【作者】 张目；

【导师】 周宗放；

【作者基本信息】 电子科技大学 ， 管理科学与工程， 2010， 博士

【摘要】 深入分析自主创新能力等非财务因素对高技术企业信用风险的影响,并将影响较为显著的因素纳入到高技术企业信用风险评价指标体系中;探索、运用先进的定量分析方法和手段对高技术企业信用风险进行科学评价,都将有助于提高信用评价结果的客观性和准确性。对畅通和拓宽高技术企业的融资渠道、提高技术成果转化率、促进高技术产业持续健康发展具有重要的理论意义和现实意义。为此,本文对自主创新能力等非财务因素对高技术企业信用风险的影响、基于分类的高技术企业信用风险评价方法等问题进行了探索性研究。主要内容概括如下:第一,为了提炼出可能对高技术企业信用风险有重要影响的行业(地区)因素,本文对高技术企业信用风险的行业(地区)差异进行了识别。首先,从借款人信用等级转移的角度,遵循CreditMetrics模型的基本假设和风险识别的前瞻性要求,构建基于Markov链的高技术企业信用风险行业(地区)差异识别系统,其中,运用基于投影寻踪和最优分割的企业信用评级模型获得高技术企业的信用状态空间和信用等级;然后,以高技术产业上市公司为例,对我国高技术产业主要行业的信用风险进行识别,同时,对我国东、中、西部地区高技术企业信用风险差异进行识别。其中,基于投影寻踪和最优分割的企业信用评级模型的建模思路为:运用投影寻踪对样本企业进行信用综合评分,将信用综合得分由大到小排序,生成有序样品序列;利用最优分割法对有序样品进行聚类,得出明确的聚类结果;将最优分割点对应的信用综合得分作为划分信用等级的阈值,从而实现对样本企业的信用评级。第二,自主创新是高技术企业生存和发展的生命线,为考察自主创新能力对高技术企业信用风险的影响,须先对高技术企业自主创新能力进行科学评价。为此,本文首先提出一种基于联系度的改进TOPSIS法。该方法将理想点与负理想点视为确定不确定系统中相互对立的集合,在考察目标方案与理想点或负理想点的联系度时,充分考虑了对立集合的存在;并通过引入联系向量距离的概念,计算相对贴近度,从而在一定程度上克服了传统TOPSIS法的不足。然后,在基于联系度的改进TOPSIS法中加入时间维,构建动态综合评价模型,对我国高技术产业自主创新能力进行分行业动态评价。第三,基于柯布-道格拉斯生产函数和净现值法,对企业违约行为进行分析,从理论上初步解析了企业自主创新能力与信用风险的关系。在此基础上,构建高技术企业信用风险分析的Cox模型,将自主创新能力、财务因素、成长性、企业规模、地区因素和行业因素等作为协变量,通过Cox回归分析,实证检验上述因素对高技术企业信用风险的影响程度和影响方向,并考察引入自主创新能力对高技术企业信用风险评估结果的影响。第四,考虑到高技术企业信用评价指标体系中存在定性指标,本文对可处理定性指标的高技术企业信用评价方法进行了研究。针对传统云重心评判法的不足,借鉴TOPSIS法基本思想,基于理想状态和负理想状态,对综合云重心向量进行归一化,并采用修正的加权偏离度来衡量云重心的变化,由此提出一种改进的云重心评判法。将该方法应用于高技术企业信用评价,可较好的处理定性概念与定量表示的相互转换。第五,针对高技术企业信用状况的两类分类问题,提出一种基于多目标规划和支持向量机(SVM)的企业信用评估模型。基于TOPSIS法,分别以“正常企业”样本逼近理想点、“违约企业”样本逼近负理想点为目标,构建多目标规划模型;运用实码加速遗传算法求解得出指标综合权重,通过构造加权样本,减少两类样本企业信用状况的重叠,可在一定程度上提高SVM的预测精度。第六,针对高技术企业信用状况的多类分类问题,基于“非降维”的思路,提出一种基于投影寻踪和K-均值聚类的企业信用评级模型。首先,运用投影寻踪对样本企业进行信用综合评分,以反映原高维数据的结构或特征;然后,利用核密度估计法对信用综合得分序列进行分布密度估计,并根据密度函数的局部极大值点来确定初始聚类中心;最后,运用K-均值算法获得最终聚类中心,并划分企业信用等级,从而实现对样本企业的信用评级。更多还原

