Free Trial Corporate Access
Reactions - Financial Intelligence for the Global Insurance Market
Reactions Copying and distributing are prohibited without permission of the publisher
Email a friend
  • Please enter a maximum of 5 recipients. Use ; to separate more than one email address.

FEATURE: Is the industry too reliant on models?

17 June 2010

Risk modellers are preparing many updates to their models worldwide, which could have an effect on pricing. But are insurance and reinsurance firms over-reliant on them?

Read more: risk models rms eqecat air worldwide catastrophe risk

Catastrophe models have become invaluable tools for the insurance and reinsurance industry, giving underwriters a scientific understanding of natural perils, in order for them to accurately estimate potential catastrophe losses to their portfolios and manage their exposures. The problems arise, however, when large, unexpected events happen. Real-life events such as September 11, Hurricane Katrina have taught the industry that models only know what has already been and this has frequently led to criticism.

In the past, risk modellers have suffered from the syndrome of workmen blaming their tools. Following Katrina, for example, many insurers and reinsurers complained that flood risk was not included in the models, meaning losses were far higher than they expected.

So with forecasters predicting an above average season for the number of tropical storms and hurricanes forming in the Atlantic basin this year, how seriously should the industry take the forecasts and risk model loss calculations and what influence will the latest risk models have on the market?

What the models offer the industry is the ability to estimate potential losses for a number of different uncertain events and a combination of variables. There has been much advancement in catastrophe models - both in the technology used to build the models and the data used to feed the models - that have improved companies' understanding of risk and loss potential.

According to Karen Clark, president and CEO of risk modelling consulting firm Karen Clark & Company and founder of AIR Worldwide, most of the recent risk model developments in the US have been with earthquake models. New studies by the US Geological Survey, a source of data for all risk modellers, has resulted in a "significant decrease in the earthquake risk, particularly in states such as California," says Clark. She says, while not decreasing pricing for every reinsurance treaty, this held down potential price increases for earthquake-exposed business.

A new RMS US earthquake model released in 2009 led to a reduction in US earthquake insured loss estimates and changed the industry's understanding of the geographical distribution of exposure. The effects of this have largely been digested by the market and now companies are waiting to see what the latest model developments will bring.

Most of the new models in the pipeline relate to wind and storm risk. In the coming year, Eqecat is releasing a basin-wide Asia typhoon model, AIR is launching updated models for US hurricane, European extra-tropical cyclone, and typhoon models for Asia, and Risk Management Solutions (RMS) is releasing a new Europe windstorm model and is in the process of building a new Atlantic hurricane model.

According to Neena Saith, senior catastrophe response manager at RMS, her firm's new Atlantic hurricane model will combine a large amount of high resolution data with a great number of claims forensics. Using historical data of hurricane activity, Saith says RMS will be able to run advanced numerical models to understand particular hurricane processes, such as the interaction between a storm and the surface upon making landfall.

Beyond this, RMS is looking to build a US flood model that will allow risk models to be run at a much higher resolution.

In June, AIR's latest US hurricane model was certified by the Florida Commission on Hurricane Loss Projection Methodology. The model incorporates "the latest research and scientific understanding of hurricane structure as well as detailed modelling for risk differentiation at a high resolution," Jayanta Guin, senior vice-president of research and modelling at AIR, told Reactions.

And in Europe, AIR's updated European cyclone fourth generation model is built upon numerical weather prediction technology that produces high resolution wind data. The updated model has been validated against a large database of surface wind observations and is ideal for use in parametric transactions. "In addition, the model introduces a novel approach to modelling clustering behaviour for storm activity over Europe, which is crucial for a complete view of risk," said Guin.

Eqecat's and AIR's development of Asia typhoon models is an interesting sign of the growing necessity for an awareness of insured property losses in an area of many fast growing economies.

Eqecat's models previously covered the individual countries of China, Japan, Philippines, Thailand, South Korea and Malaysia. "What we are doing now is building a basin-wide event set that allows our clients to understand the correlation among the risks that they have in those countries,†Bill Keogh, senior vice-president of Eqecat's strategic initiatives, told Reactions. "It is giving the market a new perspective on typhoon risk in Asia that they have never had before."

According to Keogh, Asia typhoon has three components to it. "It is not just about wind, it is also about storm surge and typhoon-induced rainfall that creates flooding. So the typhoon risk in Asia actually presents us with three different hazards, each of which needs to be modelled and understood, and then correlated."

He adds: "We incorporate all three of those in our model and we believe we are the first ones to do that. What this is providing to the market now is a full view of risks in Asia."

For Eqecat, the exceptional rate of economic growth in Asia over the past two decades has called for a comprehensive view of insured risk in Asia.

In July, it will also release a new earthquake model that implements three-dimensional vulnerability.

