Corporate Default Prediction Through Text Mining: Integrating Event, Sentiment, and Network Analyses

Abstract

The importance of textual information in corporate credit risk management is increasingly recognized. While most studies focus on the direct analysis for assessing corporate credit risk, they often overlook the potential impact of inter-company relationships on the likelihood of default. This study, focusing on both intrinsic information about companies themselves and relational information within company networks, explores the potential of advanced text-mining techniques for predicting corporate defaults. We integrate default event extraction, credit sentiment analysis, and relation analysis via co-mention networks using public news on US-listed oil companies between 2014 and 2016. We aim to demonstrate how these advanced text-derived features enhance default prediction during industry upheaval. Our findings reveal that credit sentiment emerges as a crucial predictor of default, alongside network degree and transitivity. High-risk labelled companies are more likely to default than others. Moreover, exposure to media, regardless of being positive or negative, may increase the likelihood of both default and other corporate exits, primarily mergers and acquisitions. This study emphasizes the transformative impact of text analysis on traditional credit risk assessment practices and underscores the value of relational information between companies for default prediction.