風險、公司治理與危機預警模型之建構─台灣金控與非金控銀行之實證

Risk, Corporate Governance and the Construction of Early Warning System-The Empirical Study of Financial Holding and Non-Holding Company

周淑虹(Shu-Hung Chou)

指導教授 ： 劉定焜

朝陽科技大學/管理學院/財務金融系/碩士(2010年)

摘要

西元2008年，發生一場「金融大海嘯」，史上罕見的系統性危機，債券天王Bill Gross說「全球股市、債市、房市三種主要資產齊跌」。Bill Gross口中的金融大海嘯正一波波的襲來，第一波的浪頭，吞噬了全美第五大投資銀行貝爾斯登 (Bear Stearns)。第二波，讓負債高達五兆美元的美國前兩大房貸公司 (房地美、房利美) 被接管。九月十五日，第三波襲來，全美第四大投資銀行雷曼兄弟也被捲入海底，宣布倒閉。這波金融大海嘯不只影響了千千萬萬的投資人，也影響了全球的經濟體系。就在世人慶幸走出2008年的金融海嘯當下，杜拜崩盤，冰島破產，杜拜國家總借款額度高達六百億美金，而且極為可能引發二次金融海嘯。很少人認為現在會發生「第二次亞洲金融風暴」，但也沒人知道會不會有亞洲國家在全球金融風暴中倒下，南韓看來是其中最脆弱的，南韓會成為亞洲的冰島嗎？這些皆因銀行內部控制不佳及經營品質不良而產生銀行危機，嚴重時更可能引發連鎖效應造成社會動盪不安，所以建構一健全的金融預警模型實刻不容緩。
因此，本文建構2001年第一季至2007年第三季，共27季12家金控銀行之panel data以及1997年第一季至2008年第四季，共48季之10家非金控銀行panel data，藉以探討此兩子樣本之差異。除財務變數、公司治理變數與總體經濟變數外，並納入過去文獻鮮少探討之風險變數之考量，期望能增加危機預警模型之預測能力。首先，本文分別利用因素分析法和集群分析法評估銀行經營績效，將銀行區分為3類別 (績優、普通與預警3類)，進一步運用區別分析法和Logistic模型建構金融危機預警機制，藉此推估不同分類及不同預警模型，何者具有較佳之預測能力。期望對未來後續研究者與相關單位提出具價值之參考與建議。
實證結果發現金控銀行在區別分析法下，兩兩分群之因素分析與集群分析之平均預測率分別為88.3%與83.7%，在Logistic方法下，平均預測率分別為89.4%與83.3%。因素分析與集群分析在區別分析下預測率之結果，整體來說，因素分析優於集群分析。在Logistic預測率下之結果，整體來說，因素分析優於集群分析。另外，在因素分析下區別分析與Logistic之預測率，無論是A、B，B、C或是C、A分群都是Logistic方法較好。但在集群分析下的區別分析與Logistic之預測率，則是區別分析優於Logistic。金控銀行因素分析之Logistic實證模型估計結果中，流動比率有顯著正向影響、董監事持股比率有顯著負向影響與經理人持股比率有顯著正向影響及台灣實質GDP年增率有顯著負向影響。金控銀行集群分析之Logistic實證模型估計結果中備抵放款損失率有顯著影響。
非金控銀行在區別分析法下，兩兩分群之因素分析與集群分析之平均預測率分別為91.0%與90.0%，在Logistic方法下，平均預測率分別為92.8%與93.4%。因素分析與集群分析在區別分析下預測率之結果，整體來說，因素分析優於集群分析。在Logistic預測率下的結果，整體來說，因素分析優於集群分析。另外，在因素分析與集群分析下區別分析與Logistic之預測率，無論A、B，B、C或是C、A分群都是Logistic方法較好。非金控銀行因素分析之Logistic實證模型估計結果中，利息未收現比率有顯著負向影響、營業費用率有顯著負向影響和流動比率有顯著正向影響。非金控銀行集群分析之Logistic實證模型估計結果中，經理人持股比率有顯著影響與盈餘波動風險有顯著負向影響。

關鍵字

金融預警模型 ； 因素分析 ； 區別分析 ； Logistic模型 ； 集群分析

並列摘要

There is a financial crisis happened in 2008, and is a rare systematic crisis. Bill Gross said that the three main kinds of assets are all fall. The first wave swallows the nation''s No. 5 bank, Bear Stearns of America. The second wave let the first two mortgages company been taken, that are in debt of 5 trillion. The third wave causes the Lehman Brothers Holding Inc. has to declare their bankruptcy. The financial tsunami is not merely affecting the millionaire of investors but also having influence on economy system all over the world. Under the wave of financial tsunami, Dubai disintegrates their economy, Iceland is bankrupt, and there will be a second financial tsunami. Those crises are due to unhealthy inner-control and management. When they become more serious they would cause chain-effects and unstable of our society. Therefore construct a robust financial warning system model would be an urgent matter.
Therefore, this study aims to build and construct the panel data from the first season in 2001 to the third season in 2007, the total of 27 season 12 financial holding companies. As well as the panel data from the first season in 1997 to the fourth season in 2008, the total of 48 season 10 the non-financial holding companies, so as to discuss difference of the two sub-samples. Besides the financial variables, contemporaneous governance variables and the microeconomic variables, this study takes into account the risk variable that rare discuss in previous literatures. We hope this can increase the forecasting ability of our early warning model. First, this study uses the factor and cluster analysis to evaluate the performance of bank, which is separated into 3 categories (excellent, ordinary and warning). Further by using the multinomial Logistic model and multinomial panel Logistic model we construct the early warning model of finance, and conjecture that under the different classification and model which one has the batter forecasting ability. We expect that the results would be valuable for the future study and policy decision makers.
Empirical results show that with respect to financial holding companies, under the discriminant analysis we find that the results of two separate samples with factor and cluster analysis have the average forecasting rate of 88.3% and 83.7%. Under the Logistic model we find that the average forecasting rate is 89.4% and 83.3%. As a whole, under the discriminant analysis, factor analysis is superior to cluster analysis. Under the Logistic model, factor analysis is superior to cluster analysis.
Empirical results show that with respect to non-financial holding companies, under the discriminant analysis we find that the results of two separate samples with factor and cluster analysis have the average forecasting rate of 91.0% and 90.0%. Under the Logistic model we find that the average forecasting rate is 92.8% and 93.4%. As a whole, under the discriminant analysis, factor analysis is superior to cluster analysis. Under the Logistic model, factor analysis is superior to cluster analysis.

並列關鍵字

factor analysis ； discriminant analysis ； financial warning system ； cluster analysis ； Logistic model

參考文獻

許溪南、歐陽豪與陳慶芳 (2007)。公司治理、盈餘管理與財務預警模型之建構。會計與公司治理。4 (1)，85-121。

陳弘吉、邱國欽與李羽涵 (2009)。金融預警系統之建立─以台灣商業銀行為例，朝陽學報。14，441-457。

張瑞元、林金賢 (2005)。建構銀行危機預警模型─訊號法與Panel Logit之結合。會計與公司治理。2 (1)，9-32。

劉景中 (2009)。銀行集中度、銀行市場競爭度與銀行風險─台灣實證研究。中央研究院經濟研究所經濟論文。37 (1)，101-135。

林金賢、張瑞源與王盈傑 (2006)。銀行危機可以被有效預測嗎。管理與系統。13 (2)，121-152。