Predicting Probability of Default Under IFRS 9 Through Data Mining Techniques

Abstract

Data mining techniques were employed to automatise decision-making processes in several domains. In the banking context, the introduction of IFRS 9 on Financial Instruments has impacted not only on the area of accounting and financial reporting, but also on banks’ credit risk measurement and management processes, promoting effective and efficient data mining applications. In detail, banking management can benefit from these techniques by extracting knowledge from data to support more advanced models, in particular for the assessment of credits deriving from lending activity, in accordance with the Expected Loss Approach provided by the new standard. In this study, we exploit data mining techniques to measure the probability of default of credits with specific features at the reporting date. We consider supervised machine learning to build predictive models and association rules to infer a set of rules by a real-world data-set, reaching interesting results in terms of accuracy.