News

What next for catastrophe modelling?

flooding large With income declining in a stable to softening US homeowners’ market, insurers are starting to feel the pinch as the rating agencies set ever higher risk-based capital benchmarks, forcing them to seek more reinsurance. No surprise then that the focus is shifting from simply relying on the output from catastrophe models, to using them more proactively to reduce exposures and the costs associated with them.

According to data from the US Census Bureau, 53% of the population live within 50 miles of the coast. The Bureau also states that coastal areas are up to five times more densely populated than other areas and that 19 out of the 20 most densely populated US counties are coastal.

According to Matthew Patrick, Massachusetts state lawmaker: “Several retail homeowners’ insurance companies have left the coastal market citing the cost they must pay for reinsurance and the increase in projections by catastrophe models for hurricane damage.”

But for the majority of insurers who do not choose simply to walk away from the half of the US population that chooses to live by the sea, the challenge is essentially two-fold:

Manage these exposures, so that there is not too great a concentration of risk in those particular areas which have been, or are likely to be, worst affected; and
Improve data quality and modelling assumptions so that the output is more reliable, enabling regulators to set more realistic RBC (risk-based capital) benchmarks and insurers to manage existing business and make new business decisions.

Proactivity is essential

We believe that companies need to be less reactive to the output of these models and more proactive in managing their portfolios and the modelling process to produce a more realistic result. This could prove especially important at a time when, if a model indicates that a company needs to buy more reinsurance, capacity is scarce and therefore expensive.

This means that an improvement in data needs to be on the agenda, especially as modelling companies said that a major reason for the variance in actual vs modelled losses for the disastrous 2004/2005 hurricanes was the quality of data put into the models. This relates not only to the granularity of information, ie, address, zip code, etc, but also to the amount of risk level detail for an entire portfolio, ie, age, construction type etc.

Companies need to be less reactive to the outputs of models and more proactive in managing their portfolios and the modelling processes in order to produce a more realistic result

Models could also be increasingly used to identify what risk types and characteristics performed well in recent events. For example, we now know that homes in Florida of masonry construction, built to modern building codes, generally have lower average losses than older homes – and this is already being reflected in the new focus of Florida homeowner writers.

This, coupled with more sophisticated mapping software and optimization techniques, will provide far greater transparency regarding areas of higher damageability or frequency, enabling the use of additional underwriting and mitigation techniques to reduce potential losses.

Market thinking moving on

The impact of economic losses due to increases in labor and material charges, as well as business interruption losses, will also affect market thinking. Historically, models included a demand surge function which was usually a percentage that could be added to overall losses. This was not widely used before the events of 2004/2005. However, where insurers and reinsurers were previously comfortable either ignoring this or applying a modest load of their own, there will now be an increasing trend to model and manage this aspect of potential loss.

Overall, models will continue to change as actual events test our view of the world. This will lead to the creation of more accurate models, as well as tools for new perils such as terrorism and epidemic risks being developed. Completely accurate results will remain elusive due to the dynamic nature of catastrophic events – and the fact that any calibration/validation is reliant on real events – but greater accuracy is certainly a reasonable expectation, particularly if insurers engage more effectively in the process.

Duncan Aitken