What makes companies zombie? Detecting the most important zombification feature using tree-based machine learning

Abstract

Tree-based machine learning models are crucial for identifying key features of company zombification, which remain underexplored in current literature focused solely on determinants. This study addresses this gap by employing machine learning feature analysis to identify and analyze the critical factors driving zombification, offering a fresh and data-driven perspective on the issue. Three different feature sets are examined: (i) Feature Zoo, (ii) Logistic regression-based, and (iii) Lasso-based features, focusing on critical internal characteristics of firms. These feature sets are applied to four tree-based algorithms—Decision Tree, Random Forest, Gradient Boosting Model, and XGBoost—chosen for their white model capabilities, allowing the feature extractions. The results indicate that Debt and ROA consistently have the highest feature scores, suggesting they are crucial for predicting zombie companies. Additionally, the Lasso-based feature sets provide the best evaluation metrics, indicating that the two-step filtering process effectively improves the predictive model for zombie companies. The study enriches the literature by extending the anatomy of zombie companies with a more advanced approach. The results also address Debt and ROA as the most significant features for identifying zombie firms. Managers and policymakers should prioritize monitoring Debt and ROA as early warning indicators for company zombification.