Catastrophe pricing models should be reassessed when measurable performance deterioration or fundamental shifts in hazard drivers occur. Recognizing model drift requires both statistical monitoring and attention to evolving external science and socioeconomic conditions.
Signals from model performance and climate science
Regular back-testing that tracks calibration and predictive accuracy will reveal when a model no longer reflects observed losses. Empirical indicators include systematic underprediction of losses, widening residuals, or persistent bias by region or peril. Kerry Emanuel at the Massachusetts Institute of Technology has documented increasing tropical cyclone destructive potential linked to warmer oceans, a scientific signal that should trigger model review. Attribution studies led by Friederike Otto at Imperial College London show that human-driven climate change has already altered some extreme event probabilities, meaning hazard frequency and intensity inputs can change faster than long-standing parameter assumptions.
Governance triggers and external contextual changes
Beyond diagnostics, insurers should recognize drift when external information changes materially: updated climate assessments by the Intergovernmental Panel on Climate Change, new regional sea-level projections, altered land-use patterns from urban expansion, or abrupt regulatory directives on model validation. Social and cultural factors matter: migration into higher-risk coastal or wildland-urban interface zones, affordability pressures, and shifting building codes change the exposure base and therefore the relationship between hazard and loss. Nuanced judgment is required when scientific uncertainty is high but directional change is clear.
Causes of model drift include inadequate recalibration cadence, reliance on stationary hazard assumptions, and delays in incorporating improved physical science or exposure data. Consequences of failing to act manifest as mispriced risk, accumulation of unanticipated capital strain, and potential solvency stress after major events. On the societal level, persistent underpricing can encourage maladaptive development in hazard-prone areas, while abrupt repricing without clear communication can reduce insurance availability for vulnerable communities.
Practical thresholds for action combine statistical and contextual tests: statistically significant degradation in out-of-sample performance, publication of new vetted hazard science altering key input distributions, or material changes in exposure or vulnerability data. Insurers should embed continuous monitoring, clear governance for model update decisions, and stakeholder communication protocols. Reinsurance strategies and capital buffers should reflect both the uncertainty around drift and the social consequences of coverage changes, ensuring pricing remains actuarially sound and societally responsible.