財務危機預警模型之比較

Abstract

本研究針對台灣上市上櫃公司的資料，以財務會計變數與市場變數的組合，建立Logit模型、MDA模型與離散時間危險模型等財務危機預警模型，藉以觀察加入市場變數是否可增加模型的區別能力與預測能力，並比較三種統計模型的預測準確度。本研究將變數的組合歸為四類，依序為財務會計變數組合、財務會計變數加市場變數組合、市場變數組合以及Shumway變數組合。衡量模型預測準確度的方法有違約機率分配表、錯誤分類表、ROC曲線與AUC值以及EMC值的分析。 實證結果發現，樣本內資料以Logit模型使用財務會計變數加市場變數組合的區別能力最佳；樣本外資料的預測能力則是以財務會計變數加市場變數組合與Shumway變數組合較佳，但三種統計模型的預測能力並沒有顯著的差別。綜言之，加入市場變數確實可提升模型對樣本內資料的區別能力，但對樣本外資料的預測能力則沒有顯著提升。此外，若要準確判斷樣本外違約公司的違約傾向，交替使用財務危機預警模型不失為一個良好的方法。

Based on the data of Taiwan corporations trading in TSE and OTC, this study used financial accounting variables and market variables to construct financial distress prediction models, such as Logit model, MDA model and discrete-time hazard model. With such methodology, I examined whether the added-in market variables could enhance the model’s discrimination ability and predicting capability or not, furthermore, I compared the accuracy of three statistical models. This study classified the variables into four categories, which are financial accounting variable group, financial accounting variable plus market variable group, market variable group and Shumway’s variable group, separately. The methods used in analyzing the models’ prediction accuracy are the default probability table, misclassification table, ROC curve and AUC, and EMC. The empirical results showed that the best model to discriminate in-sample data is Logit model composed of financial accounting variable plus market variable group; however, the best model to predict out-sample data is composed by financial accounting variable plus market variable group and Shumway’s variable group, but there are no difference between three statistical models in predicting capabilities. In summary, adding market variables does really enhance discrimination ability of in-sample data, but it doesn’t obviously enhance the prediction ability of out-sample data. Moreover, it is better to use financial distress prediction models alternatively in judging the tendency of the out-sample default firms.