Unlocking bankruptcy clues: A novel sentence-based machine learning approach

Abstract

Our study examines the predictive power of Management Discussion and Analysis (MD&A) sections in the context of bankruptcy prediction. Leveraging contextual sentence embeddings from a pre-trained Transformer model (BERT), we introduce a novel prediction model designed to identify MD&A sentences associated with bankruptcies. Our sentence-level model is competitive with various document-level approaches, particularly when excluding boilerplate content. Furthermore, our sentence-level approach enhances model interpretability by uncovering high-risk topics such as performance deterioration, financial losses, and cost reductions. Lastly, we provide insights into critical high-risk disclosure patterns through structural and syntactical analysis.