随机森林在制造业上市公司信用风险评价中的应用

发布时间：2018-06-26 01:17

本文选题：随机森林 + 信用风险　；参考：《暨南大学》2017年硕士论文

【摘要】：国内外学者一直将信用风险及其计量模型作为经济、金融等领域的研究前沿。随着我国经济迈入“新常态”这一重要战略机遇期,制造行业面临着宏观政策调整、产业转型升级等一系列挑战。制造业不仅是实体经济的支柱行业,而且在资本市场中也占据半壁江山,在我国金融市场日趋完善的大背景下,出现连续亏损、信用风险的企业将被市场和投资者抛弃,因此,信用风险评价模型的建立不仅可以丰富该领域的研究,还可以给信息使用者提供更加可靠的判断依据。本文在近几年国内外学者对随机森林和信用风险问题的探索、研究基础上,以制造业上市公司为研究对象,搜集并整理了160家公司两年中期报告的财务数据,共计包含五个维度的26个财务指标。研究思路包括异常样本的筛减、指标体系的构建以及随机森林模型的建立和相关参数的优化,我们把样本划分为训练集和测试集,将训练集用于模型的训练,将测试集用于模型预测效果的检验,发现随机森林有较高的分类正确率。为了更加直观的反映随机森林在解决目标行业信用风险问题上的优越性能,我们运用本次实验数据,将优化前后的随机森林模型与CART、SVM模型进行对比分析,得到了随机森林模型性能更优、判别更准确的结论。总的来说,将随机森林理论应用在信用风险评价领域,通过建立和优化模型来评价和识别风险,对于丰富理论研究和服务实体经济都有着重要的意义。
[Abstract]:Domestic and foreign scholars have always regarded credit risk and its measurement model as the frontier of economic and financial research. With China's economy entering an important strategic opportunity period of "new normal", the manufacturing industry is facing a series of challenges, such as macro-policy adjustment, industrial transformation and upgrading. The manufacturing industry is not only the pillar industry of the real economy, but also occupies half of the capital market. Under the background of our country's financial market becoming more and more perfect, the enterprises with credit risk will be abandoned by the market and investors. The establishment of credit risk assessment model can not only enrich the research in this field, but also provide a more reliable basis for information users. On the basis of the research on stochastic forest and credit risk in recent years, this paper collects and collates the financial data of 160 companies in the interim report of two years, taking listed companies in manufacturing industry as the research object. A total of 26 financial indicators with five dimensions are included. The research ideas include the sifting of abnormal samples, the construction of index system, the establishment of stochastic forest model and the optimization of relevant parameters. We divide the samples into training sets and test sets, and apply the training sets to the training of models. The test set is used to test the effect of model prediction, and it is found that the classification accuracy of random forest is higher. In order to reflect more intuitively the superior performance of stochastic forest in solving the problem of credit risk in target industry, we use the experimental data to compare and analyze the stochastic forest model before and after optimization and the SVM model of CARTN. The conclusion that the stochastic forest model has better performance and more accurate discrimination is obtained. In general, the application of stochastic forest theory in credit risk assessment, through establishing and optimizing models to evaluate and identify risks, is of great significance for enriching theoretical research and serving the real economy.
【学位授予单位】：暨南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F832.51;F424;F406.7

【参考文献】