The blind spots inside the cat model
As ILS investors move into lower reinsurance layers and offer more aggregate covers, Robert Silva argues that information about the individual properties behind a portfolio will become increasingly important to understanding the risk they are taking.
For Silva, ACAS, risk analytics expert and lead actuary at ZestyAI, the next advancement needed in catastrophe risk analysis is not necessarily another model. It is improving what goes into the models investors already use.
“The disconnect isn’t really in the cat models. It’s in what goes into them,” Silva told Bermuda:Re+ILS ahead of ILS Bermuda Convergence, held 12-14 October. “The models are built to adjust for structural vulnerability, but only if someone tells them what the structure is, and often nobody does.”
Catastrophe models can account for characteristics such as roof geometry, age and condition, but missing secondary modifiers require assumptions to fill the gaps. Silva pointed to one carrier filing in which roof geometry was known for less than 12% of policies in one line of business.
The portfolio-level issue, he argued, emerges when those assumptions systematically differ from the characteristics of the underlying book.
“Random errors on single buildings wash out across a portfolio if the portfolio aligns with those assumptions,” Silva said. “A systematic gap doesn’t, and that’s where the problems lie.”
The significance can also vary by peril. Silva sees some of the biggest potential gaps in severe convective storms, where he said roof age and condition “drive much of the loss” but are among the fields most often missing or out of date.
Wildfire presents another example, with characteristics of both the building and its immediate surroundings potentially affecting vulnerability. “The building and what's right around it, like roof type, vegetation and debris, can determine which house burns and which doesn’t,” he said.
Silva argues that aerial imagery and computer vision can fill some of those missing fields by identifying characteristics such as roof material, age and condition, vegetation and yard debris at individual-property level.
Crucially for ILS investors, his argument is not that this technology needs to create another competing view of catastrophe risk. Instead, property-level information can be mapped onto the secondary modifiers that established catastrophe models already use.
“We can let that kind of loss modelling stay with the cat models investors already use and have already vetted,” he said. “We supply the data those models need, so the question becomes simpler: is the data right?”
That question could become more important as the ILS market takes on exposures where frequency, rather than only remote tail events, plays a greater role.
“The market is already moving as ILS investors take on lower reinsurance tower layers and offer more aggregate covers. Losses in these higher-frequency layers and covers become increasingly driven by roofs, condition and other property specific characteristics that cat models know cause losses,” Silva said.
Greater property level visibility could also give investors another means of interrogating the quality of portfolios presented to them. Where underlying addresses are available, Silva said applying the same methodology across cedants could allow investors to compare book quality on a consistent basis.
Even where addresses are unavailable, he believes investors can start asking a revealing question: “What share of the exposure has verified modifiers, and what share was left to the model's assumptions?”
That could shift part of the due-diligence conversation away from competing modelled views and towards a more fundamental question: how much is actually known about the exposure entering the model?
It also underpins Silva’s view of what could look outdated in ILS catastrophe risk assessment five years from now: “Taking on frequency-driven risk without looking at the buildings.”
“Writing that cover without a consistent way to tell one cedant’s book from another will look outdated. The investors who build that method will price each cedant on its own book,” he said.
For an ILS market accustomed to interrogating different views of catastrophe risk, the next source of differentiation might therefore sit one level beneath the models themselves: the buildings they are being asked to model.
Did you get value from this story? Sign up to our free newsletters and get stories like this sent straight to your inbox.
