台灣公開發行公司一般產業財務危機預測

本論文根據台灣新報資料庫數千家公開發行企業的違約前兩年財務變數資料，根據2014年到2018年發生財務危機公司以公告日為準來估計邏輯迴歸模型。總計2014到2018年共73家危機公司發生違約，其中包括公司重整、紓困、淨值為負、繼續經營疑慮因素。本論文根據林妙宜(2002)論文中以獲利能力、現金流量、成長率、償債能力及經營能力五構面共63個比率，利用區別分析法選取有效變數，分別為常續性EPS、現金流量對總負債比率、盈餘成長率、負債比率及平均收帳天數共五項財務變數，並以其模型做為參考，經過修正後模型對台灣一般產業公開發行公司中73家危機公司，並配對73家產業相同、資產規模相同之正常公司以全體樣本運用邏輯斯迴歸分析各項財務變數顯著性並且對兩者模型的預測違約準確率做比較。研究結果如下: 一、林妙宜(2002)五構面財務模型運用邏輯斯迴歸分析之預測效果不佳。二、修正後財務模型運用邏輯斯迴歸分析預測效果優於林妙宜(2002)財務五構面模型。三、修正後財務變數模型顯著性優於林妙宜(2002)五構面財務模型。四、當時間離財務危機發生時間點越接近，預測效果越高。

關鍵字

信用風險；邏輯迴歸；違約預測

並列摘要

The paper estimates the logistic regression model based on the financial variables of the two years before the default of thousands of publicly issued companies in the Taiwan Economic Journal (TEJ) ,based on the financial crisis in 2014 to 2018. A total of 73 crisis companies defaulted from 2014 to 2018, including company restructuring, relief, negative net value, and suspicion of continuing operations. In this paper, according to Lin, MAIO-YI (2002) paper, with a total of 63 ratios of five aspects: profitability, cash flow, growth rate, debt-paying ability and operating ability. Use the discriminant analysis method to select the effective variables, which are five financial variables: perpetual EPS, cash flow to total debt ratio, surplus growth rate, debt ratio and average collection days, and use its model as a reference. After the model is revised, the model is used to analyze 73 crisis companies in Taiwan’s general industry public offering companies, and to pair 73 normal companies with the same industry and the same asset size to apply logistic regression analysis to the entire sample. And compare the significance of various financial variables and the predicted default rate of the two models. The research results are as follows: 1.Lin, MAIO-YI (2002) the model uses logistic regression analysis to predict poor results. 2.The revised financial model uses logistic regression analysis to predict the effect better than previous model. 3.The revised financial variable model is significantly better than previous model. 4.The closer time of default,the higher the prediction effect.