基于集成学习模型的上市公司财务困境判别研究

发布时间：2018-01-01 16:41

本文关键词：基于集成学习模型的上市公司财务困境判别研究　出处：《深圳大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着我国资本市场的迅速发展,企业面临的竞争日益激烈,上市企业的财务状况受到了严峻的考验。上市公司因财务状况异常而陷入困境的情况屡见不鲜,因此建立准确的财务困境判别模型显得尤为重要。从2006年到2015年10年的时间里,已经有321家公司被“ST”。上市公司被“ST”后,不但影响自己正常的生产经营,还让相关投资者和债权人蒙受损失。因此,利用上市公司发布的财务数据,建立上市公司财务困境判别模型以揭示风险,己成为上市公司管理者、投资者和债权人等相关利益方共同关注的问题。上市公司作为我国多层次资本市场建设的基石,其经营业绩、财务状况是资本市场健康发展的重要保障,建立上市公司的财务困境判别模型,也有利于资本市场实现其价格发现、风险转移、资源配置的功能。在介绍了选题背景及研究意义的基础上,本文明确了研究方法、主要内容和研究框架。随后系统梳理了国内外学者在财务困境领域的相关研究文献,对财务困境的概念界定、特征和判别的相关理论等方面进行探讨,为实证分析部分的输入指标的选择和建模方法应用提供思路。本文运用集成学习模型来提前判别上市公司会否陷入财务困境,首先对上市公司出现财务困境的相关特征进行总结归纳,本文选取了在2006到2015年期间被“ST”的A股上市公司作为研究对象,并以同行业、同规模、同类型的原则选择配对的正常企业,通过对比,来研究上市公司在陷入财务困境前三年的主要特征。考察的特征涵盖上市公司的报表项目特征,如资产、负债和所有者权益等项目,也涵盖上市公司的相关财务指标,如偿债能力、营运能力、盈利能力、现金流动能力、潜在发展能力等指标,此外还涵盖了上市公司的治理特征,如股权结构、董事会规模、大股东持股比例和外部审计等内容。在对上市公司陷入财务困境前的特征有一定了解的基础上,本文建立了相关指标体系,并且通过数据处理与严格分析,选择了11个指标作为主要判别变量,这些判别变量在一定程度上可以区分正常企业与陷入财务困境的企业。接着本文构建了随机森林财务困境判别模型,模型中包含1000棵随机树,并且在每一次运算时随机抽取6个判别变量,使用2006到2012的样本作为学习样本,利用2013年到2015年的样本作为检验样本,准确率高达83%,比传统分类方法高4到20个百分点。另外在模型中,对随机树木依次取0到500,随机抽取1到10个判别变量,随机森林算法的准确率依然保持高位。当随机树木个数大于100以后,准确率逐步稳定在80%以上;而抽取4到7个判别变量时,模型的准确率较高,这都体现了集成学习方法的稳健性。最后,从随机森林模型的结果中可以得到,每股收益与净资产收益率这两个指标对判别上市公司财务困境有着十分重要的作用,在指标体系中的重要性高达35%和20%。由于各种主客观因素,本文还有些许不足,例如样本量有限,未能覆盖到全部的上市公司,同时选取的指标也未能反映出导致上市公司陷入财务困境的所有因素。根据上述的理论分析与实证结果,本文认为基于集成学习方法对上市公司的财务困境判别模型对于企业管理者、投资者和债权人而言,都有较高的实用价值。
[Abstract]:With the rapid development of China's capital market, enterprises are facing increasingly fierce competition in the market, the financial situation of enterprises have faced severe challenge. It is often seen. listed companies because of abnormal financial condition and the plight of the situation, so it is particularly important to establish the model of financial distress accurately. From 2006 to 2015 10 years, there have been 321 companies were "ST". The listed company is "ST", not only affect their normal production and operation, also let investors and creditors suffer losses. Therefore, the financial data released by listed companies, the establishment of corporate financial distress discriminant model to reveal the risks, has become the common concern of managers of listed companies, investors and the creditors and other stakeholders. Listed companies as the cornerstone of the construction of China's multi-level capital market, its operating results, financial condition is the capital market. An important guarantee of healthy development, the establishment of discriminant models of financial distress of listed companies, but also conducive to the capital market to achieve its price discovery, risk transfer, the function of resource allocation. Based on introducing the background and significance of research, this paper introduced research methods, main contents and research framework. Then the system reviews foreign scholars in the field of financial distress related research literature, the concept of financial distress definition, characteristics and discrimination theory so as to provide ideas for the application of empirical analysis and modeling method of input selection part of the index. In this paper, using the integrated learning model to determine in advance of listed companies will fall into financial difficulties, first of all the relevant features the listed company's financial difficulties are summarized, the paper selected during the period from 2006 to 2015 by the "ST" of the A shares of listed companies as the research object, and the The same industry, the same size, the same type of normal business, the principle of selecting the pairing by comparison, to study the main characteristics of Listed Companies in financial distress three years ago. The study covers characteristics of listed companies report project, such as assets, liabilities and owners' equity project also covers listed companies related financial indicators such as, solvency, operating capacity, profitability, cash flow ability, potential development capacity, in addition to covering corporate governance features, such as ownership structure, board size, the proportion of large shareholders and external auditing content. Based on a certain understanding of the characteristics of Listed Companies in financial distress before the in this paper, establishes the index system, and through data processing and rigorous analysis, 11 indices were chosen as the main variables, these variables in a certain extent can distinguish positive Often companies and financial distress enterprises. Then we construct a random forest financial distress discriminant model, the model contains 1000 random tree tree, and in every operation were randomly selected from 6 discriminant variables, using 2006 to 2012 samples as study samples, from 2013 to 2015 as sample test samples, the accuracy rate of up to 83%, 4 to 20 percentage points higher than the traditional classification method. In the model, the random trees from 0 to 500 were randomly selected from 1 to 10, the accuracy rate of discriminant variables, random forest algorithm remains high. When the random tree number is greater than 100, the accuracy rate gradually stabilized at 80% from 4 to 7 and above; discriminant variables, model accuracy is higher, it reflects the integrated learning method robustness. Finally, random forest can be obtained from the model results, earnings per share and net assets yield two The index plays an important role in judging the listed company's financial difficulties, the importance in the index system of up to 35% and 20%. due to various subjective and objective factors, the article still has some shortcomings, such as limited sample size, not to cover all the listed companies, while the selected index also failed to reflect the result of all factors of listed companies financial distress. According to the theoretical analysis and the empirical results above, this paper thinks that the ensemble learning method based on corporate financial distress discriminant model for enterprise managers, investors and creditors, has high practical value.

【学位授予单位】：深圳大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F275;F832.51

【参考文献】