消費性貸款之風險管理 — 存活表之應用

Abstract

本研究提議金融機構應建構兩張存活表，以應用於消費性貸款的風險管理。第一張存活表是用於監控現有貸款客戶的信用風險變化，此存活表以貸款初期的申請資料以及還款期間的繳款行為變數作為解釋變數。由於收集這些行為解釋變數的資料至少需時一年，因此，此存活表可揭露所有貸款人自還款期第二年起的違約機率。另外，本研究首度提出的第二張存活表則是用來做為放貸決策之輔助工具。藉由應用這兩張存活表，消費性貸款的風險控管將更為完備。與用於監控現有貸款客戶之風險變化的信用模型不同，目前的放貸決策系統多是以申貸資料作為放貸評估依據。但此申貸資料皆為靜態資料，無法提供貸款者最近的消費行為變化，因此在風險評估上略有不足。為彌補此缺點，本研究在期初申請資料外，另採納貸款者在申貸前一年內的信用卡消費與繳費行為變數作為解釋變數。由於信用卡是目前最普及的消費金融工具，以信用卡的消費與繳費行為變數做為消費性貸款的代理人十分方便與適合。 為配合這兩張存活表的應用，本研究亦提出一套信用評分系統，其中包含消費性貸款的評分方法以及分組方法。應用台灣一家知名金融機構所提供之消費性貸款資料做實證研究，發現此信用評分系統的正確性及可行性皆極高。實證結果顯示，將信用卡消費與繳費行為變數納入解釋變數的確能提高放貸決策模型的準確度。再者，與目前最為普遍運用的邏輯迴歸分析模型相比，存活分析及存活表無論在監控現有貸款客戶之風險變化或是在放貸決策上，皆有較好的表現。

This study introduces two survival tables for the risk management of consumer loans. The first survival table, constructed by consumer loan application data and behavioral variables, is a tool to monitor the credit risk of existing borrowers dynamically. And another survival table which tells the default probabilities of new loan applicants is first introduced in this study to complete the risk management of consumer loan portfolios. These two survival tables are helpful not only in the risk management of individual consumer loans but also in that of consumer loan portfolios. Compared with the credit methods for monitoring existing loans, consumer loan grant decision models are usually composed of only static application data which are weak in seizing the changes of consumer behavior. To improve the efficiency of survival table for grant decision, not only loan application data but also credit card behavioral variables are included in survival analysis. Since credit card is the most commonly held consumer credit product and the frequency of credit card behavioral data is the same as that of consumer loan behavioral ones, the behavioral variables of credit cards are suitable proxies for these of consumer loans. To match up the practicability of survival tables, this study also introduces a credit scoring methodology which contains the computation of credit score and the classification rule that divides consumer loans into groups. Evidences from Taiwan show the soundness of the credit scoring system. They indicate that the efficiency of survival table for grant decision is improved by including the credit card behavioral variables, too. It is also pointed out that both survival table and survival analysis are competitive with, and sometimes superior to, logistic regression, in the failure prediction of existing as well as new consumer loans.