people are often unable to withstand the temptation, lack rationality,
and credit risk-related information, there is no comprehensive risk compensation
mechanism. Fully assumed all the losses caused by the credit risk.
This article uses the public financial information of listed companies to find
indicators that can represent the company’s profitability, operation, and solvency
from financial data, and use these indicators to make judgments on the financial
risks of the analysis objects, and evaluate the credit risks of corporate bonds based
on financial risks Qualitative identification to make investment decisions. Even
though financial reports, financial data, and financial ratios are often easily
criticized by people: they are easy to be faked and whitewashed. Considering the
current market and academic conditions, using financial information is the
simplest and easiest way to distinguish between credit risk and corporate bonds.
effective method. This article avoids the use of complex models that are restricted
by many conditions and the practicality of the models. Based on the advantages of
financial information, its application is easier to be accepted by investors, because
the establishment and use of models are usually sought after by professionals.
Considering time, cost and effectiveness, financial information should be the first
choice. And from the actual application effect, the credit risk of corporate bonds
can be identified. Although it is impossible to give an accurate quantitative
indicator for the identification of corporate bond credit risk, through judgment,
IV
。。。以下略