由於公司提列資產減損額度為設限資料(censoring data),過去的研究者均以Tobit模式分析,此模式受限於假設決策者在預估是否提列資產減損及減損額度的大小皆以相同形式的模式來考量;然不同決策間有其不同的決策模式,此無差異性的決策模式是否合於實際決策者或決策情境令人質疑。目前計對於設限資料的處理除Tobit模式外,尚有不同假設的模式,如two-part模式(分別處理選擇及及觀察到提列之額度模式)及sample selection模式(除分別考慮選擇模式與額度模式規格外,進一步考慮兩模式誤差間的相關性)。本研究分別應用上述三種模式,分析我國最近之資產減損提列的資料,並比較分析結果,以探究我國上市櫃公司管理當局提列資產減損的可能決策形式。 文中應用2005年至2008年上市(櫃)公司的資產減損額度資料,首先分析two-part模式及sample selection模式於此問題的適用性;再進一步與Tobit模式比較預測績效。實證結果顯示,two-part模式比sample selection模式較適合於目前資產減損額度的樣本;對於預測績效則以Tobit模式最差,two-part模式略優於sample selection模式;此結果可提供診斷資產減損分析者參考。而two-part模式較接近目前分析的樣本,顯示國內在決定提列資產減損與該提多少額度的考量上相關性很低。
Since the sample of assets write-off magnitude is censoring data type, most of previous researches used Tobit model for the analysis. The model is limited by the assumption that uses same specification for both assets write-off decision model and magnitude regression model. Nevertheless, it is in doubt that whether the indiscriminate decision mode will agree to the actual behavior of the decision makers or even the real scenarios. In addition to the Tobit model for the censoring data analysis, there are other models that relax single model specification limit, such as two-part model (considers selection model specification and observed magnitude model specification separately) and sample selection model (beside different model specifications, also considers correlation between errors distributions of the models). In this work, collected samples of listed companies in Taiwan from year 2005 to year 2008 are used to study the fitness of two-part model and sample selection model for assets write-off and compare the prediction performances of these models to the result of Tobit model. The empirical results demonstrate that two-part model is more suitable than sample selection model in the assets write-off sample. For the prediction performance, the result of Tobit model is the worst and the result of two-part model is slightly superior to the sample selection model.