Evaluation of the Altman Z-score framework for financial distress prediction in Korean firms

Abstract

Abstract

The Altman Z-score is one of the most widely used models for assessing corporate financial distress, but its coefficients were originally derived from U.S. manufacturing firms and may not generalize to other markets. This study evaluates the performance of the original Altman framework in Korean manufacturing firms using 12,651 Korean firm-year observations (2011–2024) constructed from publicly available DART financial statements and compares it with a Korean-specific logistic regression model that re-estimates the coefficients of the five original Altman variables. Model performance was evaluated using a chronological train-test split (2011–2020 training; 2021–2024 testing), confusion matrices, ROC curves, AUC, and bootstrap confidence intervals. The re-estimated Korean model outperformed the original Altman framework on the held-out test set (AUC 0.947 vs. 0.859). To assess generalizability beyond the accounting-based distress definition used for model estimation, both models were additionally validated using an independent dataset of 94 firms subsequently delisted from KOSPI or KOSDAQ. The Korean logistic model consistently achieved higher AUC across all pre-delisting horizons (pooled AUC 0.849 vs. 0.775), indicating improved discrimination of real-world corporate failure events. The study also releases the datasets and source code to support reproducibility. These findings suggest that market-specific statistical re-estimation of the Altman framework can improve financial distress prediction in Korean firms while retaining the interpretability of the original variables.