銀行放款風險雙指標預警模型－二次區別分析之運用

The Study of A Early Warning Model of Dual Index on Bank Loan Failure－Application of The Quadratic Discriminant Analysis

李松儐(Song-Ping, Lee)

指導教授 ： 古永嘉

國立臺北大學/商學院/國際財務金融碩士在職專班/碩士(2009年)

摘要

我國自開放新銀行成立，因競爭而造成授信品質低落，造成大量逾放，加上國內上市櫃公司爆發層出不窮的財務危機，使得金融業受傷甚重，甚至引發本土性金融風暴。因此金融業有必要建立一套有效的財務預警制度，以便及早對問題的授信企業有所因應。
本研究選取70家上市櫃之電子公司210筆資料為樣本，採財務比率變數及公司治理變數雙指標，以危機發生時點為基準，往前選擇三年資料進行分析，並嘗試利用Logit迴歸分析與區別分析模型來建構財務危機的預警模型。研究方法先以t-test平均數來檢測財務正常公司及危機公司資料是否呈現顯著性差異；再以逐步Logit分析及逐步區別分析判定預測企業財務危機的重要變數及重要性排序；接著分別以Logit迴歸分析、一次區別分析及二次區別分析做全體樣本的分析，並以命中率來作為驗證此模式的預測力；最後再以Logit迴歸分析及二次區別分析做分年度的命中率分析。經實證結果，本研究結論有三：
一、以t-test進行平均數的檢定，在財務指標中除財務危機公司的負債比率明顯較高外，其餘比率皆是財務正常公司比財務危機公司來得高。至於公司治理指標之董監持股及大股東持股比率危機公司高於正常公司。
二、根據逐步Logit分析及逐步區別分析篩選出前四項共同的變數依序為：每股盈餘、速動比率、董監事持股比率及現金流量比率。
三、以全體樣本來檢定，Logit分析、一次區別分析、二次區別分析，顯示二次區別分析預測力較佳達91.43%；分年度的檢定，Logit分析前三年命中率為78.57%、82.86%、85.71%，二次區別分析則為97.14%、98.57%、98.57%，也是二次區別分析預測力較佳。以二次區別分析達98.57%的高預測力，適合做為銀行放款風險預警模型。

關鍵字

財務比率、公司治理、Logit分析、一次區別分析、二次區別分析

並列摘要

Since the financial system were opened to establish new banks, Taiwan’s banks have faced fierce competition . Finance’s crisis of some enterprises inflicted heavy loss on banks even caused the local financial storm. Therefore, our financial institution should set up a effective system of financial examination and prediction to avoid bankruptcy of bank and react to the problem early.

This study collects the financial data of the 70 listed electronic companies ,210 terms including the samples of crisis state and normal state. In constructing the model, there are two main categories indicator, including financial ratio category and corporate governance category. Logit Regression model and Discriminated Analysis model are used in this study to establish precautionary finance-warning model . The methodology, first used average value of t-test to test whether crisis and normal companies are different. Then, used stepwise regression and stepwise discriminated analysis to find the important variables . Next, use the Logit Analysis, Discriminated Analysis, Quadratic Discriminated Analysis to make an analysis of all the samples. Furthermore we used the hit-ratio to verify the explanatory power of this model. Finally, we used the Logit Analysis, Quadratic Discriminated Analysis to get hit-ratio and have achieved three conclusions：

1. The results of t-test indicate that financial healthy company is higher than distress one for all ratio except debt ratio in financial ratio category . In corporate governance category the ratio of holding shares for directors , financial distress is higher than financial healthy company.

2. The empirical results indicate that the main influence factors extracted in financial index are ‘Earnings Per Share (EPS) ’”‘Equity Quick Ratio’, ‘”Directors who hold share”’, ‘Cash flow Ratio”.

3. The empirical results indicate that : (1) Test for the whole samples, the Logit Analysis, the Discriminated Analysis and Quadratic Discriminated Analysis got an accuracy is 86.67%、82.86%、91.43% , it shows Quadratic Discriminated Analysis has a better forecast ability. （2） Test by year, the Logit Analysis achieves an accuracy is 78.57%、82.86%、85.71% , and the Quadratic Discriminated Analysis achieves an accuracy is 97.14%、98.57%、98.57% . The forecast ability of Quadratic Discriminated Analysis model is suitable for bank loan-application model .

並列關鍵字

Financial ratio, Corporate governance, Logit Analysis, Discriminated Analysis, Quadratic Discriminated Analysis

參考文獻

4.李建然、陳政芳、李啟華，董監事持股集中度與會計師獨立性—對會計師出具繼續經營疑慮查核意見之影響 ，當代會計，第4卷第2期，民國92年。

1. Altman,E.I. " Financial Ratios, Discriminant Analysis, and the Prediction of Corporate Bankruptcy," Journal of Finance 22, Sep.1968, pp.589-610.

2. Beaver, W.H., "Financial Ratios as Predictors of Failure," Journal of Accounting Research 4 , Supplement 1966, pp.71-102.

3. Blum, M., " Failing Company Discriminant Analysis," Journal of Accounting Research 12, Spring 1974, pp.1-25.

4. Coats, P. and Fant F., " Recognizing Financial Distress Patterns Using a Neural Network Tool," Financial Management 22, Autumn 1993, pp.142-165.