The Intelligent Quarterly from the publishers of The Insurance Insider

Winter 2017 / 2018

Search archive

 

Poles apart

Lucy Jones

As the Caribbean islands of Dominica and Puerto Rico lay in ruins following Hurricane Maria, the industry's leading modelling agencies issued two vastly different views of the event.

RMS estimated insured losses from Hurricane Maria at $15bn-$30bn, while AIR Worldwide put out a range of $40bn-$85bn, of which losses in Puerto Rico were pegged at $35bn-$75bn. AIR later revised its estimate to $27bn-$48bn, of which $25bn-$43bn related to Puerto Rico.

More than two months on, while modellers stand by their numbers, the industry at large is asking why the estimates are so diverse, whether models are delivering value for money given the huge loss ranges and whether pricing reflects the model uncertainty that has emerged.

The disparities in industry loss estimates for Puerto Rico partly come down to a lack of information following the event.

Weather stations which measure wind speed failed during Maria, which meant both AIR and RMS had to fill in the gaps.

But the explanation can also be traced back to the inputs that risk modellers begin with - their assumptions on what properties are at risk from a storm.

Hazard risk
Hurricane Maria has revealed vastly different assumptions taken by AIR and RMS on the vulnerability of buildings in Puerto Rico.

AIR assumed wind damage would have occurred to the vulnerable upper stories of buildings in Puerto Rico, which are usually made of wood. However, RMS took the view that the island's bunker-style buildings would have been able to successfully resist the storm.

On the commercial side, Michael Young, head of Americas climate risk modelling at RMS, says that local pharmaceutical facilities, which make up half the company's exposure base for Puerto Rico, are also built to be incredibly resilient.

"The losses we expect to come from that particular sector tend to be quite low," he says.

Puerto Rico's manufacturing industry has shrunk by a third over the last 10 years, so neglecting to include this trend could lead to exaggerated losses, RMS notes.

The modelling agencies also appear to differ in their views of how quickly the island will recover and how much that recovery will cost.

On a reconnaissance trip to the island shortly after the hurricane, an RMS team found the Puerto Rican capital San Juan mostly operational. Businesses were using fuel-powered generators as part of hurricane preparedness. New hotels had performed well in withstanding the hurricane and were open.

AIR says the cost of Puerto Rico getting back on its feet could be significant, however.

"After an event of this size, cost of repair and cost of labour is not the same as before the event, as you have shortages on the island," says Cagdas Kafali, senior vice president in AIR's research and modelling group.

Some 30 percent of AIR's uppermost insured loss estimate of $43bn for Puerto Rico is attributed to this so-called "demand surge".

Converging views
One might expect that in the insurance market's exposure hotspot of the US the risk models would demonstrate greater convergence - and, indeed, this was the case for Hurricane Irma's impact on Florida.

AIR published an estimate of $32bn-$50bn for Irma, including $25bn-$35bn of US insured losses. The figures from RMS were in the same ballpark, with insured losses estimated at $35bn-$55bn, of which US private losses constituted $22.5bn-$29.5bn.

Claire Souch, director of cat risk modelling consultancy AWHA Consulting, says that the high frequency of big US hurricane events in the past 10 years has helped to develop modelled research. But for a big US flood event, such as Harvey, there is less past experience and therefore more uncertainty, she says.

For Harvey, AIR has not provided an insured loss figure for flooding but has instead issued an overall insurable loss figure of $55bn-$65bn.

RMS has given a figure of $25bn-$35bn, which includes $7bn-$10bn incurred by the US government-backed National Flood Insurance Program, plus $18bn-$25bn of commercial insured losses. RMS does not have a US flood model (it will be launched next year), whereas AIR does.

Chaos modelling
Modelling involves determining the hazard, vulnerability, exposure and locations of risk impacted by an event. However, according to Andrew Castaldi, head of catastrophe perils Americas at Swiss Re, there should be a fifth box - the modelling of unexpected components of an event, which he terms "chaos modelling".

"For example, engineering studies are looking at a particular building and how it will react within a wind field, but in reality you would have many other structures around it with, for example, gravel in the roof which turns into debris," he adds. "How is your risk impacted by an event when it is surrounded by a community of other risks?"

Some might say that's underwriting judgement but others might say it belongs in a model, he continues.

Castaldi goes as far as to say that by not looking at the exaggerations that might occur insurers could be leaving themselves in jeopardy.

"Maybe ratings agencies and regulators will start looking at these experiences and will say, 'Maybe we should look at a modelled loss, plus a certain percent to include these impacts'," he says.

Market view
The disparities in model outlooks have certainly been a cause for concern across the market.

"If a Category 4 on Puerto Rico can cause up to $80bn of loss, what would a Category 5 in Miami/Tri-County really cost?" one insurance-linked securities market source questions.

"Are licensees who use the model loading their pricing sufficiently for such huge model uncertainty?" the source adds.

Another source says the cost of taking on risk should go up, given the model uncertainty recent events have brought to light.

The uncertainty of models has occupied a less prominent position in decision makers' minds until now, but that needs to change, the source says.

But even though models differ and the ranges for individual events can be vast, insurers still reiterate their basic value to risk takers.

Having a range of views is useful and necessary, says Shree Khare, group head of catastrophe research at Hiscox.

The reinsurer's process for assessing a large loss includes taking into account the judgement of its underwriters and catastrophe modelling teams, market estimates of the industry loss which give insight into potential losses to individual carriers, specific modelled events from its modelling partners and knowledge of specific risk losses.

"Given the multiple sources of uncertainty in loss estimation, I'm not surprised to see a large range of losses from any particular vendor," he says. "Furthermore, I'm not surprised to see disparities between model vendors." However, Khare adds, it would also be useful to understand the drivers behind each of the vendor's given ranges.

Harvey, Irma and Maria have provided Hiscox and others in the industry with a huge opportunity to update their view of risk and learn and work with customers to improve their understanding of their exposures.

"Harvey is a good example of this, given [that] it stalled over Houston and caused an unprecedented amount of flooding," says Khare.

Vendor models tend to be a sanity check, says Castaldi. "If someone is between $5bn and $15bn and we're in that $8bn range, then we feel comfortable but if we're in a $30bn then we will have to look at what went wrong.

"No model is ever going to be perfect or exact but it's going to be enough to protect you from going bankrupt," he concludes.


This article was published as part of issue Winter 2017

Euromoney Trading Limited - 3rd Floor, 41 Eastcheap, London, EC3M 1DT, United Kingdom. The content of this website is copyright of Euromoney Trading Limited 2018. All rights reserved Euromoney Trading Limited actively monitors usage of our website and products and reserves the right to terminate accounts if abuse occurs.

Π