Physical climate risk in credit models
How catastrophe-model output should cross the interface into bank credit systems — and what today’s practice loses on the way.
- From expected loss to concentration. Measuring what the industry-standard interface — a single average annual loss per exposure — withholds from credit models: nothing at the rating-materiality scale, but 28–41% of 1-in-100-year expected credit loss, nearly all of it unexpected loss concentrated in a single name.
- Calibrating the margin of conservatism. Replacing judgement-based climate add-ons to PD and LGD with a number derived from damage-function uncertainty, through an admissibility protocol for vulnerability curves and a declared quantile rule that maps to prudential MoC categories.
- What to request from a catastrophe model. A methodology for credit risk managers to select, transform and qualify catastrophe-model outputs without access to model internals, with each resulting credit metric labelled by the mathematical guarantee it carries.
- The value of knowing the net loss curve. Quantifying when, and by how much, knowledge of the disaster-net credit loss distribution improves a bank’s retention and capital decisions, and how that value shifts when external protections such as sovereign disaster finance change.