The industry’s confidence in catastrophe models was tested in 2005 after Hurricane Katrina defied loss expectations. In our latest survey, Global Reinsurance asked readers how they feel about the models now. Helen Yates presents the results.

Three years ago the world watched in shock and awe when Hurricane Katrina did her best to write-off an entire US city. The loss estimates rose and insurance companies began the long and arduous task of dealing with thousands of claims. Many insurers and reinsurers quickly realised their exposures in Louisiana were much greater than they had anticipated. Fingers were pointed at the catastrophe modelling agencies for failing to predict the magnitude of the loss.

Confidence, it seems, has been restored in the three years since Hurricanes Katrina, Rita and Wilma (KRW). Whereas 69% of our respondents said they did not have confidence in the catastrophe models immediately after Katrina, only 47% still lack confidence today. “The models have clearly improved,” says Paul VanderMarck, chief products officer at RMS. “The two factors that will have the biggest positive impact on confidence will be improvements in portfolio data quality, which is currently a significant obstacle to optimal usage of the models, and actual events that demonstrate the advances we’ve made.”

Catastrophe models clearly play a crucial role in helping insurers and reinsurers better manage their exposures. Seventy-one percent of those surveyed said the models had a moderate, significant or heavy influence on their underwriting decisions. Yet half said a conscientious underwriter should assign little to no responsibility to the catastrophe models for making the right decisions. “Underwriters’ decisions are governed so much by the models and yet they should also take responsibility for their decisions,” explained one respondent.

Given a range of model outputs, it could be tempting for underwriters to opt for those outputs that allow them to write a greater proportion of risks. We asked how underwriters should deal with this conundrum, particularly in a softening market. Most said underwriters should remain disciplined in all market conditions. “You are kidding right?” was one response. “Sound underwriting begins with the precept of how much surplus is the enterprise willing to lose and then charging an adequate price to yield an adequate return for the variability of risk assumed – across the life of the enterprise. NOT what is the other guy doing in this market and can I get a piece of his pie without losing all of mine!” Whether this happens in practice is another thing, as another respondent conceded. “Conservatism is best, but unlikely to be adopted in a softening market.”

Significantly, a number of respondents said the increased emphasis on enterprise risk management (ERM) had increased the necessity and significance of cat modelling. Given its increasing importance, it is perhaps unsurprising that respondents had varied and strong views about cat modelling. However, most were in agreement that blame had been too one-sided in the aftermath of Katrina. “While cat models maybe did not fare too well in KRW, anyone without a model probably fared even much worse,” said one.

[Katrina was] “a combination of cat modellers overstating the reliability of their models, and users naively believing them,” explained one respondent, while another simply said: “A lot of workmen blaming the tools”. “To a certain extent, cat models ARE just guesswork,” said another, “but underwriters should have realised this all along and should never have been expecting the models to ‘predict the size of an event’.”

“Seventy-one percent of those surveyed said the models had a moderate, significant or heavy influence on their underwriting decisions

“Some industry observers make the mistake of believing that models will eventually replace underwriters,” comments VanderMarck. “This couldn’t be further from the truth. Underwriters absolutely need to leverage models to help develop their intuition about risk and to assess individual insureds, but they also need to take ultimate responsibility for the decisions they make about taking risk.”

So where does this leave us? Since 2005, various revisions have been made to the hurricane catastrophe models. AIR Worldwide, RMS?and Eqecat have released new “medium-term” models (with five-year horizons) reflecting a period of heightened hurricane frequency and intensity that most experts agree we are now in. Many survey respondents said they think the models are improving. “Looking at medium-term view outputs from AIR and RMS for industry loss events, they seem fairly conservative,” said one. “The problem is with underwriters playing around with the models in order to make their deals look more attractive. Data quality has also been poor but it appears that it is on an improving track.”

Despite perceived improvements, a few were still concerned about model limitations – less linked to the model science and more to the quality of data fed into them. “Every event improves understanding however the events are too infrequent to be a great predictor, especially for earthquakes,” warned one.

One respondent called the medium-term models the “latest fad” while another argued that the fact the models were altered after every major loss indicates that they weren’t accurate before the loss. “We would all prefer the models to be as stable as possible, however everyone in the catastrophe business needs to recognise that we will always learn things from future events,” responds VanderMarck. “Over time most lessons will become incremental refinements, but sometimes there will be surprises that are inconsistent with prior science and observations and that will result in larger changes to the models.”

We asked those surveyed how the industry could work better with the catastrophe modellers to improve access to relevant data. Many thought it was simply a case of finding the right motivator. “When the insured has a large financial stake in the outcome then they will take the information gathering seriously,” said one. “Market premiums are the drivers.” The benefits must be seen to outweigh the costs, most agreed. And if financial incentives fail to work, fines could be used, suggested one respondent. “Reinsurers could severely penalise companies which fail to provide accurate data through increased ceded costs.”

Politics and modelling

“The increased emphasis on enterprise risk management has increased the necessity and significance of cat modelling

In the US there is increasingly a political element to catastrophe modelling. Florida has a regulatory commission that examines catastrophe models. Instead of a medium-term view, it favours an older model, which does not reflect a period of heightened storm activity. This helps to keep homeowners’ premiums down.

Unsurprisingly, some respondents had strong views about the role of politics in insurance and catastrophe modelling. “You must be referring to [Florida governor] Charlie Crist,” wrote one respondent. “The man should be impeached and sent to re-education camp on basic economics.”

“Insurers should drop the pretence and be market players,” was another impassioned plea. “If the insurer’s underwriting judgment makes the state unprofitable, then pull out of the market. Better to have no market share than one that will cost you a decade of hard-won profit in one known and anticipated event.”

The majority of those surveyed (51%) saw greater regulatory scrutiny of the catastrophe models as a positive thing. Of those, 50% believe regulation of cat models should be done at a national level, 25% opted for state regulation, while 6% thought it required a combination of both. “In practice, federal regulation in the US would be much more efficient than individual state-level approaches and less prone to political interference,” says VanderMarck.