【Abstract】 It is helpful to improve the objectivity and accuracy of credit risk evaluation that determining the influence of the non-financial factors, such as independent innovation capacity et al, to credit risk of high-tech enterprises, and then introducing those significant factors into the credit evaluation index system of high-tech enterprises; exploring and applying the advanced quantitative analysis method and means to scientifically evaluating credit risk of high-tech enterprises. Our study will have important theory meaning and practice value to dredge and widen the financing channels of high-tech enterprises, enhance the conversion ratio of technological achievement, and promote the sustainable and healthy development of high-tech industry. Therefore, in this thesis, those problems, such as the influence of the non-financial factors, such as independent innovation capacity et al, to credit risk of high-tech enterprises, and the credit risk evaluation method of high-tech enterprises based on classification et al, are explored and investigated.The main research contents are as follows:1. For the sake of extracting the industry (or regional) factor which may have significant influence to credit risk of high-tech enterprises, we attempt to identify the industry (or regional) difference of high-tech enterprises credit risk in China. Firstly, from the perspective of borrower credit rating transfer, following the basic hypothesis of CreditMetrics model and the prospective requirements of risk identification, a system for identifying industry (or regional) difference of high-tech enterprises credit risk based on Markov chain is constructed, where the credit state space and credit rating of high-tech enterprises is obtained through a new credit rating model for enterprises based on Projection Pursuit and optimal partition. Secondly, taking the high-tech listed companies in China as samples, the empirical analysis on industry (or regional) difference of high-tech enterprises credit risk is carried out.Where the modeling approach of credit rating model for high-tech enterprises based on Projection Pursuit and optimal partition as follows: (1) Using Projection Pursuit, the comprehensive credit score of each sample is obtained. After sorting the comprehensive credit score descending, the ordered samples series is generated. (2) A clustering analysis of the ordered samples is carried out with the optimal partition method, so the clustering results are obtained definitely. (3) Each optimal partition point is regarded as the threshold to divide the credit grades, and then the credit rating for enterprises is achieved.2. Independent innovation is the lifeline of survival and development of high-tech enterprises. Therefore, in order to investigate the influence of independent innovation capacity to credit risk of high-tech enterprises, we must scientifically evaluate the independent innovation capacity of high-tech enterprises. Firstly, we propose an improved TOPSIS method based on connection degree. In this method, the ideal point and negative ideal point is regarded as mutual opposition set in a system both having certainty and uncertainty. When inspects the connection degree between the objective project and the ideal point or negative ideal point, the opposition set’s existence is considered fully. Using the connection vector distance redefined by us, the relative similarity scale is calculated. So the draw back of the traditional TOPSIS method is overcome to a certain extent. Secondly, through adding time dimension in the improved TOPSIS method based on connection degree, a new dynamic comprehensive evaluation model is constructed. Using this model, the independent innovation capacity of high-tech industries in China is evaluated dynamically.3. Based on Cobb-Douglas production function and net present value method, an analysis on the default behavior of enterprises is carried out. Therefore, the relationship between independent innovation capacity and credit risk of enterprises is preliminarily explained theoretically. On this basis, a Cox model to analyze credit risk of high-tech enterprise is constructed. Let independent innovation capacity, financial factors, growth phrase, enterprise scale, regional factor and industry factor be covariate, through the Cox regression analysis, the effect degree and direction of those factors above on high-tech enterprise credit risk is tested. And then, the influence of independent innovation capacity to the results of credit risk evaluation of high-tech enterprises is investigated.4. In view of qualitative index existing in the credit evaluation index system for high-tech enterprises, we also study the credit evaluation method for high-tech enterprises which can process qualitative index. According to the draw back of traditional MCGC, using the basic idea of TOPSIS method for reference, the comprehensive membership cloud gravity center vector is normalized based on ideal state and negative ideal state. Through the modified weighted deviation degree, the change of membership cloud gravity center is measured scientifically. Thus, an improved MCGC is proposed and applied in credit evaluation of high-tech enterprises. The empirical analysis results show that the improved MCGC can successfully process the mutual conversion of qualitative and quantitative.5. In view of the classification problem of two-types of samples, we propose a credit risk evaluation model for high-tech enterprises based on multi-objective programming and Support Vector Machines (SVM). Based on TOPSIS method, respectively taking the“normal enterprise”sample similarity to ideal point and the“default enterprise”sample similarity to negative ideal point as the goal, the multi-objective programming model is established. Using real coded accelerating genetic algorithm (RAGA), above model is solved, and then the combination weight of index is obtained. Through constructing the weighted sample, the overlap of the credit conditions of two types of samples is reduced. As a result, the predicting accuracy of SVM can be raised to a certain extent.6. In view of the classification problem of multi-types of samples, based on the idea of‘non-dimension reduction’, we propose a new credit rating model for high-tech enterprises based on Projection Pursuit and K-means clustering algorithm. Firstly, using Projection Pursuit, the comprehensive credit score of each sample is obtained, so as to reflect the structure or characteristics of original multi-dimensional data. Secondly, the distribution density of the comprehensive credit score series is estimated by the kernel density estimation method, and then the initial cluster centers are determined according to the local maximum points of density function. Finally, using K-means clustering algorithm, the final cluster centers are obtained, and then the credit grades are partitioned. Thus, the credit rating for enterprises is realized.更多还原