【作者】 边海容；

【导师】 万常选；

【作者基本信息】 江西财经大学 ， 管理科学与工程， 2013， 博士

【摘要】 市场经济的发展,使得公司间的竞争日益激烈,而全球经济一体化,带来的不仅仅是发展机遇,也暗藏了无尽的危机和风险。对上市公司而言,因为发生财务危机而被特别处理甚至被迫退市,不仅影响其自身的生存和发展,还会给投资者和债权人带来巨大的经济损失。因此,对上市公司财务危机准确、及时、有效地预警,无疑对促进资本市场和国民经济的发展,维持社会稳定具有重要作用。财务危机预警研究主要涉及两个方面：预警指标和预警模型。已有的研究在预警指标方面主要是选用财务指标,然而由于财务指标的滞后性和易被操纵等固有缺陷,也有学者积极引入如宏观经济变量、公司治理变量、产业变量等非财务指标,但是由于非财务指标类型众多,数据不易获取,而且有些指标难以量化,这些都为在财务危机预警中非财务指标的引入造成障碍。而由于网络技术的发展,Web金融信息大量涌现,其所具有的实时性、覆盖性、全面性和易获取等特点,正好弥补了非财务指标获取困难以及不完整、不全面等不足,为财务危机预警中非财务指标的获取提供了新的途径。对财务危机预警模型的研究,已有的研究成果从传统的统计模型到人工智能模型,大多是用截面数据所建立的静态预警模型,而上市公司的财务危机并不是突然发生,有一个逐渐演化的过程,静态预警模型没有考虑到预警指标的时序特征,忽略了历史数据对预警结果的影响,从而造成预警模型的早期预警效果较差,在实际应用中难以推广。本文从以上两个方面入手,一方面,研究如何将Web金融信息引入上市公司财务危机预警指标体系以及Web金融信息指标的预警作用；另一方面,研究上市公司财务危机的动态预警。围绕这两大方面,本文具体研究了以下内容：(1)研究了Web金融信息的量化问题。因为Web金融信息基本上是非结构化的文本信息,所以,要将它纳入到财务危机预警指标体系,需要将它进行合理地量化,文本内容的情感倾向值计算是文本信息量化的常用手段。本文针对Web金融信息文本,构建了金融领域情感词典,提出了基于语素分数的情感倾向值计算方法。(2)分析了Web金融信息和上市公司财务状况的关系。针对Web金融信息和上市公司财务状况的关系,本文主要研究了两个方面。首先,通过相关性分析研究了Web金融信息指标(即信息热度和情感值)与财务指标的关系；其次,运用Logistic回归分析研究了Web金融信息指标对预警上市公司是否会被ST的影响。(3)验证了Web金融信息指标对财务危机预警模型的影响。本文运用LIBSVM分别构建了纯财务指标预警模型和财务指标与Web金融信息指标相结合的混合模型,通过实证比较分析,发现加入Web金融信息指标后,预警模型在超前性、稳定性和有效性等方面都有很大程度的改善。(4)研究了上市公司财务危机的动态预警。本文首先运用计量经济学的ARMA模型,对上市公司财务状况的时序特征进行拟合；然后借用质量管理学上的控制图思想,将Web金融信息情感褒贬程度指标加入EWMA,构建了上市公司财务危机动态预警模型S-EWMA;最后对其进行实证分析,与EWMA进行了对比分析。本文的创新性工作体现在：(1)提出了基于语素分数的Web金融信息文本情感倾向值计算方法。目前,文本情感倾向性计算方面已有不少的研究成果,但是计算方法往往受到种子词选择和情感词典覆盖性等方面的限制,而且专门关于金融领域文本情感计算的研究尚未发现。本文所提出的基于语素分数的Web金融信息文本情感倾向值计算方法,具有领域针对性,能充分满足金融领域文本情感倾向性分析的要求。首先,构建了金融领域的情感词典,将金融领域的特色情感词添加到词典中,而基于语素分数的情感值计算方法很好地解决了情感词典的覆盖性问题；其次,在进行句子和文档的情感倾向值计算时,充分考虑文档结构中的否定词和程度副词对文档情感倾向所起的修饰作用,考虑了句子在文档不同位置对情感倾向的贡献不同,从而被赋予不同权重,以及子句间连接词的转折、递进、并列等模式对句子情感倾向的影响,而不是将各组成部分的情感值进行简单地求和。实验结果也验证了本文计算方法的有效性。(2)分析了Web金融信息和上市公司财务状况的相关性。本文通过对Web金融信息和上市公司财务状况的关系分析,发现Web金融信息文本情感值中含有财务指标未曾包含的与上市公司财务状况相关的信息,因此Web金融信息情感值可以作为上市公司财务指标的重要补充。(3)构建了财务指标与Web金融信息指标相结合的上市公司财务危机预警模型。本文结合Web金融信息指标和财务指标构建了混合指标的上市公司财务危机预警模型,实证结果表明,该模型在预警的有效性、稳定性和超前性等方面均优于纯财务指标模型。(4)构建了加入Web金融信息情感褒贬程度指标的上市公司财务危机动态预警模型S-EWMA.该预警模型基于财务指标的动态面板数据ARMA模型而构建,能很好地反映财务指标的时序特性,又在指数加权移动平均控制图中加入了Web金融信息情感褒贬程度指标,弥补了财务指标滞后性等缺陷,能够反映上市公司财务危机逐步演变发展的动态性及演变趋势,能够有效地预警财务危机发生的时点。实证分析表明,该模型可以较大地提高财务危机预警的超前性。更多还原

