Predicting financial distress in Latin American companies: A comparative analysis of logistic regression and random forest models

• Published in: The North American journal of economics and finance 2024-05, Vol.72, p.102158, Article 102158
• Main Authors: Barboza, Flavio, Altman, Edward
• Format: Article
• Language: English
• Subjects: Corporate default; Credit risk; Distress prediction; Logistic regression; Random forest
• ISSN: 1062-9408; 1879-0860
• DOI: 10.1016/j.najef.2024.102158

Latin America represents a growing financial market. The performance of its private sector corporations is critical, as inadequate performance and financial distress can lead to significant losses for many stakeholders. This study assesses the efficacy of Logistic Regression (LR) and Random Forest (RF) techniques in predicting corporate distress up to three years in advance. Additionally, we discuss relevant indicators and compare our findings in two different scenarios (pre versus pandemic period). The results indicate that RF outperforms LR in terms of predictive power and error levels. The most effective predictors remained consistent over the 20-year period but varied between the two models. Importantly, the performance levels remained unaffected by the COVID-19 pandemic. •Our model achieved an accuracy of over 90% in predicting firm distress 1-year ahead.•RF proves to be the best model, demonstrating minimal levels of errors.•Profitability, growth, and P/B ratios hold greater importance in the RF model.•No significant variance in predictions before/during Covid-19 observed.