中小企業向來是我國經濟發展主幹及產業發展磐石，根據2010年「中小企業白皮書」發布資料顯示，2009年國內中小企業家數占全體企業家數的97.91％；在就業人數方面占全國就業人數78.47％。由此些調查數據可體認及了解得知，在台灣要走向國際化的同時，台灣的中小企業是不可忽略的一大環節，對於穩定我國就業市場以及經濟發展具有舉足輕重的地位。 本研究利用Logistic迴歸分析，試圖從既有存在於信用評等表中之因子，去評價及衍生經過量化之財務與非財務變數，提供一套更適合個案銀行作為中小企業信用風險評價之預警模型，以降低資訊不對稱所帶來的違約風險。 樣本係以個案銀行11個變數為基礎，進而另衍生出11個變數，合計共22個變數。利用Logistic迴歸，先期進行檢定基本模型與各個衍生模型對預測違約能力的顯著性、係數正負向關係與odds ratio外，再者，選擇基本模型與衍生模型3，利用T-test及無母數檢定（Anova﹐即變異數分析）等方式，進一步探討其估算的違約機率（簡稱PD值）是否存在統計的顯著差異；最後，根據申貸案件的核准與未核准兩個情境，以及個別情境的正常戶與違約戶，更嚴謹地檢視兩種實證模型估算的PD值是否存在顯著差異。由實證結果，研究結論如下： 一、本研究無法拒絕基本模型與衍生模型3的PD值平均數及中位數相同的假設，個案銀行現行的信用評等制度堪稱理想，授信時不當接受不良客戶或拒絕合適客戶的可能性並不明顯。 二、根據申貸案件的核准與未核准兩個情境，以及個別情境的正常戶與違約戶，更嚴謹地檢視兩種實證模型估算的PD值並無存在顯著差異。此點除了反映個案銀行的基本模型和信用評分制度堪稱理想外，並可藉由衍生模型新增變數的正負值，協助個案銀行授信政策或授信條件及信用評等變數門檻標準的設置。 三、以整體性架構觀察個案銀行信用評等模型，似乎在加上本研究新增之11個財務及非財務面變數後，預測力上有顯著性提升，以供後續研究者參酌。 According to the White Paper on Small and Medium-sized Enterprises in Taiwan, published by the Taiwanese government in 2010, 97.91% of the business enterprises in 2009 were small and medium-sized enterprises (SMEs), where 78.47% of the employed population worked. As can be learned from the figures, SMEs—always a focus in and a foundation for Taiwan’s economic development—should not be overlooked when the country seeks internationalization. The research evaluates and proposes quantified financial and non-financial variables in a bank’s credit rating model for SMEs, using logistic regression on a sample of 22 variables—11 basic variables from our case bank and 11 derived ones. It further uses a T-test and a nonparametric test, i.e. an Anova, to explore if there is any significant statistical difference among the estimated probabilities of default (PDs). Our empirical analysis concludes that: 1.The research fails to reject the hypothesis that PDs in the basic model and in the Derived Model 3 have a common average and a common median. In other words, the current credit rating system of our case bank is rather ideal as it’s not significantly possible that the bank would accept a bad client or turn away a good client. 2.By further dividing the applications into approved cases and disapproved cases, as well as non-defaulters and defaulters, we scrutinized the PDs of the two empirical models and found no significant difference. This shows our case bank’s basic model and credit rating system are rather ideal; Moreover, it suggests that the derived model, where new positive or negative values are assigned to the variables, may be helpful for the bank in deciding on lending policies, lending standards and thresholds for credit rating variables. 3.Overall, with the 11 new financial and non-financial variables added, our case bank’s credit rating model shows significant increase in its predictive ability
