European catastrophe specialists are warning that even as models get more powerful, they still fall short on some of the perils and data that are driving the biggest loss surprises – a message Canadian carriers would do well to hear.
Speaking on a recent BrightTALK webinar on exposure and cat modelling, panellists noted that insured natural catastrophe losses are growing about 5-7% a year globally, and that Europe is on a similar trajectory. Crucially, that growth is outpacing GDP, meaning the catastrophe burden on the insurance system is rising faster than the economy that supports it.
Much of the increase is explainable: there is simply more value at risk, concentrated in more exposed locations, and it costs more to repair or replace damaged assets. Insurance penetration has also risen in some markets, closing part of the protection gap. But there is still a stubborn “unexplained” component, particularly for certain perils.
Jessica Turner, a catastrophe and exposure management professional, said model science has clearly improved: more computing power, finer resolution, larger event sets. Yet when you compare modelled results with actual claims, some uncomfortable gaps remain.
In northern Europe, for example, “from a modelled perspective, windstorm is the number one driver of loss,” Turner said, with flood second. For many insurers’ real-world books, the order is reversed – flood has delivered the larger losses. That discrepancy forces companies to confront a hard question: is the model’s long-term average view right, or has a 20‑year lull in wind activity lulled the industry into a new normal that might not last?
Vendors offer options to calibrate risk on shorter or longer time windows – say, the last 25 years versus a longer climatological period – but there is no single correct answer. “As risk carriers and as cat modellers, you’re going to have to engage with that topic and make a decision about what your view of risk is for your company,” Turner said.
She sees similar issues on the flood side. Vendor models, she argued, do a good job where river flood risk is obvious – floodplains near major waterways. They struggle more with surface water and pluvial flooding, especially in dense urban areas, and mostly don’t capture how long water sits in a building, even though duration is critical for damage to walls, wiring and contents.
Exposure data makes that problem worse. While most insurers track building values and locations reasonably well, Turner noted that they rarely capture the true value of contents: how expensive the kitchen is, the quality of floors and finishes, the cost of high‑end furnishings.
“That’s not a model problem,” she said. “That’s a data problem.”
Hail and severe convective storm (SCS) are another pain point. Turner welcomed the fact that European hail is now explicitly modelled, but cautioned that these are essentially first‑generation tools calibrated on historic periods. Research suggests Europe – particularly regions like Italy’s Po Valley – is already seeing a pronounced climate signal in hail.
“There seems to be a climate change signal in hail in Europe… more than in any place in the world,” she said, questioning whether models tuned to the past are adequately capturing future severity. Her advice: users should be “thoughtful about whether or not there’s enough severity in them.”
For Iwan Stalder, head of group accumulation management at Zurich, the takeaway is that insurers need to treat vendor models as a starting point, not an answer.
“We see that the models are a good starting point, but they are essentially not covering full sources of losses,” he said, adding that Zurich runs a formal model‑validation process and adjusts its internal view of risk where science and experience diverge from the standard output.
Stalder emphasised two disciplines: staying close to scientific networks – so new hazard maps and climate research can be reflected quickly in underwriting views – and building robust exposure‑data governance. Zurich has spent more than a decade creating a single, cross‑peril exposure datastore, he said, so analytics teams can focus on concentrations and drivers of loss rather than constant data cleansing.
Validation, in his view, must extend into exposure itself: checking not just locations, but total sums insured, splits between building and contents, and key construction details. With that foundation, he said, it becomes much easier to run portfolio‑wide analytics, identify cross‑peril hotspots and react quickly when new hazard information emerges.
For Canadian carriers, the parallels are almost obvious. River flood, surface water, hail, wildfire and earthquake all feature in domestic portfolios, and CAT claims are already dominated by so‑called “secondary perils”.
Owning the same vendor models as everyone else is no longer enough. Competitive advantage (and resilience) will come from how well insurers challenge those tools, improve their exposure data, and form their own view of risk before the next “unmodelled” loss shows up in the claims triangle.
