本研究應用文詞語意探勘技術,剖析公開傳媒報導中與企業經營訊息相關之資訊內涵,建立公開訊息反映企業危機事件之量化指標,並將研究結果與臺灣經濟新報資料庫(TEJ)建置公告之臺灣企業信用風險指標(Taiwan Corporate Credit Risk Index, TCRI)進行效能驗證與比較。過去財務危機預測之相關研究,大多採用財務報表等量化資訊進行分析,然而企業危機事件之徵兆往往隱藏在事件發生前之資訊中,緣此,本研究建立公開資訊之代理變數,進行企業危機預警之研究。公開資訊代理變數方面,本研究應用文詞語意探勘技術標記新聞語料,參考並延伸Demers and Vega(2011),以危機/非危機特徵詞為基礎構建危機事件發生強度之量化指標(Intensity of Default-Corpus, ITDC)。實證結果顯示距離危機發生時點愈接近之模型,其配適效果與預測能力均愈準確;本研究亦使用兩種不同的財務危機機率門檻進行判定與比較,Martin(1977)建議之實證機率門檻值,在模型的分類正確率上並未優於機率門檻值為0.5的模型,但卻可明顯降低型一誤差,達到更有效的分類。納入公開新聞資訊內涵之量化指標,有助於提升企業財務危機預警模型之預測效能並可提升預測之分類正確率。
This paper applies the technique of linguistic text mining to extract relevant information from Chinese financial news and further investigate its application of predicting probability of default. Our study examine whether the News-corpus can improve the effectiveness of the prediction of credit rating on the basis of Taiwan Corporate Credit Risk Index (TCRI), which is announced by the Taiwan Economic Journal (TEJ).The News-corpus variable of the intensity of default-corpus (ITDC) is constructed by referring to Demers and Vega (2011) and further being investigated whether news plays a relevant role. The empirical results show that the explanation power of alternative financial distress model is improved while the ITDC is incorporated. The one-quarter ahead forecast presents that the type I error is reduced and the identification accuracy is increased in the Logit Regression while the relevant news information is considered in the prediction of probability of default. Our results prove that intensity of default-corpus variable constructed from linguistic text mining of Chinese financial news does improve the effectiveness of the prediction of corporate default probability.