信用卡信用評分模型的建立與評等

Building model for credit scoring and credit rating of credit cards

廖仁傑(JEN-CHIEN LIAO)

指導教授 ： 吳博欽

中原大學/商學院/商學博士學位學程/博士(2013年)

https://doi.org/10.6840/cycu201300925

摘要

政府為尊重市場機置，強化風險控管，及銀行業者也會考量成本收益面，故為兼管風險與利潤，銀行業者必須採取差別利率(即按風險高低採取差別定價)，以有效降低信用風險及提高超額利潤，然評分模型也必須適用各種經濟環境以調整信用評分，使銀行業在追求競爭及收益的環境下，真正扮演好從事放款與投資金融中介的功能，但銀行仍必須定期公開利率資訊，讓消費者有自由選擇加入及退出的權利；本文企圖採取個案銀行的方式，介紹個案銀行如何篩選重要訊息值變數而建立信用評分卡，過去研究當中都是在有限樣本建立Logit迴歸模型，模型的係數與其後的統計檢定是否產生偏差，國內研究通常無相關探討，而視為無偏差的假設，故本文於個案銀行信用評等研究中，首度加入拔靴複製法探討模型是否產生偏差的情形，提供更加科學的驗證模型準確度，使銀行如何在有限樣本抽樣下，調整信用風險模型及信用評分，另考量個案銀行客戶在剛進件及之後產生交易行為的而建置靜態評分模型及行為評分的動態模型(利用分量迴歸建置)及考量不同景氣變動下客戶違約機率的相關性(利用copula考量客戶前後兩年違約機率相關性)，其模型是否仍適用，並如何將違約機率利用線性轉換成信用評分，而如何建構內部評等產生差別利率，另為確保信用評分模型的穩定性及可靠性，而驗證該信用評等是否穩定以符合建立一套符合巴賽爾規範下的資本計提方法；故期能盼望藉由內部評等方法，建立信用卡信用評分系統，以反映客戶的風險，使銀行獲取合理的利潤及消費者負擔合理的價格,而形成雙贏的局面 。

關鍵字

線性轉換 ； 分量迴歸 ； 拔靴法 ； copula ； 信用評分 ； 信用評等

並列摘要

The government reinforces risk control for the respect of market mechanism. Bankers consider their costs and profits, and therefore they have to adopt differential interest rates (i.e. determining interest rates based on the level of risks) for both risk control and profits in order to minimize credit risks and maximize excess profits. However, the scoring model must allow the adjustment of credit scoring for different economic environments so that banks are able to play the role of loaning and media of investment financing in an atmosphere of competition and profit seeking. Still, banks are required to publish interest rate information on a regular basis to allow consumers to have choices to get in and out. This study is intended to show how a bank filters out important information value variables and establish credit score cards using a bank as the subject of study. Previous studies focused on building Logit regression models based on finite number of samples. The coefficients used in models and whether bias occur in subsequent statistics tests were rarely discussed in the studies and the assumption of no bias was often made. For this, bootstrapping method was introduced in the study of the credit rating in the bank selected to see if bias was produced in model, thus providing better accuracy for scientific verification model. For the considerations of the static scoring model built for evaluation of new credit card applications and subsequent transactions, dynamic model for behavior scoring (established using quantile regression) and the correlation of client’s probability of default in the economic fluctuation (using copula to evaluate the correlation of probability of default in two years), it is necessary to know whether the models are still applicable, how to convert probability of default into credit scores using linear transform and how to establish internal rating for differential interest rates. Also in order to ensure the stability and reliability of the credit scoring model, the credit rating has to be verified to build a method that meets the capital requirement of Basel Accord. The hope was to establish a credit card scoring system through internal rating in order to reflect clients’ risks, allow reasonable profits and prices that are affordable to consumers and create a win-win for banks and consumers.

並列關鍵字

Copula ； Linear Transformation ； Bootstrapping ； Credit Scoring ； Credit Rating ； Quantile Regression

參考文獻

1. Altman, E. I. (1968), “Financial Ratio, Discriminant Analysis and the Prediction of Corporate Bankruptcy” , Journal of Finance, 23, 4, September, 589-609.

self-regulation and implications for environmental performance”,.

3.Beaver, W. (1966), “Financial ratios as predictors of failure. Empirical research in accounting: Selected studies”,Journal of Accounting Research 5, 71–111.

4.Bouye, E., V. Durrleman, A. Nikeghbali, G. Riboulet, and T. Roncalli (2000) ,“Copulas for Finance - A Reading Guide and Some Applications”, Working Paper.

5.Black, F. and M. Scholes (1973), “The Pricing Of Options And Corporate Liabilities”,Journal of Political Economy, 81, pp.637-659.