公司信用評等與違約機率之不同預測方式

Abstract

本研究將Kamstra 與 Kennedy (1998)所提出的合併預測模型之觀念運用在估計公司信用評等，並將所得之結果與傳統序列Logit與Probit分析模型比較。在樣本選取方面，我們以台灣地區上市上櫃公司為分析對象，樣本期間自2000年至2004為樣本內資料，用以預測2005年樣本外公司之信用評等。此外，參考國內外著名外部信評機構之評分原則，認為在衡量信用評等時必須同時考慮產業風險與企業個別風險。職是之故，將樣本公司分為傳統業、製造業與電子業分別預測，由於金融服務業的財務報表有別於其他產業，乃將之排除不用。分析變數考量財務比率、市場變數與總體經濟之不同構面，初步涵蓋62個變數，並以統計處理與模型精簡化原則控制每種模型所含之自變數。在違約機率預測方面，我們利用歷史資訊計算各信用評等之破產機率，取其平均值用以計算通式。實證結果發現，在三種產業之中，隨著信用狀況之惡化，評等與違約機率呈現出指數型態的關係式，此結果也與國外信評機構所作之分析報告相符合。

The aim of this study is to introduce an alternative method for the prediction of credit rating and default probability. Following the concept of combining forecast proposed by Kamstra and Kennedy (1998), the method of combining forecasting technique is used in the prediction of credit ratings rather than individual forecast. In addition, we further calibrate the ratings to default probability. The data is collected from the Taiwan Economic Journal database, covering firms listed in TSE, OTC, and emerging market. The sample period here are quarterly recorded in quarter one 2000 through quarter three 2005. The data recorded in 2000 through 2004 are assigned to build the forecasting model, while the data in 2005 are applied to testify the model. Moreover, considering the influence of industry and firm specific factors, the data is divided into three industry categories. The empirical results suggest that the predictive power of combining forecasting method is indeed better than that of the individual forecast. The other important task is to calibrate the ratings to default probability. We find empirical support that the default probability increases exponentially with decreasing creditworthiness.