因應實質盈餘管理及裁決性應計數之信用風險評估模型調整

Abstract

傳統之風險評估模型其基礎建立在財務報表相關資訊上，但在以其評估企業信用風險時，或宜注意財報相關資訊是否有被操弄，此可能導致違約風險之錯估。 本研究就此議題研究實質盈餘管理及裁決性應計數，以中國上市公司為樣本估計其相關數值，再依本研究所訂定之標準轉為虛擬變數，復將此以線性之方式加入風險評估模型中，分別就學習期及預測期看其是否能提供邊際解釋力與預測力；此外，基於債務協商、債務重整…等事件不算在既有違約樣本中，但有違約之實質，為避免回歸式之估計產生偏誤，本研究也納入隱性違約之定義，並就一樣本中納入隱性與真實違約樣本，及二樣本中僅納入真實違約樣本分別看其成效。 實證結果發現，在學習期時，加入實質盈餘管理及裁決性應計數無論對真實違約或是隱性與真實違約樣本，皆有助於提升模型之調整判定係數，即其對模型具有邊際解釋力；在預測期時，就僅納入真實違約樣本加入實質盈餘管理及裁決性應計數之虛擬變數，顯著提升預測力；在對納入隱性與真實違約樣本下，其所能產生額外之效益則有限。

Traditional credit risk rating models based on financial ratios are prevalent; nevertheless, problems with credit scores might emerge from some earnings manipulaitons. This study aims to explore the influences of real and accrual-based earnings management schemes on the effectiveness of accounting credit risk models, using data from publicly listed companies in Mainland China. Specifically, this study adopts as dummy variables that reflect the significance of real and accrual-based earnings manipulation input factors to the credit risks rating model, aiming to explore the extent to which such modification adds to in-sample fitting (learning) period and out-of-sample (forecasting) period explanatory power of the forecast models. Moreover, this study identifies three types of firms including firms that legally defaulted during the sample period (hereafter the actual default firms), the firms that encountered substantial debt negotiations or reconstructions but did not legally defaulted during the sample period (hereafter the stealth default firms), and the firms that had not been subject to either of the defaults during the same period (hereafter the continuing firms). Thereby it conducts tests with (1) the actual default firms serving as the experimental group and (2) both actual default firms and stealth default firms serving as the experimental group. The empirical results shows: during the learning period, adding real earnings management and discretionary accrual to the models enhance the explanatory power, regardless of explaining actual or stealth defaults. As for the out-of-sample (forecasting) tests, including real earnings management and accrual-based earnings management measures in the predicting models, the forecast accuracy appears to increase significantly. However, the effect of adding the earnings manipulation measures on stealth plus-actual default sample are insignificant.