應用邏輯斯迴歸與支持向量機建構兩階段提前清償模型

Abstract

當企業以授信方式向銀行或金融機構借款時必須支付利息及本金，而企業面對市場利率變化時，可能會選擇支付借款罰金而提前清償。對銀行或金融機構而言，當企業發生提前清償行為時，會使預期之利息收入減少，現金流量發生變化，使銀行或金融機構在財務管理上碰到許多問題。因此，企業提前清償機率的預測精確與否，對提升銀行或金融機構的獲利有舉足輕重的影響。過去國內、外文獻所提出之提前清償模型多是用在個人住宅貸款，可能不適用於建構台灣企業提前清償模型。此外，現有的提前清償模型多是利用統計方法如：邏輯斯迴歸模型、判別分析等，將借款客戶分成正常及提前清償案件兩類。利用這些判別模型進行提前清償判別時，常會出現某類案件(如：提前清償案件)準確率高，但另一類案件之準確率偏低的現象，即使此種判別模型的整體準確率不錯，但由於對某類案件之誤判機率過大，對銀行或金機構而言實用性不高。因此，本研究利用一套兩階段的提前清償模型，以提升現有判別模型判定之準確率。本研究利用邏輯斯迴歸(logistic regression)與支持向量機(support vector machine, SVM)方法建構此兩階段提前清償模型，以供銀行或金融機構來制定最佳之放款策略。本研究最後利用國內某金融機構所提供之中小企業借款歷史資料，驗證了本研究之兩階段提前清償模型確實有效可行。

It is common to pay interest when enterprises borrow funds from the banks or financial institutions. Generally, enterprises will prepay their mortgage loans with prepayment penalty if the market interest rates decline. If interest rates remain stable or increase, enterprises may put off prepayments until rates decline or other circumstances arise. This prepayment behavior will decrease the banks or financial institutions’ expected interest and make an unexpected cash flow changes. Therefore, accurately predict prepayment behavior plays an important role for banks or financial institutions to enhance their profitability. According to domestic and foreign studies, it is known that prepayment models are mainly utilized in mortgage loans of individual real estate so that the existing prepayment models may not be applicable for Taiwanese enterprises. In addition, the existing prepayment models mostly use single statistical method (such as: Logistic Regression and Discriminant Analysis), and classify the loan enterprises into prepayment/non-prepayment groups respectively. It is often found that the accuracies for both groups are not balanced, i.e. the accuracy for a particular group (such as the prepayment group) is significantly higher than the other. Although those prepayment models may have high average accuracy for both groups, they are not practical to use by banks or financial institutions. In this study, we propose a two-phase prepayment model using logistic regression and support vector machine (SVM) to promote the accurate identification rate. The proposed prepayment model can assist banks or financial institutions to develop the better lending strategy. A real case from a Taiwanese financial institution is provided to demonstrate the effectiveness of the proposed prepayment model.