2017 is set to become the costliest year in history for weather disasters. Weather‑related insured losses for 2017 have been estimated at US$134bn and some 60% of this figure comes from the three major events of the 2017 hurricane season: Harvey, Irma and Maria (HIM).
In this context, catastrophe modelling has assumed an increasingly important role in recent years for the (re)insurance industry. It has become highly sophisticated, a unique combination of actuarial science, engineering, meteorology and seismology. These models combine analysis of long‑term climate change trends, economic growth and coastal construction with windstorm patterns and effects of storms over land. However, the 2017 hurricane season has raised significant questions about the methodology and accuracy of catastrophe modelling.
Following the onset of hurricanes HIM, modelled loss estimates varied significantly, and it is the level of fluctuation that has drawn criticism from some in the industry and prompted calls for better models. The widest variations were seen for Maria, where estimates ranged from US$15bn up to US$85bn. The actual insured losses from Maria are yet to be confirmed, but Swiss Re has estimated them at US$32bn. In total, hurricanes HIM caused US$215bn of overall losses, of which US$92bn is expected to be insured, according to Munich Re. Catastrophe modelling has never been an exact science and modellers accept the ranges of errors and uncertainties in their models.
What makes catastrophe modeling difficult?
Hurricanes HIM demonstrated that there are many reasons that it is difficult to model hurricane losses accurately. Whilst catastrophe modelling can predict a given level of physical damage following a storm, the quantum of actual insurance claims is another matter and can depend heavily on the policy terms and an insurer’s claims adjusting practices. For example, policy provisions for basis of indemnity can significantly affect the amount of a claim. Another issue is that catastrophe models can struggle when strong wind events are also major flood events. This was the case for Harvey, whose impact at landfall was relatively modest, but was followed by record levels of rainfall over Houston – the fourth largest US city. This was difficult to model, because catastrophe modelling does not typically take account of rainfall and flooding.
Sometimes there are specific circumstances in play which significantly affect insured losses whilst eluding the most sophisticated models. One example is Superstorm Sandy in 2012, which led to US$300m in lost fine art in the many expensive beachfront homes that were damaged.
The speed at which residents leave storm‑threatened areas can also have a significant impact on ultimate insured losses. For example, the exposure to car insurers varied wildly between Irma and Harvey because many residents had left Florida before Irma hit, but had not left Houston before Harvey. There are now fears that certain auto insurers in Houston may go out of business.
These complicating factors for property damage (PD) modelling also have a knock‑on effect on business interruption estimates, which are typically modelled as a function of PD estimates. Moreover, the margins can be extremely fine. It is estimated that a mere 20cm rise in the sea level at the southern tip of Manhattan Island increased storm surge losses from Superstorm Sandy by 30% (around US$8bn).
How can models be improved?
Some in the (re)insurance industry have put forward suggestions as to how models could be made more accurate and useful for future events such as HIM. David Flandro of JLT Re recommended "a combination of models or...completely different methodologies from the usual distribution‑driven, stochastic analyses currently widely used". Flandro also discussed the option of moving away from emphasis on severity, windspeed and trajectory to an approach encompassing multiple storm trajectories, modelling them every few miles apart in real time. He said "storm surge and flooding could be calculated for each of these scenarios and ramped up to different severities".
Others have commented that closer attention should be paid to the quality of the loss estimates generated by the different models and on how well the scientific data is implemented to produce reliable loss estimates.
With the new hurricane season set to begin officially on 1 June, catastrophe modelling will again take on an important role, especially given that forecasters are already predicting a high level of activity. It remains to be seen whether the modelling of future weather events will prove easier than HIM. In any event, insurers will be striving to keep premium rates high so as to build up reserves to cater for the next high loss year.
In October 2017, HFW London hosted a presentation and panel discussion about the impact of hurricanes HIM with Gerard Kimmitt and Sheshe Evans from HFW’s Houston office together with HFW London partners Chris Cardona and Andrew Bandurka. To read more about the discussions, follow the link: http://www.hfw.com/Hurricanes-Harvey-Irma-and-Maria-October-2017.