樣本選擇偏誤於企業財務危機預警模型之研究：以台灣上市公司為例

傳統財務危機預警模型建立，皆是以已通過審核申請者樣本建立模型，忽略未通過審核申請者樣本，然而以這些樣本所建構出來的模型，因爲不能反映母體的變動程度與變數間的相互影響效果，亦未考慮樣本選擇偏誤問題，故會影響模型的配適度與預測能力。本文則是加入拒絕推論技術建立修正後Heckman兩階段樣本選擇模型，以台灣上市公司爲例，與傳統財務危機預警模型進行比較；研究結果發現，財務危機模型建構中的審核模型與違約模型兩階段間存在顯著之相關，若不採取樣本選擇模型，將對模型預測結果產生很大的偏誤；而觀察模型的配適度與預測能力後亦可發現，修正後Heckman兩階段樣本選擇模型的配適度與預測能力確實優於傳統的財務危機預警模型。

關鍵字

樣本選擇；拒絕推論；財務危機預警模型； Probit模型

並列摘要

Traditionally, most scholars use the sample of accepted applicants in building financial distress prediction models and neglect the sample of rejected applicants. These models cannot reflect the variation of population and the interactive effects among all variables, and do not consider the problem of sample selection bias. Therefore, the fitness and prediction ability of the models would be affected. In this paper, the reject inference technology is considered in the financial distress prediction as building the modified Heckman two-stage sample selection model. Using the modified model and Taiwan's listed companies as examples, we could find that the application and the default stages are highly correlated in distress prediction. In other words, if the modified model is not used, the sample selection bias would result. After observing the fitness and prediction ability of our modified Heckman two-stage sample selection model relative to traditional financial distress prediction models, we discover that the prediction performance of the former is superior.