我国A股上市公司财务危机预警模型实证研究

发布时间：2018-03-03 12:34

本文选题：上市公司　切入点：财务危机预警　出处：《厦门大学》2014年硕士论文　论文类型：学位论文

【摘要】：近年来,我国A股市场上出现了许多因财务问题被特别处理的上市公司。由于目前我国证券市场还处于弱势有效阶段,集中表现为信息的不对称性,所以财务危机不仅使公司自身蒙受巨大损失,还给利益相关者带来了经济损失和负面影响,甚至会对市场环境造成恶劣冲击。因此,研究如何有效利用公开数据,建立可靠、稳定的财务危机预警模型对解决上述问题能发挥较为积极的作用。在介绍了选题背景及研究意义的基础上,本文明确了研究框架和方法。随后,系统梳理了国内外学者在财务危机领域的相关研究文献,对财务危机的概念界定、形成原因和预警理论等方面都进行充分的探讨,并系统介绍了以往研究中所采用的特征指标和建模方法。据此,界定了实证分析中财务危机的具体定义,确定了指标的初选方向和建模方法。结合选定的建模方法,本文详细阐述了随机森林算法的相关理论。这一理论结合了非参数决策树和Bagging算法,对过拟合免疫,适用于解决输入变量多、先验信息不足等复杂问题。本文着重介绍了随机森林的特征选择功能,并指出其变量重要性的计算结果是有偏的这一不足,引入了可以计算变量无偏条件重要性的条件森林Cforest。在此基础上,以A股市场上的234家正常公司和78家ST公司为样本,本文分别运用随机森林和Cforest筛选特征指标,对比了两者的选择结果,并从财务学角度阐明了Cforest的计算结果更具合理性。在确定特征指标和最优参数之后,本文建立了基于随机森林算法的财务危机预警模型,在不同的市场行情下运用该模型进行预警,并用混淆矩阵评估其性能。评估结果显示该模型在两种市场条件下都有较高的准确率,具有良好的自适应性和稳定性。为了进一步验证随机森林模型的高效性,本文还利用了另外一种变量选择方法Lasso,建立了基于Lasso-logistic回归的财务危机预警模型,并对该模型性能进行评估。评估结果显示,虽然Lasso-logistic预警模型也能较有效地筛选指标和进行预警,但是与随机森林预警模型相比仍然存在一些的局限和不足。根据上述的理论研究和实证分析结果,本文认为基于随机森林建立的财务危机预警模型在市场实践中具有较高的实用价值,并为上市公司、投资者和债权人等其他市场参与者提供了相关参考建议。最后,为能给财务危机预警实践提供更多帮助,本文展望了未来进一步研究的方向。
[Abstract]:In recent years, there have been many listed companies in China's A-share market that have been specially dealt with because of their financial problems. At present, the securities market in China is still in a weak and effective stage, which is concentrated on the asymmetry of information. Therefore, the financial crisis not only makes the company suffer huge losses, but also brings economic losses and negative effects to the stakeholders, and even has a bad impact on the market environment. Stable financial crisis warning model can play a more active role in solving the above problems. On the basis of introducing the background and significance of the research, this paper clarifies the research framework and methods. Then, it systematically combs the relevant research literature of domestic and foreign scholars in the field of financial crisis, and defines the concept of financial crisis. The formation reasons and early warning theory are discussed, and the characteristic indexes and modeling methods used in previous studies are systematically introduced. Based on this, the specific definition of financial crisis in empirical analysis is defined. The primary direction and modeling method are determined. Combined with the selected modeling method, this paper elaborates the related theory of stochastic forest algorithm, which combines the nonparametric decision tree and Bagging algorithm, is immune to over-fitting, and is suitable for solving the problem of multiple input variables. This paper mainly introduces the feature selection function of random forest and points out that the calculation result of its variable importance is biased. A conditional forest Cforestwhich can calculate the importance of variable unbiased condition is introduced. On this basis, with 234 normal companies and 78 St companies in A share market as samples, the selection results of random forest and Cforest were compared. The results of Cforest are more reasonable from the point of view of finance. After determining the characteristic index and optimal parameters, a financial crisis warning model based on stochastic forest algorithm is established in this paper. The model is used for early warning under different market conditions, and its performance is evaluated with confusion matrix. The evaluation results show that the model has high accuracy under both market conditions. It has good self-adaptability and stability. In order to further verify the efficiency of the stochastic forest model, this paper also uses another variable selection method, Lasso, to establish a financial crisis early warning model based on Lasso-logistic regression. The performance of the model is evaluated. The results show that although the Lasso-logistic early warning model can screen indicators and carry out early warning more effectively, there are still some limitations and shortcomings compared with the stochastic forest early warning model. According to the above theoretical research and empirical analysis results, this paper thinks that the financial crisis early warning model based on stochastic forest has higher practical value in the market practice, and it is listed company. Other market participants, such as investors and creditors, provide relevant reference suggestions. Finally, in order to provide more help to the practice of financial crisis warning, this paper looks forward to the direction of further research in the future.
【学位授予单位】：厦门大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F275;F276.6;F832.51

【参考文献】