資料採礦在財務危機預警模式之應用

Application of Data Mining on Financial Distress Models

李建成(Carlos Lee) ； 陳怡君(I-Chung Chen)

《Journal of Data Analysis》 1卷2期 (2006/04) Pp. 85-108

https://doi.org/10.6338/JDA.200604_1(2).0005

摘要

當企業發生財務危機或是經營不善時，對於企業本身或是政府單位、投資者及銀行都會造成嚴重的衝擊。因此，若能在事先發現危機的徵兆，建立一套預警模型來預測危機的發生，將可以幫助企業、投資者、銀行或是政府單位降低損失。本研究以資料採礦流程中CRISP-DM的角度出發，了解模型的建構過程，同時，蒐集台灣上市公司之財務性資料以及非財務性資料，例如：會計資訊、股權結構以及董事會結構等變數，藉由結合傳統統計技術以及資料採礦技術，利用羅吉斯迴歸以及類神經網路，分別建構危機前一年、危機前二年以及危機前三年的預警模型，並比較區別能力，找出適當的財務預警模型以及影響財務危機發生的變數，最後，並透過C5.0了解顯著變數間之互動性。研究結果顯示，類神經網路的整體正確率均較羅吉斯迴歸模型佳。此外，利用羅吉斯迴歸所建構之正確率最高的模型發現，其顯著變數分別為：「稅後淨利率」、「財務槓桿度」、「現金流量比率」、「董監質押比率」、「流動比率」。由此可知，影響危機事件的發生，除了獲利能力外，流動性、槓桿度以及董監質押狀況也是主要的因素之一。另外，將這5個變數，利用C5.0，共產生5條規則，從中可以了解變數之間的互動關係，將可以根據這些規則，幫助政府單位，或是銀行業甚至是投資大眾更加確認危機公司的特性，將有助於降低危機事件的發生。

關鍵字

財務危機 ； 資料採礦 ； 羅吉斯迴歸 ； 類神經網路

並列摘要

The main purpose of this research is to apply crisp-dm and data mining techniques, as well as traditional statistical methods to develop financial distress models. Crisp-dm is a popular method of data mining process. We collected financial and non-financial data, for example, accounting information, ownership structure, and board structure, from some of the publicly traded companies in Taiwan. We then used two kinds of method, logistic models and neural networks, to analyze those data. For each method, three different financial distress models were built, ranging from one year to three years prior to the financial distress. Correct rates were compared with all of the models. We try to form the best logistic model to find the significant variables that affect financial distress happened. Then, we use C5.0 to find the interactive of the significant variables.The results we found based on our research are summarized as following: 1. The total correct rates of the Neural Networks are better than logistic models. 2. From the best logistic model, we found that ”net income margin”, ”degree of financial leverage”, ” cash flow ratio”, ”pledged shares ratio of directors' and supervisors' ”, and ”current ratio” are the significant variables. Therefore, besides profitability we understand that liquidity, leverage, and pledge are important causes. 3. We use C5.0 to have 5 rules. It will understand interactive of the significant variables and help government, banks, and investors to assure the characters of financial distress then reduce the possibility of loss.

並列關鍵字

Financial Distress ； Data Mining ； Cross Industry Standard Process for Data Mining ； CRISP-DM ； Logistic Regression ； Neural Networks ； C5.0

參考文獻

Cabena, P.,Hadjinian, P.,Stadler, R.,Verhees, J.,Zanasi, A.(1998).Discovering data mining from concept to implementation.Upper Saddle River, NJ.:Prentice Hall.

Berry, M. J. A.,Linoff, G. S.(1997).Data Mining Techniques: for marketing, sales, and customer support.New York:John Wiley & Sons.

Zhang, G.,Patuwo, B. E.,Hu, M. Y.(1998).Forecasting with artificial neural networks: The state of the art.International Journal of Forecasting.14,35-62.

沈大白、張大成、劉宛鑫(2002)。運用類神經網路建構財務危機預警模式。貨幣觀測與信用評等。38，95-101。

俞慧華(2001)。改良式類神經網路模式於信用卡顧客關係管理之研究(碩士論文)。台北科技大學商業自動化與管理研究所。