Cat Modeling in a New, More Open Phase: Aon Benfield/Impact Forecasting

By | June 23, 2015

Over 30 years ago, Karen Clark produced the first catastrophe model – for hurricanes. Since then the use of these valuable tools has changed the face of the re/insurance industry. They are indispensable for the developed world in studying and calculating potential catastrophe risks of all kinds – windstorms, floods, hurricanes and other tropical cyclones, wildfires, tornadoes and terrorist attacks.

There are hundreds, probably thousands, of them. Aon Benfield’s Impact Forecasting held a conference in London on June 17, attended by more than 200 people, to analyze the current state of cat modeling, how it’s grown, how it works and where it might be going next.

The conference, entitled “Impact Forecasting Revealed,” chaired by the division’s global head Adam Podlaha, made it clear that there’s no unique model for any given risk. Any insurer, reinsurer, or broker who assesses property/casualty risk needs to have a wraparound picture of that risk, which requires input from different sources.

Podlaha listed the line-up in ascending order, beginning with “scenarios,” i.e. overall views of a given geographic area where risks have been identified, and moving on to “probabilistic models,” where the likelihood of a catastrophic event is calculated, and the potential losses are identified.

"Climate is what you expect, weather is what you get,” Impact Forecasting's Steve Bowen, quoting Mark Twain

The next step is to narrow the point of focus by “customizing” the model, or in British terminology to create a “bespoke,” i.e. tailor-made, model for re/insurers and brokers to consult in order to understand the precise risk(s) they are asked to cover.

Podlaha explained that this requires combining all the available knowledge relevant to the risk, but, as that knowledge may be contained on more than one platform, it needs to be collated for a precise risk assessment. Impact Forecasting therefore developed “ELEMENTS,” its loss calculation platform, which is compatible with most other formats, including Oasis, the open source platform introduced in London last year.

Following its release, Trevor Maynard, Head of Lloyd’s Exposure Management and Reinsurance team, explained that the Oasis concept breaks up catastrophe models into their respective components – hazard, vulnerability, damage and insured loss. Each one is assessed at various levels, and then integrated to give the clearest picture possible for any given set of risks.

By integrating all of the elements that combine to define a risk Aon Benfield can produce the “primary tools,” required by underwriters to fully comprehend the nature of the risk they have been presented with and the price range into which it falls.

The process comes full circle when an event occurs, especially one that hasn’t been adequately modeled, such as the Thai floods or the recent earthquake in Nepal. Such events generate catastrophe reporting and analysis, which in turn enlarges the database for any similar events.

Weather and Climate

Weather related events generate the greatest number of catastrophic losses. Whether cyclonic storms – hurricanes and typhoons – thunderstorms and tornadoes, windstorms in Europe and Nor’easters, or the floods and mudslides from excessive rainfall, modeling them is a top priority for re/insurers and brokers.

Impact Forecasting’s Associate Director and Meteorologist Steve Bowen said that, while there are differences between weather and climate, they are certainly related. “Climate is what you expect, weather is what you get,” he said, quoting Mark Twain. The slow but inexorable rise in ocean temperatures underlies most changes in formerly “normal” climates. “Sea level temperatures have been warmer than average for 363 consecutive months,” he said.

When water warms it expands, raising sea levels and increasing the melting of polar ice. It also causes “more atmospheric instability,” which increases the frequency and the power of the storms that occur. Global warming has been accompanied by, and is probably in part caused by, an increase in carbon dioxide (CO2) in the atmosphere, which Bowen said is now over 400 parts per million (ppm), 93 percent of which is “stored in the oceans.”

The concentration is climbing higher, and is expected to reach 450 ppm by 2050, a concentration that hasn’t been reached in millions of years. The year 2050 is also when 66 percent of the global population is expected to be living on or near coastlines.

Thunderstorms, cyclones and floods have caused over $1 trillion in private and public losses since 1980, Bowen said. Eighty-five percent of those losses, however, were attributable to the growth in value of the properties affected; only 15 percent were attributable to weather or climate activities. In future years, however, those two factors will each augment the other in causing greater casualties and economic losses.

The scientific community is working hard on gathering more data and using more and better scientific analysis to produce more accurate cat models, said Dail Rowe, who leads a team at WeatherPredict. “We have lots of data on auto accidents, but not nearly enough on weather [related events].”

To fill in these gaps in the data, those working on the problem turn to science to help them find answers. “We blend science and data,” Rowe said. He cited the work done on how high and low pressure variants affect storm systems, noting that European windstorm models now take into account the phenomenon described as “clustering.”

