Do parsimonious bankruptcy prediction models still work after COVID-19? Logistic regression evidence from central Europe

Purpose The purpose of this paper is to develop and evaluate logistic regression (LR) based bankruptcy prediction models for enterprises in the Visegrad Group (V4) countries in the post-COVID period. The research assesses the stability, interpretability and predictive efficacy of accounting-based early warning models across multiple forecasting horizons and national contexts. Design/methodology/approach The empirical analysis is based on firm-level financial data for 24,922 enterprises from V4 countries, obtained from the ORBIS database. Financial indicators from 2020 to 2022 are used to predict enterprises' financial condition in 2023. LR models are estimated for three horizons using a common set of predictors. A pooled V4 model and country-specific models are developed and evaluated using a 70:30 training-testing split, with performance assessed via accuracy, sensitivity, specificity, area under the curve (AUC) and pseudo-R2. Findings The results indicate that LR models provide strong predictive performance, particularly at shorter horizons, with high accuracy and AUC values across the V4 region. Total indebtedness and interest burden emerge as the most significant and consistent predictors of financial distress, while liquidity indicators play a limited role. Predictive performance consistently deteriorates with an extended horizon, exhibiting cross-country differences, with weaker sensitivity in the Hungarian models. Originality/value This study provides one of the first post-COVID evaluations of multi-horizon LR bankruptcy models in the V4 region using a harmonized dataset and unified modeling framework. By emphasizing explainable and economically interpretable models, the paper complements the growing machine learning (ML) literature and offers updated early warning insights for European transitional economies.