EURO 2025 Leeds
Abstract Submission

1375. Credit Scoring Model of Small Sized Firms Using Default Data during COVID-19 Pandemic

Invited abstract in session WC-9: Methods and models in portfolio and risk management, stream OR in Finance and Insurance .

Wednesday, 12:30-14:00
Room: Clarendon SR 2.01

Authors (first author is the speaker)

1. Hiroumi Naganuma
Risk Management Department, Micro Business and Individual Unit, Japan Finance Corporation
2. Masaru Iwase
Risk Management Department, Micro Business and Individual Unit, Japan Finance Corporation
3. Masahiro Toshiro
Risk Management Department, Micro Business and Individual Unit, Japan Finance Corporation
4. Shinsuke Sasaki
Regional Community Policy Department, The University of Shimane
5. Kenzo Ogi
School of Commerce, Senshu University
6. Norio Hibiki
Department of Industrial and Systems Engineering, Keio University

Abstract

Japanese banks employ credit scoring models with default as the dependent variable to manage their debtors' credit risk. During the COVID-19 pandemic, while the financial condition of firms deteriorated, the government's economic measures reduced the number of corporate bankruptcies. This gives the lower correlation between financial indicators and default occurrence and may reduce the accuracy of models. Therefore, handling data affected by the COVID-19 pandemic is a critical issue to model the credit scoring for Japanese banks. In this paper, we built nine logistic regression models with different learning periods and compared the accuracy ratios (ARs), using Japan Finance Corporation's data on 799 thousand loans from FY 2014 to FY 2021, which include both pandemic and non-pandemic data periods. Our results indicate that incorporating more recent data improves model accuracy, even if including the COVID-19 pandemic period. The model with the highest AR of 58.3% on average is built using the data from FY 2014 to FY 2021. On the other hand, this result is consistent with the previous studies such as Yanagisawa et al. (2007) and Sawaki et al. (2017) who suggest that the accuracy of credit scoring models for Japanese SMEs is affected by the amount of data and learning periods. In addition, our paper can strengthen these studies because we show it even in the case of handling unusual data such as during pandemics.

Keywords

Status: accepted

Back to the list of papers