It's a given that catastrophe models play a key role in helping insurers estimate the losses that could result from cat events such as earthquakes and hurricanes. But these models don't work in isolation. To get even more accurate assessments, insurers increasingly are using them in conjunction with other models-commercial or homegrown-and with data from other sources such as satellite and radar systems.
One recent trend that's spurring the use of multiple cat models is the growing pressure from regulators and ratings agencies in the wake of widespread disaster-related losses for insurers and reinsurers.
For example, ratings firm Standard & Poor's in September 2011 reiterated its call for the use of multiple cat models by insurers. To avoid instances of "model shopping," the ratings firm says it prefers that insurers and reinsurers use models from at least two of the big three modeling firms—Risk Management Solutions (RMS), AIR Worldwide and EQECAT—when assessing catastrophe risk for natural peril catastrophe bonds. "Loss estimates from an event, or the likelihood of an event, can differ significantly between modeling agencies, according to how the data is interpreted," the S&P notes. "We therefore consider that a multiple-model approach would give existing and potential investors a better perspective on the range of potential outcomes."
Sibylle Steimen, head of cat Risk Management at Allianz Re, the internal reinsurer of Munich, Germany-based Allianz group, says the firm uses models from multiple sources. Steimen says in addition to a primary model from one of the big three, the company employs "second opinion" models from another vendor as well as models from brokers and those the company has developed on its own.
"We try to have a broad view," she says. Creating internal models to measure risks such as damage from hail storms and floods gives the firm an even broader view of catastrophe loss data, she says. "If you're exclusively relying on brokers' work [and commercial models] you've lost a bit of control over cat modeling and analysis," Steimen says.
Overall, the frequent use of cat modeling gives Allianz a long-term baseline for catastrophic events. "It's one way to measure the situation over time and compare [catastrophic events and how they affect the group] over a longer period of time," Steimen says.
Before cat models are used by the firm they are validated by in-house experts who have the right scientific expertise. If required, models are adjusted to better reflect the company's own catastrophe exposure, Steimen says.
A Broader Process
Likewise, Matthew Jones, head of catastrophe management, global underwriting at the Schaumburg, Ill.-based Zurich North America says its important for insurers to understand both the scope and the limitations of modeling tools. Jones says the company uses modeling products to inform its "Zurich View" of catastrophe risk, which then feeds into many areas within the company-from risk appetite assessment to risk-based capital calculation to primary pricing. "The main benefit of such tools is that they provide an exposure- based view of risk, which is needed given the low frequency of the type of events we are trying to understand," Jones says. "However we never use such tools without adjustment and do not treat them as 'black boxes.'"
Establishing the Zurich View of risk always involves using multi-model output together with information from external experts, as well as the company's own claims experience and expertise. Zurich uses CAT model output from all the big three vendors, as well as other sources, Miller says.
Farmers Insurance Group is also leveraging multiple models to determine the frequency and severity of losses resulting from specific natural perils in specific geographic locations by licensing models from AIR and RMS, says Derek Gullage, VP, reinsurance at Farmers. Gullage says that Farmers has found the models from both vendors to have strengths and weaknesses.
"We are not a company [that] takes a single output and builds a business plan," he says. "Rather, we use these outputs as one point of view, but continue to have a more robust view of natural catastrophe risk that is not based on a single data point from a single model. Using both [enables] us to minimize the effect of any one model change."
Like Steimen and Jones, Gullage is quick to stress that models are only part of a broader process. "We should point out that the models are only a tool, and we use this tool in concert with our own experience, knowledge and history of losses," he says.
Aon Benfield, a global reinsurance intermediary and capital advisor firm that helps its insurance firm clients evaluate CAT models, typically encourages the use of multi-model views, says Paul Miller, head of international catastrophe management at London-based Aon Benfield.
"That said, a majority of our clients [generally] use a core model, then may apply adjustments depending on their comfort with the scientific robustness of the model and whether events have come along that prove the model to be inaccurate," Miller says.
Over the past year events such as the earthquake and tsunami in Japan "have proved that [individual] CAT models are not necessarily providing as adequate a view of risks as may have previously been thought," Miller says. "That has led clients to consider whether it's appropriate to be using some sort of blended model approach or to continue to use a core model but further develop their adjustment factors."
Miller says he's not sure about the extra benefits of blending models. "It's a struggle to truly blend the models by trying to match up the event sets," he says. "The simulated events within each of the model solutions are different, so blending [models] isn't as easy as it sounds."
A better approach might be to use a core model but have a more thorough understanding of its strengths and weaknesses, and keep aware of all other models in order to continually monitor the strengths and weaknesses of each as the models are updated, Miller says.
Additional Augmentation
In addition to using models from multiple vendors, insurers are augmenting models with data from other sources whenever it makes sense.
Zurich North America augments models with additional data, Jones says. "The main benefit of supplementing CAT model output with extra data is that the additional data can be more up-to-date, more relevant or more accurate for the intended usage," he says.
An example of this is assessing flood risk. "Probabilistic event-based flood models are not available from model vendors for many regions, but often flood maps are available and can be used together with other information to form a view of flood risk," Jones says.
Aon Benfield's clients are relying more on supplemental information such as geospatial data when using CAT models, Miller says. The firm has developed a geospatial tool that allows clients to visualize their risks more accurately and to look at data sets such as flood plain information that they can bring into modeling systems to view potential outcomes and risks.
The use of additional data is "something [firms] should be getting a handle on to help form an opinion, rather than purely relying on CAT models," Miller says. "In the United States [in particular] there's a wealth of claims data than can be used by clients to see what the models are missing."
While Allianz is not currently augmenting CAT models with other data, the firm has plans to do so in the future [see sidebar] to provide additional data when needed, Steimen says.
"You will always have certain gaps in the catastrophe landscape; maybe there are lines of business or certain perils that are hard to model," Steimen says. "I see a need for other ways to tackle these kinds [of gaps] and it could be we make use of satellite imaging or other tools."
Bob Violino is a business editor and writer based in New York.
