Insurance and reinsurance industry catastrophe risk models constantly evolve and advance, but when adjustments are required, typically, companies must decide whether to “bend or blend,” according to reinsurance broker Guy Carpenter’s Global Chief Catastrophe Modeler, Imelda Powers.
Speaking with Reinsurance News, Powers explained that many insurers and reinsurers utilise a mix of prescribed formulas, self-developed models and licence vendor catastrophe models to develop their own view of risks, but stressed that when adjustments are needed, “there are typically two approaches – bend or blend.”
Powers, explained: “Bending a model refers to the transparent adjusting of its components to develop a customized view of risk for a particular region, peril and portfolio. Blending models involves the mixing of event results from multiple models to develop a final view of risk.”
Commonly, insurers and reinsurers rely on one licensed model from a single vendor as a result of cost considerations, despite potentially being able to obtain a second or third opinion from another vendor for further credibility.
When utilising a single vendor model, users can ‘bend’ the model by adjusting “the hazard module, event frequency or loss severity to develop their views of risk,” said Powers.
For insurance and reinsurance companies that are able to take advantage of multiple vendor models, comes the ability to validate each of them to inform their own perspective, explained Powers.
“For example, they may prefer Model A and B for residential and commercial risks, respectively. The process may include weighting models differently or braiding events from different models to form a new set of synthetic events for a particular region, peril and portfolio,” said Powers.
“Tailoring one model’s assumptions to customize the results to a particular insurer’s underwriting strategy and claims practice is a much more common practice than blending multiple model outputs. While licensing cost, infrastructure footprint and the expertise to validate individual models are initial hurdles, the challenge of blending them also weighs heavily in the decision to bend or blend,” she continued.
Interestingly, Powers offered some insight into the motivations of companies that prefer the ‘bend’ approach to their catastrophe risk modelling needs.
A preference for ‘bending’ could be driven by the fact it’s expensive to validate multiple models and challenging to articulate the proper weighting for each, said Powers.
Adding: “If the weightings are the same, regulators may question if the insurer is indifferent about its view of risk. It is more efficient to understand one model and fill in the gaps to customize it for the company’s needs.”
She continued to note that a preference for ‘bending’ might also be a result of the fact it’s difficult to blend personal and commercial losses from different event sets, with insurers preferring “the model that more closely fits the major line of business, bending its results for the other line.”
Thirdly, “It is costly to validate each model, and challenging to braid wind events from the U.S. model to those of the Gulf of Mexico model to accumulate onshore and offshore losses for a synthetic master event set. It is more effective to choose the model that better fits the larger exposure and bend the results of the other model to be fit for purpose,” explained Powers.
At the same time, companies with a preference to ‘bend’ models, might also have motivations for ‘blending,’ explained Powers.
A reason for this could be to “take advantage of multiple expert perspectives to the degree each matches underwriting strategy and claims practice,” she explained.
Powers also explained that there could be a situation where Model A is superior in commercial lines but Model B is superior in personal lines, and results are required on an all lines basis.
And thirdly, a motivation for blending could be the following:
“For correlated losses from the same peril, Models A and B are superior for different countries in the same region. For example, Model A is superior for U.S. hurricanes while Model B is better for Gulf of Mexico hurricanes, requiring a braided event set based on event categories and landfall locations from each,” she explained.
The above examples do not cover all scenarios, as noted by Powers, who said: “This does not include situations where, for uncorrelated losses, Models A and B are superior for different countries and perils; for example, Model A is preferred for U.S. hurricanes and Model B is preferred for Japanese earthquakes. Using the preferred model for each country/peril mix to form a combined view of risk does not constitute true blending because these analyses represent disjointed and uncorrelated analyses.
“In these situations, the decision is based on the cost considerations of one vendor compared with multiple vendors, and bending a less-preferred model may be adequate to meet the company’s need.”
