산업․시장 경쟁구조 요인을 반영한 부도 예측: Explainable AI 접근

초록

This study aims to address the limitations of conventional bankruptcy prediction models that primarily rely on firm-level financial data and fail to adequately incorporate industry and market structure competition structure factors, while simultaneously enhancing both predictive accuracy and interpretability. We measure bankruptcy risk using two dependent variables: the continuous indicator of Distance to Default (DD) and the binary indicator of default occurrence (Default_D). The key explanatory variables include the Herfindahl- Hirschman Index (HHI), industry growth rate, market share, and firm revenue growth rate. Methodologically, we combine traditional econometric techniques, including logistic regression, fixed effects panel regression, and Cox proportional hazards regression, with explainable AI models. Specifically, we apply the Explainable Boosting Machine (EBM) and XGBoost with SHAP, conducting both global and local interpretability analyses. The empirical analysis is based on data for Chinese A-share listed firms from 2016 to 2023 obtained from the CSMAR database. The results show that, in the DD models, industry concentration positively affects bankruptcy risk, whereas industry growth rate, market share, and firm revenue growth rate have negative effects; all are statistically significant. In contrast, when using the binary default indicator, only firm revenue growth rate has a statistically significant negative effect. In terms of predictive performance, XGBoost outperforms all other models, followed by EBM, while logistic regression shows relatively lower accuracy. Interpretation results confirm DD as the most influential variable across all models, with industry-level variables also contributing to predictive improvements. EBM further reveals a nonlinear relationship between industry sales growth and bankruptcy risk, while SHAP-based interpretation of XGBoost provides policy-relevant insights. This study contributes to the literature by extending bankruptcy risk analysis beyond firm-level financials to incorporate quantitative industry and market competition structure factors, and by proposing an integrated predictive-interpretive framework that offers high academic and policy relevance through explainable AI techniques.

키워드