信用風險模型探討- 實質盈餘管理及裁決性應計數之複合加權模型

Abstract

過往針對實質盈餘管理及裁決性應計數之研究，以及會計違約預警模型(Altman Z-Score Discriminant Analysis Model)與市場違約預警模型(KMV Model之研究，多為獨立之研究，並未見到有將其複合使用以預測違約之綜合研究。 本研究以中國大陸上市公司作為研究樣本，由Z-Score模型開始，並加入實質盈餘管理及裁決性應計數以檢測其有效性，並再將其顯著之結果與KMV模型結合，得到複合模型，並透過實證結果驗證其模型之違約預測能力優於會計違約預警模型。 實證結果發現，傳統KMV模型對於企業是否發生違約之預測效果不彰；但因企業在發生違約前或有盈餘操縱行為，利用複合模型或可獲致優於Z-Score模型之違約預測效益。首先，本研究將實質盈餘管理及裁決性應計數以虛擬變數之形式加入模型，以實質盈餘管理中的異常生產成本效果最佳，顯示公司在違約之前，生產成本已異於正常營業收入之比例。而利用此異常生產成本加入KMV模型所建立之複合模型，異常生產成本在前10%之群組，其預測效果雖未顯著優於Z-Score模型，但仍然維持與Z-Score模型相當之預測水準；而在異常生產成本前30%之群組，其預測效果則優於Z-Score模型。整體而言，複合模型之預測能力較Z-Score模型為佳，對於違約之預測有實質之幫助。

This study establishes a hybrid credit risk model that encompasses real earnings management and discretionary accruals. Its findings may help extend prior studies, which most typically perceive real earnings management, discretionary accruals, Altman Z-Score discriminant analysis model and KMV model as being independent. Despite that firms under financial distress may engage themselves in earnings management and may thus report overstated accounting measures adopted by the Z-Score model, there is few, if any, extant paper systematically adjusting for such potential bias by using both Z-Score and KMV predictors with weights on these two predictors varying with the measures that reflct the significance of earnings management. Based on Mainland China listed companies data sample, in the first step, this study encompasses real earnings management and discretionary accruals into Z-Score model, and test for its significances. According to the significant level observed in the first step, this study incorporates KMV model into the original accounting-based model and reaches a hybrid model. The empirical result shows that (1) the conventional KMV model fails to identify the default firms well beforehand, and (2) incorporating Z-Score model and KMV model into hybrid model helps enhance the default predictability. Specifically, including abnormal production cost as a variable in the prediction model helps increase the explanatory power more than other real earnings management and discretionary accruals variables. Based on this empirical evidence, this study includes abnormal production cost, Z-Score and KMV variables into a hybrid model. This study documents that given a dummy variable for the extent of earnings manipulation with a cut-off of top 10% abnormal production cost, the estimation performance of the hybrid model is similar to that of the Z-Score model. However, if the criteria conditions on a 30% cut-off for abnormal production cost, the result shows a better prediction via the hybrid model compared to Z-Score model estimation.