Business - Creating a storm.

(From Reinsurance)

Byline: Adrian Leonard.

Modelling is an obvious part of the back-to-basics prescription for catastrophe reinsurers. As Aon Re explains in its recent review of the 2004 renewal, "modelling underpinned the ratings, and underwriters were not yet prepared to set prices below modelled levels. Coverage could not be obtained at prices below those thresholds, but capacity was abundant above them". Likewise reinsurance buyers cannot afford to be exposed: if corporate policy dictates purchasing cover up to a hundred-year return period, reinsurance managers will follow the models' guidance closely.

Against this backdrop, Risk Management Solutions (RMS) introduced a third generation of model technology for US earthquake and hurricane hazards in February 2003. Dubbed RiskLink 4.3, it affected reinsurance buying patterns. "The impact has been quite significant. The new model drove up PML figures by as much as 40% for certain perils with higher return periods," reports David Glover, managing director for non-marine reinsurance at Aon. One of his clients, an insurer in the hurricane-exposed north-east US, increased its catastrophe coverage by $100m to $250m based on the new model's amended calculation of its 100-year loss.

The impact on mid-west windstorm PMLs appears less dramatic, but the model change prompted increased buying for Florida and California exposures, particularly by US insurers writing catastrophe-driven excess and surplus lines portfolios. "There was a big buying spree at the end of 2003, when people were buying between $50m and $100m of additional cover," Mr Glover says.

Reinsurance asked Dr Robert Muir-Wood, RMS chief research officer, to explain the changes incorporated into release 4.3. He says there are two drivers of model evolution. "First is new scientific understanding; for example, of how hurricanes behave, and of how buildings fall apart in high winds. Science advances quite rapidly; this third-generation model has taken advantage of a lot of new science."

Second is the regular incremental increase in computing power. "Old limitations on the ambitions of how we can build a model have been swept away by more powerful computers. We can now model the events themselves in more detail and at greater resolution." For example, RMS uses about 400,000 possible hurricane tracks in its simulations. "In 1997 it would have taken a few weeks to run the event set once. When we developed the new model, we ran it repeatedly, day after day, making minor changes each time."