【作者】 罗圣雅；

【导师】 王光伟；

【作者基本信息】 苏州大学 ， 金融学， 2011， 博士

【副题名】基于罗吉斯模型、KMV模型之研究

【摘要】 自1997年亚洲金融风暴以来,世界各国陆续传出地雷股之消息,许多大型企业瓦解的事件也纷纷发生,使得公司债违约情形日益严重,再加上衍生性金融商品的推陈出新,金融机构在竞争激烈的市场中,为求生计及提升市占率情况下,相对利差逐渐缩小,银行风险性资产受到严重侵蚀,又如面临经济景气因素、公司治理或内部控制不良、金融商品操作不当等情形时,只要稍有不慎,逾放、呆帐便因应而起,造成金融机构体质不良,此伤害不只对该金融机构之股东、投资者形成直接损失,若因此被政府接管,亏损部份亦由所有纳税义务人全民买单,更进一步影响整个经济体系,严重时将形成国内金融危机,银行与金融机构可能产生更多的违约风险,因此企业信用风险之衡量与管理亦日趋重要。本文除了尝试以实务上较具应用性的Logit模型与KMV模型,并比较是否纳入总体经济变量及公司治理变量的Logit模型与KMV模型三种财务预警模式,在三种不同的切割值下,何者预期台湾企业发生财务危机之可能性的正确率较高,最后,本文亦应用严谨的绩效评估模型进行检定,以供银行授信的同时,进行信用风险评估的参考,有效建立企业财务预警制度,进而达成企业信用风险管理之目标。实证研究是以2000年至2009年发生财务危机的国内上市柜公司为研究对象,所筛选的变量除了KMV模型所应用的变量外,还考虑了总体经济变量及公司治理变量来建构罗吉斯预警模型,并依据不同的绩效指针来评估模型。依据实证结果分析,可得到以下四点结论:(1)模型的正确率以考虑总体经济变量及公司治理变量的Logit模型最佳。无论样本内外,都以考虑总体经济变量及公司治理变量的Logit模型最佳,其次为未考虑总体经济变量及公司治理变量的Logit模型,最差为KMV模型。这表示KMV模型仅考虑资产、负债、利率等变量,忽略了景气指针及公司治理变量而减低其预测能力。另外,不同的公司股价会因人为操作有高估或低估的情形,所以股价的市值及波动性不能反映真实价值,造成信用风险变量的显着性及解释能力较不佳,但部分公司预测能力会有提升的效果。(2)无论何种模型在样本内、外之绩效比较,都以切割值0.5为最佳。不同的分类切割值会影响模型的判别绩效,但理论上并无法验证何种切割值最佳。因此,本研究比较了三种不同的切割值,分别为03,0.5及0.8。实证结果显示,无论何种模型在样本内、外之绩效比较,都以切割值0.5为最佳。(3)由型一误差观察,仍以考虑总体经济及公司治理变量的Logit模型最低。但若考虑不同的切割值则以0.3较佳。型二误差的绩效则与正确率比较结果一致。而样本外的预测,亦发现与样本内比较一致。但以切割值为0.5时最佳。(4)总体而言,样本内的绩效要比样本外为佳。本文的实证结果显示,KMV模型的样本外绩效优于样本内。探究其因,可能与估计的参数适合于样本外数据评估。但Logit模型的样本外明显较样本内为低,其可能的原因是受限于企业本身内涵之信用风险涉险程度之不同,再者是因为模型无法学习到训练样本信息而造成测试样本有判断能力降低的现象。对于企业财务危机的研究,相关文献已累积许多成果,但利用样本内数据所求得之最佳违约点系数,再根据KMV模型及是否纳入总体经济变量及公司治理变量纳入Logit模型的三种财务预警模式进行样本外预测,最后将样本内数据与样本外数据进行绩效比较者并不多见。由本研究的分析得知,纳入总体经济变量及公司治理变量确实能够提升模型财务危机的预测能力。而这结果亦可做为金融机构在评估企业授信风险时非常重要的参考指标,亦即能有效预测企业之财务危机,以提供企业、债权人、投资者及政府机关作为有效的投资决策及风险管理的依据,进而降低投资风险及损失。更多还原

