Machine learning in corporate credit rating assessment using the expanded audit report

Abstract

We investigate whether key audit matter (KAM) paragraphs disclosed in extended audit reports—paragraphs in which the auditor highlights significant risks and critical judgments of the company—contribute to assess corporate credit ratings. This assessment is a complicated and expensive process to grade the reliability of a company, and it is relevant for many stakeholders, such as issuers, investors, and creditors. Although credit rating evaluations have attracted the interest of many researchers, previous studies have mainly focused only on financial ratios. We are the first to use KAMs for credit rating modelling purposes. Applying four machine learning techniques to answer this real-world problem—C4.5 decision tree, two different rule induction classifiers (PART algorithm and Rough Set) and the logistic regression methodology—, our evidence suggests that by simply identifying the KAM topics disclosed in the report, any decision-maker can assess credit scores with 74% accuracy using the rules provided by the PART algorithm. These rules specifically indicate that KAMs on both external (such as going concern) and internal (such as company debt) aspects may contribute to explaining a company’s credit rating. The rule induction classifiers have similar predictive power. Interestingly, if we combine audit data with accounting ratios, the predictive power of our model increases to 84%, outperforming the accuracy in the existing literature.