Early warning detection using Logic Learning Machine: Evidence from private firms

Purpose. The paper aims to assess the ability of explainable artificial intelligence (XAI), specifically Logic Learning Machine (LLM), to predict early signals of distress in private companies. Design/methodology/approach. We examined a sample of Italian private firms that activated early recovery procedures, which are matched to healthy firms. The proprietary algorithm developed by Rulex Innovation Labs is used to discriminate between distressed firms and healthy companies based on a set of publicly available data. Results are then compared with those obtained using other (widely employed) methods. Findings. The analysis shows that the LLM method is able to classify distressed firms with high accuracy, outperforming logit models and other AI-based methods. Originality/value. We contribute to the literature on the use of AI in insolvency prediction by exploring the predictive ability of XAI. We also extend the literature on insolvency in private firms, which represent a fundamental part of the economic system and are subject to less scrutiny than public firms. Practical implications. Our results have practical implications considering the recently enforced EU Insolvency Directive, which imposes the implementation of early warning tools that should be easy to use for all entities across all Member States. By using publicly available data on early distress procedures activated by companies, we build an early warning detection system that can be easily employed by companies of all sizes and types.