This occurs when a second or third storm system follows directly after the first storm, creating a series of linked weather events. The classic example is Lothar and Martin, two storms that followed one another across France, Germany and Switzerland within 24 hours of one another 1999 and caused significant damage and loss of life.

Improvements to Catastrophe Models

The challenges described above have resulted in the development of new techniques and scientific applications in the construction and use of catastrophe models. It’s no longer sufficient to use past claims and specific storms as the main data source. These are now a starting point to bring in other data and to apply scientific analysis to create better models.

Aon Benfield’s Patrick Daniell described the application of engineering techniques in assessing the structural vulnerability of properties subject to European windstorms. “We applied different engineering elements, analyzed them and combined them with [existing] claims data,” he explained. The resulting “bespoke model,” produced on Impact Forecasting’s ELEMENTS platform, analyzes residential, commercial, industrial and agricultural properties.

Dave Martin, technical director of Australia’s Ambiental, described how the company created a model for hazards, exposures and loss estimates for its flood risk data products. The work has been validated by actual events, and gives an accurate picture of river and coastal flood potentials in Australia down to the postal codes.

Aon Benfield’s Charlie New went through the intricate technical stages required in the catastrophe modeling workflow. While the ultimate goal is to provide underwriters with the model(s) they require to make accurate coverage decisions, the amount of material required must be collected, analyzed and incorporated into existing systems. At this point technology and the people who know how to use it are critical.

After he had described the various platforms and services involved in creating accurate catastrophe modes, New said “Aon Benfield is now a software provider.” Even 10 years ago that would have been a rather startling comment. The explosion in the creation and dependence on cat models makes it today simply a statement of fact.

What the Future Holds

Technologies are rather like bicycles; they must continue to move forward, or fall down (to be replaced by newer, better technologies). In the last morning session a number of Impact Forecasting’s experts summarized some of the areas that their catastrophe modelers are working on.

Alexandros Georgiadis, who has responsibility for developing the division’s probabilistic windstorm risk model for Europe, focused on the damages these storms cause to Europe’s forests, particularly the rather extensive ones in Scandinavia. The models require data on the specific type of tree – spruce, pine and birch are the most common – as well as the soil(s) in which they grow, their location (higher or lower), and the potential, although diminished, valuation of a tree that’s been cracked or uprooted.

Cristina Arango, who specializes in earthquakes, explained how their team is working on models covering the U.S., Turkey, Southeast Asia, the Arabian Peninsula and Chile – among others. She highlighted the problem of having insufficient data in some areas – such as the Arabian Peninsula – and too much data in others, like Japan. In the former case it’s really only possible to construct scenarios, but in the latter probabilistic models can be quite complicated to produce, especially when the quakes frequently cause tsunamis.

Petr Puncochar reviewed the flood models Impact Forecasting is working on, noting that the division plans to release six of them this year. They include an updated U.S. model based on data from 22,000 rain gauges, prompted by the floods earlier this year in Texas and Oklahoma.

The floods in Canada in 2013 have also hastened extensive flood model revisions. In the Asia Pacific region getting information can be difficult. In which case Puncochar explained that Impact Forecasting uses “the data available” for both fluvial (river) and pluvial (rain) caused flooding. A similar situation exists in some European countries, notably Poland, where the claims data is only as recent as 2010. The data is good enough in both France and Brazil, however, to enable Impact Forecasting to generate a “bespoke” flood model for those countries.

The weather isn’t the only source that produces catastrophic events, as some of them are man-made. Mark Lynch, Impact Forecasting’s terrorism expert, have modeled the varying potential effects of explosions. The type of explosive used, the weight/volume of the charge and where it would potentially be placed all have to be considered in creating the model, he said.

In addition, considerations of the urban density, which includes the “geometry of streets” in potential target cities, the type of construction of the buildings at risk, as well as where they are located all affect the eventual model.

Using Frankfurt as an example, Lynch explained how the team “measures the propagation of pressure [from a blast wave],” analyzing the uncertainty of explosions in different locations, as well as the accident and health factors present in relation to a potential target. The greatest loss of life from terrorist bombings occurs when a building collapses.

Aon Benfield’s Impact Forecasting is one of many companies currently using state of the art data, science and technology to constantly create and update catastrophe models for its clients. It will not be too long before accurate models exist for most of the perils present in the developed world, and the models will be on their way to covering emerging markets as well. What effect that will have on the re/insurance industry remains an open question.

Was this article valuable?

Here are more articles you may enjoy.