基于贝叶斯网络的上市公司财务危机预警研究

发布时间：2018-08-09 08:10

【摘要】：自2007年,美国次贷危机引发的金融危机席卷全球,使得各国金融市场剧烈震荡,经济也深受其累。然而,2009年爆发的欧债危机再一次将全球经济拖入泥潭。这不仅使市场风险急剧上升,也使风险管理面临更加严峻的挑战。因此,如何强化风险管理意识,提高风险预测精度,维护经济社会的和谐稳定,既是政府经济管理部门面临的主要任务,也受到了学术界的广泛关注。值得注意的是,随着中国证券市场的繁荣发展,越来越多的公司通过上市获得资金以扩大发展,上市公司已然成为我国经济发展的核心力量,加上财务危机在企业集团内部具有传染性,进一步引起政府经济管理部门和投资者等利益相关者对中国上市公司财务危机的高度关注。上市公司一旦发生信用违约,不仅会给投资者带来巨大损失,甚至还可能导致企业破产、社会动荡等严重后果。因此,构建科学有效地财务危机预警方法,具有重要的现实意义。基于此,本文以我国上市公司为研究对象,首先,运用正态性检验、参数与非参数检验和多重共线性检验提取出能显著刻画上市公司财务危机的特征指标;进而,既引入NB模型克服BN在初始网络结构学习上过于依赖样本数据导致网络结构学习复杂度增加的局限性,同时又引入基于约束的TPDA算法克服NB模型在结构学习中过度依赖的条件独立假设的局限性,构造出一种改进的贝叶斯网络模型——TPDA-NB模型对上市公司财务危机进行预警研究,并运用性能评价指标将TPDA-NB模型与NB模型、Logistic模型、神经网络模型进行对比分析;最后,运用配对样本T检验对各模型预测精度的差异性进行显著性检验,实证研究结果如下:(1)将Logistic模型与NB模型、TPDA-NB模型对比发现,在预测精度与预测稳定性上,不仅Logistic模型与NB模型之间存在显著性差异,而且Logistic模型与TPDA-NB模型之间的差异更加显著;(2)将神经网络模型与NB模型、TPDA-NB模型对比发现,在预测精度与预测稳定性方面,神经网络模型与TPDA-NB模型之间存在显著差异,而与NB模型之间的差异性较弱;(3)更为重要的是,TPDA-NB模型能够有效地提升NB模型对上市公司财务危机的预测精度和稳定性。以上实证研究结果表明:运用TPDA-NB模型能够较为准确的预测我国上市公司财务危机,这在风险管理领域具有广阔的应用前景。对于投资者而言,能够运用TPDA-NB模型提前捕捉风险信号,进而做出合理的投资决策以规避风险带来的损失;对于相关的政府经济管理者而言,能够运用TPDA-NB模型对可能发生风险问题的领域进行预测,及时制定合理的监管政策,从而稳定市场秩序,促进经济的持续健康发展。
[Abstract]:Since 2007, the financial crisis caused by the subprime mortgage crisis in the United States has swept the world. However, the European debt crisis in 2009 once again dragged the global economy into a quagmire. This not only causes the market risk to rise sharply, but also makes the risk management face more severe challenge. Therefore, how to strengthen the awareness of risk management, improve the accuracy of risk prediction, and maintain the economic and social harmony and stability is not only the main task of the government economic management department, but also has been widely concerned by the academic community. It is worth noting that with the prosperity and development of China's securities market, more and more companies have obtained funds through listing to expand their development, and listed companies have become the core force of our country's economic development. In addition, the financial crisis is contagious within the enterprise group, which causes the government economic management departments and investors and other stakeholders to pay close attention to the financial crisis of listed companies in China. Once a listed company defaults on credit, it will not only bring huge losses to investors, but also may lead to enterprise bankruptcy, social unrest and other serious consequences. Therefore, the construction of scientific and effective financial crisis warning method has important practical significance. Based on this, this paper takes the listed companies of our country as the research object. Firstly, using normal test, parameter and non-parameter test and multiple collinear test, we extract the characteristic indexes which can depict the financial crisis of listed companies. The NB model is introduced to overcome the limitation that the learning complexity of the network structure is increased due to the excessive reliance on the sample data in the learning of the initial network structure of BN. At the same time, the constraint based TPDA algorithm is introduced to overcome the limitations of the conditional independence hypothesis that NB model is over-dependent in structure learning, and an improved Bayesian network model, TPDA-NB model, is constructed to study the financial crisis of listed companies. TPDA-NB model, NB model, Logistic model and neural network model are compared and analyzed by performance evaluation index. Finally, the difference of prediction accuracy of each model is tested by paired sample T test. The empirical results are as follows: (1) comparing Logistic model with NB model TPDA-NB model, it is found that there are significant differences not only between Logistic model and NB model, but also between Logistic model and NB model in prediction accuracy and prediction stability. The difference between the Logistic model and the TPDA-NB model is more significant. (2) comparing the neural network model with the NB model TPDA-NB model, it is found that there are significant differences between the neural network model and the TPDA-NB model in terms of prediction accuracy and prediction stability. But the difference between NB model and NB model is weak. (3) more important is that TPDA-NB model can effectively improve the accuracy and stability of NB model for financial crisis prediction of listed companies. The above empirical results show that the TPDA-NB model can accurately predict the financial crisis of listed companies in China, which has a broad application prospect in the field of risk management. For investors, the TPDA-NB model can be used to capture the risk signal in advance, and then make reasonable investment decisions to avoid the risk of loss; for the relevant government economic managers, The TPDA-NB model can be used to predict the areas where risk problems may occur, to formulate reasonable supervision policies in time, to stabilize the market order and to promote the sustained and healthy development of the economy.
【学位授予单位】：成都理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：F275

【参考文献】