【作者】 徐佳丽；

【导师】 彭寿康；

【作者基本信息】 浙江工商大学 ， 金融学， 2010， 硕士

【摘要】 自1990年12月19日和1991年7月3日上海证券交易所和深圳证券交易所分别成立,并提供股票交易服务以来,中国证券市场经历了快速发展的历程。在这快速发展过程中,因发生财务危机而被特别处理的上市公司数目也在逐渐增加。’为了能够保护利益相关者的权益、增强企业抵御风险的能力,对上市公司财务危机预测的研究成了财务理论界和实务界关注的焦点。企业财务危机预测研究作为企业经营的指示灯,不仅具有较高的学术价值,而且有着很大的现实意义。企业财务危机的预测研究是通过筛选有信息含量的指标,借助一定的分析方法构建预测模型。其中,如何准确、合理的选择预测指标和方法是企业财务危机预测研究的关键,然而传统的指标筛选法(如T检验、逐步回归等)和预测模型存在一定缺陷,阻碍了企业财务危机预测研究的进一步深入。出于对这些问题的浓厚兴趣和对其应用价值的关注,本文在对国内外文献回顾和总结的基础上,通过构造企业远期偿债能力的度量指标ω,通过引入信号噪音差方法筛选预测指标,通过对朴素贝叶斯分析法的改进等几种新的现代技术方法,构建企业财务危机的线性与非线性预测模型,并通过比较模型预测准确率的高低,来对模型进行比较。本文以制造业上市公司为代表,借助SAS统计软件和Data Miner数据挖掘软件,构造不同方法下的企业财务危机预测模型进行研究。本文通过考察传统的指标筛选方法(T检验和逐步回归),得出T检验的缺陷在于不能给出单个指标的具体预测价值,并且T检验筛选出的预测指标数量偏多,影响了模型构建的效果,增大模型的使用成本。逐步回归方法的缺陷在于估计所得的指标系数无法说明指标和预测结果的关系,仍然没能给出单个指标的预测信息量,本文引入的信号噪音差方法解决了这些问题。文章同时也利用由逐步回归筛选的指标和信号噪音差所选指标分别构建了多元线性判别分析模型和朴素贝叶斯模型,实证结果显示信号噪音差所选指标构造的模型预测精度高于逐步回归筛选指标所建模型,并且信号噪音差所选指标构造的决策树模型具有较高的预测能力,肯定了信号噪音差在指标筛选方面的优势。鉴于以往研究一般利用公司的横截面数据来构建模型,不同的研究样本和研究数据可能得出不同的预测变量,造成预测模型的外推能力不强,并且难以反映企业从财务正常到财务危机的渐进过程。本文通过分析企业陷入财务危机的原因,提出一个新的预测变量,并利用财务报表时间序列数据和贝叶斯推断方法估计该变量,由该变量所构建的单变量预测模型和多变量预测模型都取得较好的预测准确率,为财务危机的预测研究提供了参考。更多还原

【Abstract】 Since December 19,1990 and July 3,1991,the Shanghai Stock Exchange and Shenzhen Stock Exchange set up to provide stock trading service,the Chinese security market has experienced rapid development course. In this rapid development process,the number of listed companies with special treatment has been increasing,because of financial crisis.In order to protect the interests of stakeholders, and enhance listed company’s ability to resist risk of financial crisis,the research of listed company’s financial crisis prediction model has been the focus of theory and practition community. As the indicator of business operation, the research of financial crisis not only has high academic value, but also has great practical significance.Prediction of enterprise financial crisis is through the selection of indicators with information content, using certain methods to build prediction models.Among them how accurately choose the predictor and analysis method is the key to the research, however, the traditional indicators of screening methods (such as T tests,stepwise regression, etc.) and prediction models have some defects,hindering further development of enterprise financial crisis prediction.For the interest on these two issues and concern about the value of their applications,based on the review and sum up of the literature at home and abroad,this paper constructed a measure of corporate long—term solvency indicator,introducted signal-noise method and through the improvement of Naive Bayesian analysis,etc.Building linear and nonlinear prediction models,and comparing level of prediction accuracy of each model.This paper take manufacturing sector as example, using SAS statistical software arid the Data Miner software,constructed different financial crisis prediction models.-In this paper,by screening traditional indicator choosing methods (T test and stepwise regression),finding defects in T test are that it can not give a single indicator specific predictive value,and with huge number indicators,affecting the model building effect,increasing the cost of model using.The drawback of stepwise regression method derived from the indicators estimated coefficients can not account for the relationship between targets and indicators,meanwhile still unable to predict the amount of information given in a single indicator,this paper introduces the signal to noise ananlysis to solve these problems.The article also used indicators selected by the stepwise regression and signal to noise analysis constructing multivariate linear discriminant analysis model and a naive Bayesian model, empirical results show that the prediction accuracy of models using indicators selected by signal to noise analysis is higher than that of stepwise regression,and the decision tree model constructed by indicators selected by signal to noise analysis has a high predictive ability,confirmed the signal to noise analysis’advantage. Past studies have generally used the company’s cross-sectional data to build models,different study samples and research data may get different predictor variables,resulting in poor extrapolation power of predictive models,and is difficult to reflect the gradual process for business changing to financial crisis.This paper analyzes the reasons for companies in financial crisis,proposing a new predictor variable,and the use of financial statements for time-series data and Bayesian inference methods to estimate the variable,the single-variable forecasting models and multi-variable forecasting models constructed by it achieving higher prediction accuracy,providing a reference for financial crisis prediction research.更多还原