應用Copula模型於信用卡風險推估

Abstract

由於國際清算銀行（Bank of International Settlement，BIS）所屬的巴塞爾銀行監理委員會（Basel Committee on Banking Supervision簡稱巴塞爾委員會）在2001年一月修正了1988年所公布之巴塞爾協定並公布了新版的巴塞爾協定草案(Basel II)中，提出了對銀行風險模型建置的兩項建議，一為採用巴賽爾協定草案中的標準化公式。另一項則為銀行在監理者的同意之下，建立適用該銀行的內部模型。因此各銀行紛紛開始尋求發展符合自身情況的模型。 隨著市場的變動以及經濟的發展，個人財富所得顯著的增加，因此銀行致力於開發個人消費金融產品，特別是信用卡在近幾年因資訊科技的發展與金融國際化的趨勢，成為人們支付帳款的重要工具，伴隨信用卡而來的手續費及高額循環利息，讓銀行獲利的有了新來源。然而信用卡是一種信用借款的工具，客戶不需提供任何擔保品，只有根據客戶本身的信用狀況給予不同的消費額度，因此若消費者違約無力償還消費款項，即形成銀行的呆帳，對於銀行的財務影響甚大，構成風險管理上的威脅。 而信用卡消費者的行為模式、信用評分的方法又與過去的產品差異甚大，加上銀行之間又有不同的產品組成，因此並沒有所謂的標準化信用卡風險模型，提供了銀行更多的機會去發展符合本身體質的管理模型。 本研究希望透過將各群信用卡顧客的違約機率變化率作為風險因子，並將各群之間的相關性結構利用Copula模型模擬建構，結合風險損失的概念，計算出預期損失，期能得出接近真實分配的狀況，提供銀行作為風險控管的依據。

Basel Committee on Banking Supervision of Bank of International Settlement revised Basel I in 2001 and announced Basel II. It proposed two ways for the establishment of risk models. The first is adopt the standard models which Basel II suggests. The other is that banks could develop internal ratings based models to estimate their own risk. With the development of economy, personal wealth increases significantly. The market of consumers is growing. Therefore, banks devote to provide more consumer banking products. Especially credit card has become an important tool to pay the bill. Banks could profit from its fees and high revolving interest. However, credit card is kind of a fiduciary loan. Banks could give them different credit lines and clients don’t need to provide collaterals. As a result , when clients couldn’t repay money, it forms a bad debt. This is a great challenge to risk management and threatens the financial structure of the bank. In addition, the behavior of credit card consumers and the evaluation of credit quality is different from other products. Thus there is no standard credit model, and this offers banks more opportunity to develop their own model. In this research, I want to use the change of default probabilities of each group as the risk factor. Then I will use Gaussian Copula as the correlation structure and combine important elements to calculate expected loss. I expect to simulate a situation close to real distribution so that it could be a basis for banks to manage their risk.