KMV法適合估計台灣上市櫃公司的信用風險嗎?

Is KMV Model Suited to Estimate the Credit Risk of Listed Companies in Taiwan?

許燕玲(Yen-Ling Hsu)

指導教授 ： 巫春洲 ； 江長周

中原大學/商學院/國際貿易研究所/碩士(2007年)

https://doi.org/10.6840/cycu200700208

摘要

本文主要探討KMV法於估計台灣上市櫃公司之信用風險的適用性。由於Duan，Gauthier和Simonato (2004) 證明了在Merton (1974) 選擇權評價模型的架構下，由KMV法求出之估計值，等同於以Duan (1994, 2000)提出的資料轉換法得出之最大概似估計值。因此，本文分別以KMV模型與Duan (1994, 2000) MLE模型，進行Merton (1974) 模型參數值與違約距離 (distance-to-default) 的估算，並比較兩模型於公司信用風險的預測準確率。此外，陳業寧等人 (2004) 發現Z評分 (Z-score) 模型於台灣企業發生財務危機的預測較Merton (1974) 模型為準確。故本文進一步自KMV與MLE模型中，選出準確率較高者，和採用與Altman，Hartzell，和Peck (1995)提出的修正後Z評分模型相同變數之區別分析 (discriminant analysis)，進行比較分析。由Logit 迴歸分析法 (Bongini et al., 2002) 及檢定力曲線(Kealhofer和Kurbat, 2001) 可發現，針對取樣自1998年至2004年間，於台灣證券交易所及中華民國證券櫃檯買賣中心掛牌的216家公司，三模型於公司發生財務危機之預測準確率，均以區別分析為最高，Duan (1994, 2000) MLE模型為次之，KMV模型則為最差。由於區別函數之建構乃採用前述216家公司之樣本資料而得，若就此分析結果逕行判定區別分析對於公司信用風險的估計較另兩模型準確，似乎過於草率，故本文另選取60家於2005間掛牌的樣本外公司，進行更為客觀的樣本外分析。實証結果顯示，不論是樣本內或樣本外分析，公司信用風險的估計，均以區別分析的準確率最高，Duan (1994, 2000) MLE模型次之，KMV模型最差。

關鍵字

信用風險 ； 區別分析 ； KMV ； MLE

並列摘要

Duan, Gauthier, and Simonato (2004) demonstrate that, in the framework of the option pricing model of Merton (1974), the KMV estimates are identical to maximum likelihood estimates proposed by Duan (1994, 2000). Therefore, this paper utilizes the model of KMV and Duan (1994, 2000) MLE to estimate the parameters of Merton (1974) model and the distance-to-default, them compares their abilities in predicting corporate financial distress. Chen et al. (2004) obtain the result that the accuracy of the forecast of insolvency in Taiwan from 1998 to 2001, the Z-score model of Altman (1968) is significantly better than Merton (1974) model. Consequently, this paper chooses the more accurate model from the KMV and MLE model to compare with the discriminant analysis of using the same variables in the revised Z-score model proposed by Altman, Hartzell, and Peck (1995). No matter the analysis of the Logit regression (Bongini et al., 2002) or the power curve (Kealhofer and Kurbat, 2001), using the data of Taiwan's listed companies from 1998 to 2004, the results both show that the most accuracy of predicting corporate financial distress in Taiwan is the discriminant analysis, next is Duan’s (1994, 2000) MLE, and the KMV model is the last. Owing to the discriminant function is constructed from the preceding sample, this paper chooses extra 60 firms listed in Taiwan during 2005 to analyze the accuracy of estimating corporate credit risk. The empirical results of in-sample and out-of-sample all reach the same conclusion that, the most accuracy of the forecast of insolvency in Taiwan is the discriminant analysis, next is Duan (1994, 2000) MLE, and the KMV model is the last.

並列關鍵字

Credit risk ； MLE ； KMV ； Discriminant analysis

參考文獻

2.沈中華、林公韻 (2005)，違約機率預測與極端值-Robust Logistic Regression，金融改革與發展研討會論文。

3.陳惠玲、黃政民（1995），「財務報表分析與企業信用評等」，台灣經濟新報文化事業股份有限公司。

6.陳業寧、王衍智、許鴻英 (2004)，台灣企業財務危機之預測：信用評分法與選擇評價法孰優？，風險管理學報，第六卷第二期，頁155-179。

8.Anderson, R., S. Sunderesan, and P. Tychon, (1996), “Strategic Analysis of Contingent Claims,” European Economic Review, 12, 871-881.

9.Altman, E., (1968), “Financial Ratios, Discriminant Analysis and the Prediction of Corporate Bankruptcy,” Journal of Finance, 23, 589-609.