No silver bullet for cat model uncertainty; new models muddy landscape

There is “no consistent silver bullet” for explaining why catastrophe models produce different loss estimates. Equally, the growing availability of alternative models is raising fresh questions about how ILS investors should interpret competing views of risk.

That was a key question discussed by a panel of experts speaking at at ILS Bermuda Convergence, held October 12-14.

The panel, moderated by Dr Joanna Syroka, director of new markets at Fermat Capital Management, comprised Karen Clark, co-founder and CEO of Karen Clark & Company; Robert Silva, risk analytics expert and lead actuary of Zesty AI; and John Wardman, CCO of MaxInfo.

The executives examined the sources of model variability, the challenges posed by unknown property characteristics and whether greater access to multiple models necessarily leads to better underwriting decisions.

“There is no consistent silver bullet that we can point to,” Wardman said, explaining that discrepancies must be traced through the constituent components of catastrophe models, including hazard, vulnerability and loss calculations.

“How do the models deal with unknown characteristics of buildings? That in and of itself can create very big loss differences,” he said as an example.

The sources of disagreement can also vary considerably between portfolios, making it difficult to establish a consistent explanation for divergent results.

For investors, improved property-level information could help address some of that uncertainty.

Silva argued that more comprehensive and consistent data on building characteristics can have a substantial impact on modelled outcomes, particularly where differences in probable maximum loss (PML) estimates affect catastrophe bond exposures.

“We're really here to provide the kinds of data that cat models have been asking for a couple of decades now, and it is very accurate and comprehensive data,” Silva said. The impact of this data is that “you should see a much smaller range of probable outcomes from different loss events.

“If you lower a PML by 20%, that really could be the difference between your cat bonds paying nothing out and paying quite a bit of the capital out.”

However, the discussion exposed contrasting views on whether the proliferation of catastrophe models is itself beneficial.

Clark questioned the value of continually expanding the number of models available to insurers and investors, particularly given the additional work required to assess competing methodologies.

“I don’t actually think there’s a desire for 10 hurricane models. Because then it actually adds to the cost because now you have to validate and evaluate 10 models instead of two or one,” she said.

She argued that confidence in the underlying models should take precedence over the sheer number of available perspectives. “Rather than more models, I think we want more credible, reliable models.”

That position drew a challenge from Wardman, who argued that differences between models can reveal uncertainty that might otherwise remain hidden when investors rely on a single view.

“The nice thing, though, in our in our world, at least from magnify perspective, is that you can rapidly get those evaluations, rapidly come to a sense of confidence in each one of those models, and then. When you see the spread of different results, really, you’re just looking at more of that uncertainty space that the one cat model collapses into a single number anyway,” he said.

“From our philosophical perspective, we would argue more models, more information, to then be able to make a better decision.”

Although platforms such as Oasis and the Verisk Model Exchange are reducing technical barriers to accessing alternative models, the panel also identified obstacles to their adoption within the transaction process.

Clark argued that established broker and deal-management practices can prevent alternative models from being incorporated into catastrophe bond offerings, even where insurers already use those models for underwriting and exposure management.

“I think one of the things is investors need to speak up that [they] do value alternative views, second-generation models,” she said.

With model differences potentially influencing loss estimates, pricing and investment decisions, the debate left investors confronting a question that additional technology cannot answer on its own: whether greater confidence comes from narrowing the range of modelling assumptions or understanding more fully the uncertainty those assumptions reveal.

Did you get value from this story? Sign up to our free newsletters and get stories like this sent straight to your inbox.