本文透過銀行經營原則及國內外相關文獻，客觀選取適合評估我國銀行經營績效之財務變數，建立實證模型，由銀行發生財務惡化之機率為基礎，進行台灣地區銀行經營績效之研究。其次，藉由計量統計方法，萃取數個財務績效指標，建立一套金融預警系統，並搭配其他統計方法加以驗證所求得財務指標之正確性後，以發揮「事前」防弊的功能，使其能及早發現並導正績效較差的銀行。 使用敘述統計與逐步LOGISTIC迴歸模型進行台灣地區銀行經營績效評估及風險指標建立相關研究，由概似比統計量做適合度檢定，結果模型顯著，故其參數估計不全為0。在模型準確效力方面，本模型之和諧率（Percent Concordant）為96.9%，且Somers’D指標均高達96.1%及Gamma 和諧指標高達98.3%，足以說明本研究建立之模型是相當優秀的模型。另將全體樣本依違約銀行與正常銀行分為兩組樣本，並以Kolmogorov-Smirnov檢定法(簡稱KS檢定法)檢定該兩組樣本之違約機率分配，於顯著水準0.05下，該兩組樣本之違約機率有顯著差異。足見本模型具有良好的辨識能力。 此外本研究以發生財務壓力機率4%以上為具財務危機銀行。將全部樣本分為預測具財務危機樣本及預測無財務危機之樣本，透過SAS軟體之運算，將所有32項財務指標之平均值、中位數、最小值、最大值及95%之預測信賴區間求出，可做為監理檢查單位之參考數據，且至96年為止，金管會所接管之銀行，除未列為研究樣本者，均為本模型估算出之問題銀行，故若以本模型推估台灣地區銀行違約之機率，將可豫於未發。 The Empirical Study of Banking Operation Performance and Risk Evaluation in Taiwan Abstract This paper discussed the principles of banking management and the establishment of empirical model by selected relative financial variables objectively to evaluate the banking operation performance in Taiwan. Based on the probability of financial deterioration and econometric measurement, some financial indices were chosen to establish a financial warming system to test and verify the accuracy of banking operation to elaborate the function of bankruptcy beforehand and guide the inefficient banks in the right direction. Descriptive statistics and stepwise regression by logistic were used to analyze banking operation performance and risk evaluation in Taiwan. The empirical model was test significantly by likehood ratio testing the good of fit. As far as the accuracy of model, the percent concordant is 96.9% and Somers’D is 96.1%, the Gamma percent concordant also reach 98.3% to proof that this empirical model is excellent to forecast. Moreover, to test the validity of banking default, the samples were divided into two groups- normal banks and default banks and the difference of two groups was test significantly by Kolmogorov-Smirnov statistics This study also found that if the banks with financial pressure probability over 4% were categorized as financial crisis banks. In addition, if the samples were separated as financial crisis and non-financial crisis banks to calculate the mean , median , minimum, maximum, and 95% confidence interval of 32 financial variables. The calculated data was supported usefully to the Financial Supervisory Commission in Taiwan to evaluate the banking operation performance and forecast the probability of bank default.
