In recent years, there have been a number of factors that have, to some extent, limited progress in credit risk modeling due to a greater focus on regulatory compliance (IRB and IFRS 9 parameters). However, there is a growing interest in the industry to improve credit risk models by incorporating artificial intelligence techniques, and supervisors are increasingly receptive to such techniques,
In this sense, major institutions are working, to a greater or lesser extent, on incorporating artificial intelligence techniques to improve their credit risk models. Typically, the focus is on credit risk models that do not impact IRB portfolios. However, some institutions are integrating AI techniques into IRB scoring and rating systems, subject to supervisory approval.
AI applied to Credit Risk Models
This means that banks face challenges related to defining a strategy with proper prioritization of initiatives, establishing a control framework that complies with regulatory requirements (AI Act) and having the appropriate information, technology and skills.
Management Solutions can support the integration of AI in the different stages of the credit risk model lifecycle as well as in cross-cutting issues (data, IT infrastructure and architecture, framework and regulation). In addition, MS has a proprietary tool - ModelCraftTM - as an accelerator.
Our value proposition is organized into 4 work streams:
1. Model lifecycle: we offer our analytical capabilities in 3 areas:
-
Definition and prioritization of use cases and development of PoCs, taking into account the entity's needs and industry results.
-
Support in model development, automation and industrialization processes involving AI, as well as in the definition and implementation of Explainable AI (XAI) for better interpretation of AI model outputs.
-
Adaptation of internal validation and model risk frameworks to the use of AI and support in internal validation and model risk management (MRM) exercises.
2. Data: identification, capture and organization of data from new sources and enhancement of Data Government and Data Quality issues.
3. IT Infrastructure: creating a technology roadmap and evolving the IT infrastructure and architecture for developing and validating AI models, including transitioning to open source programming languages such as Python and adopting MLOps approaches. ModelCraft™ could be considered as an accelerator for introducing AI into the modeling processes.
4. Regulation: improving modeling frameworks, policies and procedures to ensure proper integration of AI techniques in compliance with new regulations (such as the AI Act).