Banks are required to estimate credit risk in a forward-looking manner, particularly under the IFRS 9 regulatory framework. A key component of this process is the calculation of Expected Credit Losses, where default probabilities need to reflect current and expected economic conditions rather than long-term averages.
In practice, this requires a transition from Through-the-Cycle probabilities of default to Point-in-Time estimates. A crucial step in this process is identifying which macroeconomic variables are most relevant for explaining changes in default rates. As there is no prescribed method for selecting these variables, different approaches are used across institutions.
At Amsshare, we work with these types of models in practice and therefore assess not only how they are implemented, but also how they can be improved. This paper presents a structured and data-driven approach to selecting macroeconomic variables, applied to European corporate default data.
Summary
This paper examines how banks can incorporate macroeconomic information into credit risk models under the IFRS 9 framework, with a specific focus on the transition from Through-the-Cycle to Point-in-Time probabilities of default.
A key challenge in this process is the selection of macroeconomic variables that explain movements in default rates. Since no standard methodology is prescribed, this paper introduces a data-driven approach based on the LASSO BIC method. This technique automatically selects the most relevant variables while balancing model complexity and explanatory power.
The results show that certain macroeconomic variables, such as interest rates and credit spreads, play a consistent role in explaining default rates over time. At the same time, the analysis demonstrates that the importance of these variables can change, highlighting the need for periodic re-evaluation rather than relying on a fixed model specification.
These findings are directly relevant for banks and financial institutions that are required to implement IFRS 9 models. They show that combining regulatory requirements with robust variable selection techniques can improve the reliability of credit risk estimates, while also providing a more structured and transparent modelling approach.