"We recognise through examining thousands of pieces of claims data from earthquakes that the way buildings respond to earthquakes can change during an earthquake," says Keogh. "After the initial ground motion and damage to a building, the building becomes more vulnerable to additional ground shaking. So as an event occurs for a longer duration it has the potential to be more vulnerable than we thought."

Number of Atlantic hurricanes
Near-term predictions
Long-term average Actual AIR Eqecat RMS
2006 5.9 5 8.4 8 8.4
2007 5.9 6 6.8 8 8.4
2008 5.9 8 6.8 8.1 8.4
2009 5.9 3 6.8 8.1 7.6
Total 23.6 22 28.8 32.2 32.8
Source: Karen Clark & Company

Risk of reliance

So how reliant should companies be on these models, and do they have a tangible effect on the market?

The financial impact of an updated model is hard to quantify because models are one among several components that determine pricing. However, as AIR points out, model enhancements do lead to improved risk differentiation. In the past this has led insurers to become more aware of exposure, which had a direct impact on insurance and reinsurance underwriting and portfolio management.

According to Tatsuhiko Hoshina, director and chief underwriting officer at Tokio Millennium Re, the Bermudian reinsurance operation of Japanese insurer Tokio Marine, companies do react to these models and this presents a threat to the market.

"It's a strange phenomenon from the cedents' point of view as the risk exposure is exactly the same, but just because the models change it has an effect. I think that's one of the biggest threats that we have in the market," Hoshina told Reactions. "Also companies are just relying on the output from these models and not knowing what has actually gone in. Whatever comes out from the different models they take an average and that is the threat that we have.

"There is a big uncertainty within the models especially if you blend it like that â€" you are just adding more uncertainty by using multiple models," he adds.

Bryon Ehrhart, CEO of Aon Benfield Analytics and chairman of Aon Benfield Securities, is less convinced that people have such a knee-jerk response to model changes.

Speaking on a catastrophe management panel at a Standard & Poor's insurance conference in New York in June, Ehrhart said: "People still take the models and make a judgement about what they need to do. Do you see people actively reducing their PMLs [probable maximum loss] and reducing the amount of cover because of the earthquake model? Not really."

He pointed out instead that market pressure has a pronounced effect, using the example of a European wind risk model change a few years ago. "The estimate of European wind went down pretty dramatically and people initially held the coverage that they had, but as pressure mounted in the past year and half people did start to reduce coverage," he said.

Ehrhart said it is interesting to see how dependent the industry has become on these models considering that "almost all of the volatility and vulnerability estimates are made by people that have zero capital risk."

However, Heike Trilovszky, head of corporate underwriting at Munich Re, says the industry reaction to models in the US differs from the rest of the world. "In the other [non-US] markets it's not that there is one model provider who comes up with a new version and then everybody uses that model and all of a sudden has a new view on the risk," Trilovszky told Reactions. "Some non-US markets don't have models at all, or if there are models, there are many models. It is not yet as standardised as in the US.

"RMS has a very strong position in the market and many people measure their exposure on the basis of that model. That is why in the US the RMS influence on prices is so material."

Number of US landfalling hurricanes
Near-term predictions
Long-term average Actual AIR Eqecat RMS
2006 1.7 0 2.4 2.3 2.4
2007 1.7 1 2 2.3 2.4
2008 1.7 3 2 2.3 2.4
2009 1.7 0 2 2.3 2.2
Total 6.8 4 8.4 9.2 9.4
Source: Karen Clark & Company

An uncertain future

Forecasters are predicting an above-average season in terms of the number of tropical storms and hurricanes forming in the Atlantic basin this year.

However, Clark questions the usefulness of the scientific forecasts that predict the number of tropical storms and hurricanes that form in the Atlantic each year, calling it a very "uncertain science".

"There does seem to be a consensus between the forecasters in that most of the predictions are well above average for the season; they tend to line up as they are based on the same indicators. However, predictions are typically wrong," she says.

"What matters to the insurance industry is: if it makes landfall, where does it make landfall?" says Clark. "When it comes to a major hurricane hitting land, no one can predict that and so you don't know if it will hit a highly populated area."

Clark points out that when Hurricane Andrew hit in 1992, it was a low frequency year for hurricanes in the Atlantic. The same was seen back in 1938 with the Great New England Hurricane that also occurred in a low frequency year. "Some of the worst loss years have been in low frequency years," says Clark. "It's just a random major event hitting a highly populated area and that is what is important to the industry. Losses are what are important."

Predicting if and where hurricanes might make landfall is a very challenging task that involves much higher levels of uncertainty than the predictions of storm and hurricane counts. As a result, only a few of the organisations that provide seasonal forecasts include landfall forecasts.

