The traditional catastrophe model remains a powerful framework
Catastrophe models generally combine three core components: hazard, exposure and vulnerability. Hazard describes the frequency and intensity of events. Exposure describes the assets, people or economic values in harm's way. Vulnerability translates hazard intensity into damage. A financial module then converts damage into insured or economic loss.
The Norges Bank Investment Management material in the source compilation explains how catastrophe models use stochastic event sets to simulate disasters that are physically plausible even if they have not occurred in the historical record. This has made catastrophe modelling central to insurance pricing, reinsurance, capital and accumulation management.
Climate change weakens the stationarity assumption
Historical calibration often assumes that the statistical relationship between past hazard and future hazard is sufficiently stable. Climate change challenges that assumption. If the frequency, intensity, spatial footprint or seasonality of a peril changes, a model calibrated mainly to historical observations can become systematically biased.
The source material explicitly notes that as climate change alters the frequency and intensity of many perils, the predictive power of historically calibrated catastrophe models may weaken. Climate-conditioned models address this by incorporating forward-looking climate information into hazard modelling.
Exposure and vulnerability are also non-stationary
Non-stationarity is not limited to hazard. Exposure changes as populations move, cities grow, supply chains concentrate and assets are built in vulnerable locations. Vulnerability changes through building standards, flood defence, adaptation investment, ageing infrastructure, insurance behaviour and technological change.
A model can therefore be wrong even if the hazard component is correct. For example, the same flood depth may produce different losses after a region improves flood protection or after critical infrastructure becomes more interconnected. Climate risk modelling should treat hazard, exposure and vulnerability as evolving systems rather than fixed inputs.
Compound and cascading events create a second challenge
Traditional models are often built by peril. Real losses can cross peril boundaries. Heat, drought and wildfire can occur as a compound sequence. Storm and flood can interact. A physical event can damage infrastructure, interrupt suppliers, create liquidity needs and cause business interruption far from the original location.
The NGFS short-term scenario material notes the lack of statistical models for the full joint distribution of compound hazards and therefore uses physically coherent storylines for severe events. The Bayesian and causal-network paper in the compilation makes a related point: simple branch structures can become brittle when feedback loops and multi-causal failure dominate.
Validation must change with the climate
Traditional backtesting remains necessary but cannot be sufficient when the data-generating process is changing. A model may fit the past well and still be poorly calibrated for the future. Validation should therefore combine historical performance with forward-looking diagnostics.
Useful tests include sensitivity to climate-conditioned hazard frequency, alternative event sets, tail calibration, regional bias, vulnerability assumptions, exposure growth, adaptation and dependency structures. Benchmarking against multiple models can help identify model-form risk, but common dependencies among models should be recognised.
The decision should determine the conditioning
Not every application needs the same degree of climate adjustment. A one-year insurance policy may require a different treatment from a 30-year mortgage, an infrastructure investment or a long-tail liability. The source material notes that near-term adjustments may be modest for some annually renewing insurance products, while long-duration decisions are more sensitive to climate trends.
The correct approach is therefore horizon-specific and use-specific. Climate-conditioned catastrophe modelling should improve the model where the future hazard distribution is relevant to the decision, while avoiding complexity that does not change pricing, capital, underwriting or risk appetite.
Conclusion
The key requirement is decision usefulness. Climate analysis adds value when it improves risk identification, pricing, capital allocation, portfolio management or governance, while making uncertainty and model limitations explicit.
Sources
Norges Bank Investment Management. Economic impacts and pricing of climate risk. Discussion Note, 2026. https://www.nbim.no
Network for Greening the Financial System. Short-Term Scenarios. https://www.ngfs.net
Velasco-Reyes, E.R. and Pui, A. Rethinking Uncertainty: Why Disaster and Climate Risk Models Must Move Beyond Logic Trees. 2026.
Disclaimer: This article is professional risk-management analysis and is not investment, legal or regulatory advice.