Credit Risk Prediction Using Calibration Method: An Application In Financial Scorecard

Lee, Choon Yi (2022) Credit Risk Prediction Using Calibration Method: An Application In Financial Scorecard. Master dissertation/thesis, UTAR.

Preview

PDF Download (1697Kb) | Preview

Abstract

Machine Learning models have been extensively researched in the area of credit scoring. Banks have put in substantial resources into improving the credit risk model performance as improvement in accuracy by a fraction could translate into significant future savings. Given the lack of interpretability in machine learning models, it is often not used for capital provisionin g in banks. This paper uses the Taiwan Credit Card dataset and illustrates the use of machine learning techniques to improve assessment of credit worthiness using credit scoring models. In factor transformation for a credit scorecard construction, Decision Tree technique showed the ability to produce quick and predictive transformation rule. Besides, model comparison result showed that Artificial Neural Network and Gradient Boosting Approach have great predictive power compared to traditional logistic regre ssion scorecard . Credit underwriting decision could be improved by implementing a better discriminatory power scorecard, as more good customers are likely to be better than score cut thus accepted by banks. Probability of Default (PD) Calibration moff and aps model scores to output PD that reflects portfolio underlying performance. This paper illustrates approach to perform PD calibration for machine learning models that can be used to align with banks internal application scorecard strategy . Calibration Plot and Binomial Test assessment showed that traditional scorecard approach performed better with least risk of underestimation of actual PD . Both tests suggest estimation purpose.

Actions (login required)