台灣地區企業信用評分系統的建置、驗證和比較

Abstract

本文根據台灣新報資料庫所載數千家公開發行企業的三年財務報表資料，估計以違約與否為應變量、財務比例為解釋變量的Logit迴歸模型，並根據估計結果建置適用於企業信貸的信用評分卡。本文除詳細描述信用評分卡的建置過程外，特別強調信用評分卡雖可視為一個預測虛擬變量的計量經濟模型，但卻又和典型的實證經濟研究有很大的不同，這是因為不論在估計Logit模型之前還是在獲得迴歸係數之後，都必須採行諸多繁複的操作和分析，尤其重要的是建置信用評分卡前一定要保留部分樣本以驗證信用評分卡的預測績效。本文根據十多項驗證統計量統建模樣本之後的兩年資料來評估信用評分卡的預測績效，並證明我們所建置的信用評分卡在所有的績效指標上都優於台灣經濟新報所發佈的法本評等。在2006年底實施「新巴賽爾資本協定」的前夕，不論是對即將採行內部評等法以衡量信用風險的銀行，還是對政府金融監理機關，本文都有很適時的實用價值。白經濟學理來說，信用評分模型的發展有助於解決借貸市場中先天存在之訊息不對稱所造成的逆選擇和道德危機問題，從而增加借貸市場供給和需求兩方面的效率，更進而影響貨幣政策傳導機制，深具政策意義。

Using financial report data of thousands of public-issued companies over three years drawn from the TEJ (Taiwan Economic Journal) database, we develop credit scorecards for business landing based on the estimates of a Logit regression model. This paper emphasizes that the in-depth analysis and transformation of the original data before the estimation as well as testing and validation of the estimated model are much more important in developing credit scorecards than in a typical empirical econometric analysis. In particular, reserving a large percentage of data for the validation purpose is absolutely necessary when developing credit scorecards. Using two additional years of data as the validation sample, we show that our credit scorecards perform much better than the basic credit rating provided by the TEJ in terms of several performance statistics. Before implementing the new Basel Capital Accord at the end of 2006, this paper presents a timely contribution to both the banks which intend to adopt internal rating based approach and the governmental agencies which regulate the banking industry in managing credit risks.