1364. Credit Scoring Model for Education Loan Using Data on Customers Rejected in Loan Screening
Invited abstract in session MC-9: New challenges for risk management, stream OR in Finance and Insurance .
Monday, 12:30-14:00Room: Clarendon SR 2.01
Authors (first author is the speaker)
| 1. | Masahiro Toshiro
|
| Risk Management Department, Micro Business and Individual Unit, Japan Finance Corporation | |
| 2. | Yusuke Hikidera
|
| Micro Business and Individual Unit, Japan Finance Corporation | |
| 3. | Kenzo Ogi
|
| School of Commerce, Senshu University | |
| 4. | Norio Hibiki
|
| Department of Industrial and Systems Engineering, Keio University |
Abstract
Financial institutions employ credit scoring models to screen loan applications for approval or rejection. The models calculate credit scores based on explanatory variables such as customer attributes, past transactions, and credit history. Previous studies on models for education loans, such as Hibiki et al (2011) and Bandyopadhyay (2016), proposed models built to evaluate the default occurrence using data only on customers approved in loan screening. However, approved and rejected customers are likely to have different attributes, and models built using data only on approved customers may be inadequate for screening all loan applications. The contribution of this study is to clarify the effectiveness of using data on rejected customers in model building using 1.2 million education loan applications to Japan Finance Corporation from FY2011 to FY2022. We build a model to evaluate the default using data on approved customers only, and a model to evaluate the rejection using data on both approved and rejected customers. A total of 42 explanatory variables are selected in the two models, but only 5 are common. We then propose to integrate the credit scores calculated from these two models to screen loan applications. In validation comparing the AUC calculated as positive for both defaulted and rejected customers using out-of-sample data, the integrated score is consistently more than 5 points higher than the credit score calculated by the model to evaluate the default.
Keywords
- Financial Modelling
- Finance and Banking
- Risk Analysis and Management
Status: accepted
Back to the list of papers