羅吉斯迴歸在分期付款違約之應用

APPLICATION OF DEFAULT OF INSTALLMENTS IN LOGISTIC REGRESSION

吳柔瑾(Jou-Chin Wu)

《Journal of Data Analysis》 17卷2期 (2022/06) Pp. 1-30

https://doi.org/10.6338/JDA.202206_17(2).0001

摘要

分期付款已成為現代消費者常有的一種消費方式，無卡分期，亦即不使用信用卡即可享有消費分期，也為當今趨勢。然而，為了有效掌控承辦此項業務的風險，提供分期付款的公司必須嚴格審核新進案件，以降低違約帶來的損失。本研究之對象為某借貸公司，由於借貸公司非銀行，無法依法取得聯徵中心的信用評分，因此本研究以過去顧客分期資料建立羅吉斯迴歸模型。由於違約數與非違約數並不相同，故以1:1精細抽樣使資料平衡後，分別以向前選取法向後選取法及逐步選取法建立模型，得向前選取法及逐步選取法較適合此資料。模型中顯著的自變數有年齡、教育程度、居住狀況、婚姻狀況、產品別、適用群組、正常繳款次數、延滯繳款天數介於30至60天、分期期數、辦理分期金額、年資、罰單總筆數等等。期待研究結果能為借貸公司帶來更有效率的審核方式，並提供消費者更多元且自由的消費方式。

關鍵字

分期付款 ； 違約機率模型 ； 羅吉斯迴歸

並列摘要

Installment payment has become a common way of consumption for modern consumers. Installment without cards, that is, you can enjoy installment of consumption without using credit cards, is also the trend nowadays. However, in order to effectively control the risks of undertaking this business, companies that provide installments must strictly review new cases to reduce the losses caused by default. The research object is a lending company. Because the lending company is not a bank, it cannot legally obtain the credit score from the Joint Credit Information Center. Therefore, this study uses the past customer installment data to establish a Logistic regression model. Since the number of defaults and non-defaults are not the same, I oversample the data to be balanced, the model is established by forward, backward and stepwise selection methods respectively. The forward selection method and the stepwise selection method are more suitable. The significant independent variables in the model are age, education level, living status, marriage status, product type, applicable group, normal payment times, delayed payment days between 30 and 60 days, period of installments, the amount of installments, working experience, and number of fines, etc. It is expected that the research results will bring more efficient audit method to lending companies and provide consumers with multiple and free consumption mode.

並列關鍵字

default of installments ； probability of default model ； logistic regression

參考文獻

Hull, J., 2003, Option, Futures, and Other Derivatives, Fifth edition, Prentice-Hall Press.

Kubat, M., 2015, An Introduction to Machine Learning，pp 43-48.

Kutner, M. H., C. J. Nachtsheim & J. Neter, 2004, Applied Linear Regression Models, pp 555-577.

Mason, S. J. & N. E. Graham, 2002, Areas beneath the relative operating characteristics (ROC) and relative operating levels (ROL) curves: Statistical significance and interpretation. Q.J.R. Meteorol. Soc.

Zou, K. H., A. J. O'Malley & L. Mauri, 2007, Receiver-operating characteristic analysis for evaluating diagnostic tests and predictive models. Circulation, 6.