【Abstract】 With the development of market economy, the competition among companies becomes more and more intense. The global economic integration, not only brings the development opportunities, but also the endless crises and risks. Due to financial crises, listed companies have to be treated specially or forced to delist, which not only affects their own survival and development, but also brings enormous economic loss to investors and creditors. Therefore, accurate, timely and effective early warning of listed companies’financial crisis will promote the development of the capital market and the national economy and meanwhile maintain social stability.Existing works of financial crisis early warning focus on two aspects:early warning indicators and model. For the first one, previous studies used to select financial indicators. However, these indicators have the inherent shortcomings, such as lag, easily manipulated. So, such non-financial indicators like macroeconomic variables, corporate governance variables, industry variables, etc are presented. However, non-financial indicators of diversity, non-available of data and hard index quantification, all lead to be difficult to introduce non-financial indicators. With development of network technology, it has rapidly evolved to an emerging a large number of Web financial information. The characteristics of real-time, diversity, comprehensiveness and accessibility is just to compensate the defects of non-finance indicators and provides a new way to get non-financial indicators for financial crisis early warning.For the model of financial crisis early warning, existing works mainly build static warning models based on one-period cross-sectional data by using traditional statistical models or artificial intelligence models. As we are seen that the listed company’s financial crisis will not be happened suddenly and will be last a gradual process of evolution. However, in static warning model, the timing characteristics of early warning indicators have not been taken into account and the historical data’s the impact on the results have been ignored as well. Thus, it is less effective for the models’early warning and hard to popularize in practical applications.This thesis focuses on two parts of work. On the one hand, we study how to introduce Web Financial Information into indicator system of listed companies’ financial crisis early warning, and its role in early warning. On the other hand, the dynamic warning of listed companies’financial crisis is presented.The following problems are addressed in this thesis.(1) We initiate the problem of quantification about Web Financial Information. Due to the unstructured text, Web Financial Information must be quantified reasonably can we introduce it into indicator system of the financial crisis early warning. Calculation the text sentiment tendencies value is a common means of text messages quantification. We construct the sentiment dictionary of financial domain, and propose a calculation method of sentiment tendencies values based on evaluation scores of morpheme.(2) We analyze the relationship between the Web Financial Information and listed companies’financial situation. In this thesis, two aspects are studied. At first, we study the relationship between indicators of Web Financial Information (information heat and emotional value) and financial indicators through correlation analysis. Subsequently, we utilize the Logistic regression analysis to study whether the indicators of Web Financial Information will affect the early warning.(3) We verified the impact of indicators of Web Financial Information on the model of financial crisis early warning. By LIBSVM, an early warning model of pure financial indicators, and an early warning model of mixed indicators with the Web financial information indicator are constructed respectively. The empirical comparative analysis show that, the model of mixed indicators is better than the model of pure financial indicators in the validity, stability and advancing of early warning.(4) We also research into the problem of the dynamic warning of listed companies’financial crisis. Firstly, we use the econometrics ARMA model to fit the Timing Characteristics of the listed companies’financial position. Secondly, learned from the thought of control chart in the quality management, a dynamic early warning model for listed companies’financial cricis(S-EWMA)is constructed by adding the emotional inclination value of Web financial information to EWMA. Finally, the comparison between EWMA and S-EWMA is performed through empirical analysis.The contribution of this thesis can be summarized as follows:(1) We propose an approach of sentiment tendencies calculation based on morpheme value for web financial text information. The research community of text sentiment tendencies calculation has been made significant progresses, however, the existing reported solution are still far from perfect. The main issue is that the current methods are limited by the seed opinion lexicon selection and coverage of it. Furthermore, the findings on text sentiment tendencies calculation of financial text have not been found yet. The proposed method in this thesis could fully meet the demand of text sentiment tendencies analysis to financial information. Firstly, a sentiment dictionary of financial-domain is constructed, in which sentiment words for financial-domain are identified and the sentiment tendencies classification(calculation) approach solves the problem of coverage. Secondly, many effect factors are considered in the sentiment tendencies of the whole document, such as the modification funcition of negative word and degree-averb, different position of sentence, and turning, parallel, progressive mode.The experimental results have proved that the proposed method is feasible and effective.(2) We analyze the relationship between the web financial information and listed companies’financial situation. We found that the opinion orientation of Web financial information contains some information of listed companies’financial situation, which have not been included in the financial indicators. Thus, the opinion orientation of Web financial information could be used as important information supplement of listed companies’financial indicators.(3) Combing web financial information indicator with financial indicators, an early warning model is constructed. The experiment results show that the model of mixed indicators outperforms the pure financial indicators in the validity, stability and advancing of prediction.(4) An dynamic early warning model-S-EWMA for listed companies’ financial crisis is built, in which the opinion orientation of web financial information is added. The model has the following the advantages:firstly, by the dynamic panel data of financial indicators, the model could reflect sequential variation characteristics of financial indicators. Secondly, due to adding the opinion orientation of web financial information in the Exponentially Weighted Moving Average model(EWMA), the proposed model could make up financial indicators’lag and other inherent defects and meanwhile reflect the gradual evolution and evolution trend of listed companies’ financial crisis. Finally, it could give the time point of financial crisis of early warning effectively. Empirical analysis shows that the model is superior to other models.更多还原