【Abstract】 Since the Asian financial crisis of 1997, most of the countries all over the world had spread the tank stock news one after another, collapse happened to many large enterprises in succession, which made the company default more and more serious with each passing day, with the weeding through the old to bring forth the new of the derivatives, with a view to making a living and enhancing the market share, the relative spread of the financial institute in the fierce competition market gradually shrunk, and the bank risk-adjusted asset was seriously eroded. Faced with the situations, such as the business cycle factor, poor corporate governance or internal control and improper financial product manipulation, as long as the financial institute was a little incautious, the overdue loan and bad debt would arise hereby, which caused the poor financial institute constitution, such injury not only caused direct damage to the stockholder and investor of the financial institute but the loss, if taken over by the government thereby, would be settled accounts by all the taxpayers as well. If seriously, the injury formed the internal financial crisis, and the bank and financial institute might generate more default risk; thus, the measurement and management of the enterprise default risk also became important with each passing day.The research motives of the study not only try the Logit model and KMV model of more applicability in practice but also compare three kinds of financial early warning model: whether to incorporate the Logit model of the macroeconomic variables and corporate governance variables and the KMV model. Which has a higher accurate rate to expect the possibility that a financial crisis happened to Taiwan’s enterprises under three kinds of different cut values? In the long run, the article also applies the rigorous performance evaluation model to conduct a test for the credit risk evaluation reference while the bank credits to establish effectively the financial early warning system of an enterprise to further achieve the objective of the enterprise credit risk management. The study uses the internal listed companies and OTC companies to which a financial crisis happened from 2000 to 2009 as the research object. The screening variables are the variables applied by the KMV model. The study also considers the macroeconomic variables and corporate governance variables to construct the logit early warning model and to evaluate the model in light of different performance index. Pursuant to the results of the empirical analysis, the four following conclusions can be acquired: (1) The Logit model which considers the macroeconomic variables and corporate governance variables has the best accurate rate. No matter it is inside or outside the sample, the Logit model which considers the foresaid is the best. The Logit model which does not consider the foresaid is the next. The worst is the KMV model. This indicates that the KMV model only considers the asset, liability and interest rate variable but neglects the business indicator and corporate governance variables, so its predictability reduces. In addition, different company stock prices are over-estimated or under-estimated on account of manmade manipulation. So the market value and fluctuation of the stock price cannot reflect the real value, which makes significance of the credit risk variable and poorer explanatory ability, but part of the companies have the enhanced effect. (2) No matter what kind of model is inside or out the sample, 0.5 is the best cut value of the performance comparison. The cut value of different classifications will influence the identification performance of the model, but it cannot be verified in practice that what kind of cut value is the best. Thus, the study compares three kinds of different cut values, separately 0.3, 0.5 and 0.8. The empirical result shows that no matter what kind of model is inside or outside the sample, 0.5 is the best cut value of the comparison performance. (3) Observing from the type I error, the Logit model which considers the macroeconomic and corporate governance variables is the lowest. However, in consideration of different cut values, 0.3 is better. The performance of the type II error is consistent with the comparison result of the accurate rate. The prediction outside the sample is found to be consistent with the comparison inside the sample. But when the cut value is 0.5, it is the best. (4) As a whole, the performance inside the sample is better than that outside the sample. The empirical result of the study shows that the performance outside the sample of the KMV is superior to that inside the sample. The reason may be that the estimative parameter is appropriate for the data estimation outside the sample. But the estimative parameter outside the sample of the Logit model is significantly lower than that inside the sample. The possible reason is subject to the difference of the credit risk involving extent of the implication of the enterprise per se; moreover, it causes the phenomenon of the judgment ability reduction of the testing sample that the model cannot learn to train the sample information.As for the research of an enterprise’s financial crisis, the relevant literature has accumulated many results, but it cashes in on three kinds of financial early warning model to make a prediction outside the sample: the best default coefficient calculated by the data in sample, the KMV model and the Logit model of whether to incorporate the macroeconomic variables and corporate governance variables. Finally, the model which makes a performance comparison between the data inside and out sample is seldom seen. The analysis of the study obtains that incorporating the macroeconomic variables and corporate governance variables can indeed enhance the financial crisis predictability of the model, which can also serve as a very important reference index in evaluating the credit risk of an enterprise; that is, it can effectively predict the financial crisis of an enterprise to provide an effective investment decision and risk management reference for an enterprise, a creditor, an investor and the government authority to further reduce the investment risk and loss.更多还原