Despite nature's best efforts, the widespread use of catastrophe models meant reinsurers' balance sheets escaped any major storm damage in 2004, says Hemant Shah

The 2004 Atlantic hurricane season was remarkable in many ways.

Insured losses within the US, across the offshore energy industry, and throughout several Caribbean islands, are likely to exceed $30bn, making this cluster of hurricanes the highest insured loss in any season on record.

Interestingly, this total was an aggregation of loss across four hurricanes, three of which - Charley, Frances and Jeanne - made a direct landfall in Florida; Ivan struck the neighbouring state of Alabama, but caused significant loss in Florida as well.

On average, only three hurricanes (Cat 1-5) and one major hurricane (Cat 3-5) make a US landfall every two years. The last year that the US experienced hurricane activity on a similar scale to 2004 was 1964, while the only recorded year in which Florida was hit by four hurricanes was 1837. So, not only was the 2004 season unusually active, it was also unusually severe.

Three of the storms made landfall at Cat 3 or greater intensity; two of them hit the US with Cat 4 winds. The Caribbean experienced even greater intensities. Three of the storms achieved at least Cat 4 winds at some point along their journey toward the US. In aggregate, the hurricanes of 2004 tallied the greatest "accumulated cyclone energy" (ACE) index since 1950, and posted the largest number of intense hurricane days (23) since 1926.

Thanks to the widespread use of catastrophe models, insurers and reinsurers have anticipated and planned for levels of hurricane loss well in excess of that experienced in 2004. For most portfolio losses incurred in Florida for the season were consistent with the 25 to 50-year return periods on their modelled Exceedance Probability (EP) curves. In contrast to the chaotic industry response which followed Hurricane Andrew in 1992, which was characterised by several insolvencies and severe market dislocations, the industry today absorbed a much greater impact in a deliberate and orderly manner. The adoption of third-generation catastrophe models, together with the more analytical and granular business practices catalysed by their use, have enabled the industry to construct diversified books of business, backed by adequate capital and reinsurance structures.

There are, however, many lessons to be learned from the 2004 season for modellers, insurers and reinsurers alike. Modellers now have the opportunity, and, I believe, the obligation, to investigate vast databases of detailed claims data to calibrate and refine their models. The season generated over 2.2 million claims in the US, a figure almost three times that of Hurricane Andrew, and thanks to years of investment in information technology and catastrophe management infrastructure, insurers are now able to capture and report data at levels of resolution previously unavailable. Modellers also need to develop more sophisticated approaches to quantifying multi-event risk, and in particular, the physicality of the correlation between hurricane activity, severity and geography, a phenomenon known as clustering.

And modellers, insurers and reinsurers must insist on a more robust quantification - and consistent application - of factors such as post-event claims cost inflation (also known as demand surge).

For reinsurers and insurers, the season provided an opportunity to evaluate, not only the performance of the models, but also the totality of their catastrophe management infrastructures. Exposure data quality, while much improved relative to five or ten years ago, remains a problematic issue.

Some firms experienced surprises, due to out-of-date, incomplete, inaccurate or simply miscoded data. A drive to further increase data resolution is important, but more work needs to be done by all involved to validate the underlying accuracy of exposure data, particularly where systemic issues may be of concern. A related issue is that some portfolios manifested a certain level of "basis risk" between the actual exposures and the assumptions embedded in the models being used.

It has been ten years since catastrophe models broke into the mainstream of insurance and reinsurance risk management practice. The essential distinction is no longer whether or not such models are used; today's best practice boundary is more subtle. Leading insurers and reinsurers do not simply use models or make use of catastrophe model output; they will have also internalised an understanding of the models, their capabilities, and their limitations. Ultimately, the experience of the 2004 season is a useful reminder that models are simply a set of tools, while modelling is, at its best, a way of doing business.

- Hemant Shah is president and CEO of Risk Management Solutions.