Catastrophe models are very useful in assessing risk exposure, and it is no surprise that they have become essential tools for any insurer underwriting catastrophe loss coverage. But in recent years, they have evolved from useful supplemental tools to forces in their own right. Today, model revisions can become disruptive events affecting large segments of the industry - as we have seen over the past year with major changes to models by RMS and AIR.
The tail is wagging the dog. As valuable as they are, these models reflect highly imperfect science and carry levels of uncertainty far greater than their influence would suggest.
In the interest of helping clients use these tools effectively and appropriately, Guy Carpenter has undertaken a research effort into catastrophe models and how they can best be used. Our findings suggest that overall uncertainty levels can be mitigated to some extent - and thus better results be obtained - by employing multiple vendor models when evaluating catastrophe risk.
Among our findings:
Vendor-reported uncertainty bands vary widely. As the chart indicates, there is a vast difference in the vendor-reported uncertainty bands from the three major vendor models along an exceedance probability (EP) curve for the same insured portfolio. (We believe the true uncertainty band is actually underreported in all three cases.)
Figure 1: Vendor Model Uncertainty Bands
Source: Guy Carpenter
Not all models are created equal. Different vendors use different methodologies in developing their models. They have different estimation, fitting and smoothing techniques, as well as different representations of scientific details. Using multiple models “diversifies” the risk of error from these choices by allowing for independent errors to cancel each other. It also allows clients to use the best tool for a given evaluation, as certain models tend to perform better than others for certain perils, regions or class of risk.
More data leads to less uncertainty. While various models are founded on a common set of data, they are not founded on identical data. Each has its own interpretation of the historical record, analysis of detailed scientific data (e.g., wind fields or earthquake propagation), sources of vulnerability data (damageability data) and data on site conditions. Using multiple models effectively increases the amount of data bearing on the analysis.
Guy Carpenter works with clients to run model validations on a case-by-case basis, the results of which can suggest a preferred model for a company or a blending of multiple models. We have made a substantial investment in building the industry’s leading research and analytics capabilities in order to deliver unparalleled insight, tools and information to our clients. A better understanding of catastrophe models - and helping clients make the most of them - is an early return on that investment.