無母數集群模式於企業信用評等之應用

Non-parametric Clustering for Corporate Credit Ratings

蘇美芳(Mei-Fang Su)

指導教授 ： 盧以詮

元智大學/資訊學院/資訊社會學研究所/碩士(1998年)

摘要

在預測債券等級的相關研究中，主要的預測目標是Moody’s 或是S&P’s 等債券評等機構所公佈的債券等級。但在台灣的金融市場中，中華信用評等公司並不公開非金融業的企業評等結果，因此在缺乏非金融業的公開評等的情況下，就難以利用各種需要目標值的方法論來進行評等。本研究使用統計k-mean集群分析來進行評等，並以我國經濟發展中的重要產業－電子業上市公司為研究對象，選擇研究樣本為八十四年三月至八十七年三月的季財務資料，其中八十四年三月至八十六年十二月為訓練資料，而八十七年三月為測試資料。我們先將電子業分割成六個子產業，這是因為各子產業有其獨特的產業特性；再依各子產業進行統計k-mean集群分析，而群集數目之選定乃依各子產業之SPR及R2來決定。初始集群結果我們初步以台灣經濟新報因素權重作為給定等級的範例，以提供專家作為賦予等級意義的參考；最後我們提出區別分析方法，將專家調整後的各集群資料形成模式，以作為未來預測信用等級之用。在測試樣本中，集群分析在區分最高信用等級及最低信用等級的樣本中有較好的效果，而在區分中間信用等級的效果較差。

關鍵字

信用評等 ； 集群分析 ； 區別分析

並列摘要

Most researches in bond ratings use historical ratings as learning targets. The targets are usually the public ratings such as the ones from the Moody’s or the Standard & Poor’s. In the absence of target ratings, the methodologies by supervised learning become impractical. We propose a statistical clustering model, known as the k-mean clustering, for classifying corporate credit ratings. We collect the sample data from all listed electronics industry companies in the Taiwan stock market. We use quarterly financial data between 1995 and 1998. Since the electronics companies have their individual characteristics, we divide the electronics industry into six sub-industries. The statistical k-mean clustering is then applied to each sub-industry. The number of clusters is evaluated by values of the SPR and the R2 within each sub-industries.　We then provide a set of factors’ weightings of Taiwan Economic Journal as a reference to help experts determine the ratings. Finally, we provide a multivariate discriminant analysis using the previous clustering results for future classifications. The results show that the discriminant analysis has a better accurate rate in discriminating companies with either the best or the worst credit ratings.

並列關鍵字

credit ratings ； cluster analysis ； discriminant analysis

參考文獻

2.George E. Pinches and Kent A. Mingo , “A Multivariate Analysis of Industrial Bond Ratings”, Journal of Finance, March 1973, pp.1-18.

3.Horrigan, J.O., “The Determinants of Long Term Credit Standing with Financial Ratios “, Empiricl Research in Accounting Research, 1966, pp.44-62.

4.Yi-Chuan Lu and Hilary Cheng, “Neural Clustering for Corporate Bond Classification”, working paper(submitted for publication), 1998.

6.Kaplan, Robert S., and Gabriel Urwitz, “Statistical Models of Bond Ratings: A Methodological Inquiry”, Journal of Business, Spring 1979, pp.231-261.

7.M. Porter, “Competitive Advantage:Creating and Sustaining Superior Performance”, New York, Free Press, 1985.