整合金融機構風險評估與信用評等模式之研究

Abstract

風險評估及信用評等結果是金融機構用以評量借款企業償債能力的重要依據，然而在現今經濟不景氣的大環境下，節節升高的逾放比率使得越來越多的金融機構必須檢討其現有信用風險評估模式的缺失，以對貸款企業作出更正確有效的放款決策。一般信用風險評估制度之設計大多是根據以借款企業的品格、能力、資本、業務狀況及擔保品等五大構面，而影響此五個構面之因素很多，要從這麼多個影響企業信用的因素中對企業做出正確的信用評等實非易事。現有之中、外文獻雖發展出許多信用評等模式來探討此問題，但多以上市上櫃公司為研究對象，這與台灣一般放款公司實際接觸的中小企業借款者有所出入，因此這些文獻所建議之模式其應用價值有限。因此，本研究乃針對台灣金融機構之中小企業借款者，發展出一套風險評估與信用評等整合流程，此流程之構建主要分五階段：(1) 選擇變數與收集資料；(2)利用分析層級程序 (Analytic Hierarchy Process；AHP) 法找出各變數之適當權重；(3)分別利用複變數判別分析(Multivariate Discriminant analysis)、羅吉斯迴歸(Logistic regression)與核心法(Kernel method)構建放款風險評估模式以區分正常與違約資料；(4)構建正常與違約資料之多等級企業信用評等模式將借款企業作多等級的評等分類；(5)構建違約資料存活期預估模式與違約機率表，可提供金融機構決策者關於借款者還款期長短與應收擔保品多寡之資訊，以有效幫助其降低放款風險。本研究並利用反應曲面法(Response Surface Method；RSM)快速有效地找到核心法中最佳參數水準之設定值。本研究最後透過台灣某金融機構所提供中小企業借款者之實際歷史資料，驗証了本研究所提之風險評估與信用評等整合模式確實有效可行。

The risk assessment and credit rating are two important indicators for financial institutions to evaluate the payment capability of the loan applicants. However, the raising ratio of the bad loans drives financial institutions to review and reconstruct their credit risk assessment models to make right and efficient decisions of loaning under the economic depression. In general, most of the models are built based upon five dimensions: applicant’s personality, payment capacity, capital, business condition, and collaterals. Many studies proposed several credit rating models for evaluating the listed companies, but these credit rating models would be invalid when they are employed for evaluating the credit of small and median enterprise(SME). Therefore, an integrated risk assessment and credit rating model is developed for the loan applicants of small business enterprise. The proposed procedure has five stages: (1) selecting variables and collecting data, (2) finding the appropriate weight of variables by using Analytic Hierarchy Procedure, (3) constructing a model of loan risk assessment to discriminate between good and bad loaners using Multivariate Discriminant analysis, Logistic regression and Kernel method separately, (4) constructing a multi-level credit rating model for good and bad loans, (5) constructing a prediction model of survival for bad loaners and default probability table which can provide financial institutions information to make decision about the proper length of payment periods. Response Surface method(RSM) is also used in this study to find optimal parameter setting level of Kernel method. Finally, a real case is provided to demonstrate the effectiveness of the proposed procedure.