스코어 얼라인먼트 기반 중소기업 부도예측모형의 운영 효율화

This study utilized data from small and medium-sized enterprises (SMEs) that have accessed the Korea Credit Guarantee Fund from 2014 to 2021 to construct a model for predicting the likelihood of SME defaults. Similar to international and domestic credit bureaus using logistic regression models for transparent and straightforward explanations of credit score, this research also employed logistic regression models for the same purpose. Particularly, to enhance discriminative power, binning was performed on all variables, which were then used as input variables. Considering the number of SME and the target rate, five segments were created, with modeling conducted for each segment. When compared with other methods such as XGBoost and Neural Network, the logistic regression model based on binning and segmentation was found to provide excellent explanatory power and competitive discriminative ability in prediction results. Furthermore, this research proposes a method that allows financial institutions operating credit score-based cutoffs to apply the model as intended during its development without needing to rebuild it, even when the relationship between credit scores and credit risks changes due to economic fluctuations. This involves establishing a baseline between the logit of default probability and the score at the time of model development using PDO (Points to Double Odds) and odds-based scoring, and rotationally transforming the line between logit and score at the application time back to the baseline established at development time, thereby enabling efficient model operation.