Machine learning for credit applications: a recent EBA discussion paper
On November 11, 2021, The European Banking Authority published a discussion paper on Machine Learning for Internal Rating-Based approach. This paper looks like the first serious step towards accepting the use of machine learning models for credit applications and for calculating the capital requirements of credit portfolios.
In recent years, there has been much interest (and excitement) in banks and consulting firms about exploring machine learning models for credit-related topics: ranging from assessing the creditworthiness of new credit applicants to monitoring existing loans and flagging those that are potentially “troublesome”. However, scepticism about such models has been quite significant, partly fuelled by the fact that models from this new toolkit have not been accepted by the regulators: e.g. for capital requirements calculations.
The application of machine learning in financial institutions is seen as offering several potential benefits: faster decision making, significant cost savings due to lower human involvement in loan application assessment and monitoring with potentially better forecasts of clients’ default behaviour; resulting in better risk management and estimation of capital requirements. However, applying machine learning to the highly sensitive and reputationally precarious fields of credit has many potential pitfalls; the EBA discussion paper provides a good and comprehensive overview of those issues.
- The recent EBA discussion paper on Machine Learning (ML) for Internal Rating-Based approach opens the door for ML usage for Pillar 1 capital requirements.
- Banks should form an opinion about potential use of machine learning in their business. This will soon be possible not only for credit application and arrears management but also for the required capital calculations.
- Banks should start getting experience with ML models by running them parallel with the current regulatory allowed models for credit risk.
- Machine learning is a fast-evolving, complex, and error-prone domain of expertise. Hiring and creating the needed skills set on all levels is highly recommended.
- Probability & Partners expects that ML and alternative data will transform the competitive landscape for banking tremendously and relatively fast. Therefore, banks should become more agile and keep their finger on the pulse of these recent developments.