Title: Predicting credit rating migrations by combining financial, market, and textual data
Authors: Manon Reusens - KU Leuven (Belgium) [presenting]
Kamesh Korangi - University of Southampton (United Kingdom)
Seppe vanden Broucke - KU Leuven (Belgium)
Christophe Mues - University of Southampton (United Kingdom)
Cristian Bravo - Western University (Canada)
Bart Baesens - KU Leuven (Belgium)
Abstract: Estimating credit risk is critical for all entities in business. Therefore, several proxies for the creditworthiness of companies exist, for example, credit ratings issued by credit rating agencies, and much research is conducted into the predictive factors of credit risk. In literature, credit risk is often estimated using financial data. However, we predict credit rating migrations by combining financial data, market data, and textual data. Moreover, we analyze how text can facilitate the predictions of these migrations and what its added value is on top of solely using financial and market data. Furthermore, we also compare different ways to incorporate textual data. In the past, text was often summarized into so-called NLP scores, like subjectivity and polarity, before predicting credit risk. However, through recent advances in the field of Natural Language Processing, full texts can easily be incorporated with pretrained transformer models. We compare multiple models and further analyze the effects of using full text over only incorporating these summarizing NLP scores.