Are you a Zombie Firm? An Early Warning System Based on Machine Learning Methods

Abstract

In this paper, we propose an empirical approach that finds the features that matter to categorize zombie firms, separate them from the non-zombies and the recovered, and ultimately predict tomorrow’s zombies and recovered zombie firms. We apply our approach to listed US and European firms. Using machine learning models for feature selection and logistic regressions, we show that an ensemble of firm-level variables on the firm capital, financial, and industry structure lead to
the prediction of zombies and recovered zombie firms. The final model, our early warning system, produces variables that could be of use in monitoring firms’ zombie status versus recovered.

Keywords: machine learning, zombie firms, financial distress

JEL Classification: C55, C63, D22, G32, G33