In the evolving landscape of climate risk and financial regulation, few topics have provoked as much discussion as the integration of environmental data into internal rating-based (IRB) models.
The head of Internal Ratings-Based modelling at a leading UK building society told CeFPro’s 2025 Climate Stress Testing Europe conference that his organization was actively exploring the benefits of adding energy performance certification to its climate risk management strategy.
The unique nature of its mutual model and robust capital position stands the organization - one of the largest of its kind in Europe – apart from its key competitors.
With a mortgage portfolio supported by decades of historical data, including rating-level information back to 2008, the society is well-positioned to interrogate the implications of climate risk for capital models.
The executive’s presentation began by acknowledging the regulatory tensions surrounding climate risk integration into Pillar 1 capital requirements.
In the speaker's view, embedding such risks into IRB frameworks is not only misaligned with the regulatory timeline but also incompatible with the nature of existing IRB modeling methodologies.
Developing or adjusting IRB models is a multi-year endeavor, as demonstrated by the UK’s hybrid model journey, which started in 2017 and has only recently seen a second institution gain approval.
A key part of the discussion focused on the treatment of Energy Performance Certificate (EPC) data.
While previous sessions highlighted the scarcity and inconsistency of EPCs, this perspective acknowledged those limitations but moved further to investigate the data's surprising empirical impact.
Merging internal mortgage and default data with EPC records sourced from a national housing association created a robust sample for analysis.
After rigorous cleaning and adjustment, a strong statistical relationship emerged between EPC ratings and default probability.
This correlation, while notable, came with caveats. It remained unclear whether EPCs causally influence credit performance or whether they act as proxies for other unobserved borrower characteristics such as income or wealth.
For instance, better-rated homes might simply be more common among wealthier borrowers, who inherently present lower credit risk.
Even after accounting for numerous potential confounding variables, the residual impact of EPCs on predicted default remained statistically significant.
Crucially, the speaker said, when EPC data was introduced into existing IRB PD models - already built with rigorous selection processes that filter over a thousand variables down to a small core set - the models demonstrated improved performance.
This finding challenged conventional organizational assumptions that the inclusion of climate-related data into highly mature IRB frameworks would add little value.