Clark believes there is too much uncertainty around year-to-year hurricane activity and insured losses to make credible short-term predictions.
According to a Karen Clark & Company report on the performance of the near-term models that have been designed to project insured losses in the US from Atlantic hurricanes for the five-year period ending in 2010, the losses for the cumulative 2006 through 2009 seasons have often been overestimated. As such, Clark recommends insurers, reinsurers and regulators evaluate the efficacy of the near-term models in light of this uncertainty.

Keogh says Eqecat's initiatives for education and training related to uncertainty remains the single most important resource that Eqecat has for its clients. "For many people when they think about uncertainty they think about the mathematical expression of uncertainty and numbers part of it," says Keogh.

Eqecat tries to hammer home the message that similarly built structures, subject to same peak gust wind speed or same ground motion, can have very different responses and experience dissimilar damage profiles.

"What we call that is real physical uncertainty and it's a way for people to understand that the uncertainty isn't just about mathematics. The uncertainty of the model is about the physical uncertainly about how buildings respond to different physical responses and to external forces. We have a lot of information about that and we use that information to inform how our models treat uncertainly," he says.

US insured losses from hurricanes ($bn)
Near-term predictions
Long-term average Actual AIR Eqecat RMS
2006 10 0 14 13.6 14
2007 10 0 11.6 13.5 14
2008 10 13.3 11.6 13.7 14
2009 10 0 11.6 13.7 12.6
Total 40 13.3 48.8 54.5 54.6
Source: Karen Clark & Company

Keogh says with risk modelling, it is vital to understand that models are tools and should be used as such. He does not claim that Eqecat holds a crystal ball and urges practitioners to have a high appreciation for the underlying uncertainly in the models.

So what does Keogh say models have to offer? "Our models, at best, will provide our clients with realistic expectations about risk," says Keogh.

Looking ahead, Keogh expects that models will become more accurate with regards to knowing what the uncertainties involved are. Although he notes: "Accuracy is not a word we tend to use in modelling. The underlying sciences, physics and engineering that go into the model get better, but at the end of the day the output you are getting will always have uncertainly in it.

"That is something that we want practitioners to embrace and understand so they are able to use them as tools as opposed to thinking that they have the answer when they get the output."

By Annabelle Palmer -

Risk models could help reduce poverty

The use of catastrophe models is not just about protection for the beach house owner in Florida or a homeowner on a fault line in California. Recently, insurance concepts of extreme risk modelling and risk sharing across finance, development and other industries have been extended to help development and aid organisations look for new solutions help poorer nations and emerging economies cope with catastrophes.

According to Rowan Douglas, chairman of the Willis Research Network (WRN), an academic and the insurance industry collaboration project funded by the reinsurance arm of Willis, the use of catastrophe models and the management of extreme risks are vital to sustainability, financial security and poverty reduction.

The methods employed by insurers to model extreme risks can be important in helping the world's most vulnerable communities cope with the social and economic impacts of floods, droughts and other climate-related threats.

Speaking at the World Bank's inaugural 'Understanding Risk' conference, held in Washington, DC in June, Douglas noted that the scientific models used by the insurance industry to assess the financial risks of extreme weather events could help governments and aid agencies develop new approaches for sustainable development and poverty alleviation.

During a keynote speech, Douglas illustrated how the tools, models and techniques used to help insurance companies sustain 1-in-200-year shocks could help deliver sustainability and financial security, and combat poverty in those regions most at risk to extreme weather.

According to research conducted by the World Bank, it is the poorest regions in areas such as south-east Asia that will be most affected by extreme weather events, including floods, cyclones and drought. India alone could see a 30% to 40% decline in agriculture productivity as a direct result of more extreme weather.

"We tend to manage for normality, but it is the extreme events that often matter most," said Douglas. "Whether it's in our financial institutions, societies or environment, sustainability is achieved by avoiding or managing the impacts of undesirable extremes within tolerable parameters. Therefore, we are proposing a new lens through which we look at the issue of sustainability, one which repositions the fundamental concepts and roles of insurance at the heart of delivering sustainability, financial security and poverty reduction."



Have your say
  • All comments are subject to editorial review.
    All fields are compulsory.


July/August 2016 
In this month's Reactions
 London Market Awards
 John Cavanagh

Click here to view this issue

  • Tropical Storm Madeline 27 Aug 2016 - Tropical Storm Madeline formed in the East Pacific Ocean in the early ...
  • Hurricane Gaston (07A) 26 Aug 2016 - Gaston has tracked northwest over the central Atlantic Basin over the past ...
  • Typhoon Lionrock (12W) 26 Aug 2016 - Typhoon Lionrock has strengthened over the past two days as it tracked ...
For more catastrophe reports, data and news, click through to the RMS/Reactions Catastrophe Centre.