Why hurricane risk modelling has to change
It has been a record year for hurricanes, with 30 named storms battering the North Atlantic since the start of the summer.
Thankfully, insurance losses have been comparatively moderate. That’s because the six hurricanes that made landfall in the United States did not come ashore in densely populated areas. But in a year when insurers are already feeling the weight of coronavirus-related losses, it could have been much worse.
From 1995 onwards, North Atlantic tropical cyclone activity has been markedly higher than the long-term historic average. But even by the standards of the last 25 years, 2020 was significantly more active in every respect.
With tropical cyclone activity increasing in the North Atlantic and ever more records being set in recent years, is it time for the insurance industry to take another look at the way we calculate hurricane risk?
Watching and waiting
Everyone in the industry will remember an anxious weekend watching the cone forecast for Hurricane Dorian in 2019, after a devastating landfall in the Bahamas, with a track simulating a full-frontal hit on Florida’s densest coastal sections and insured loss estimates of over USD 100 billion. In the final hours, the storm altered course and spared people and property in Florida from catastrophic impact.
Nevertheless, 2020 kept everyone nervous. With more than 30 storms, the likelihood of an unfortunate combination of storm track and landfall location materialising was high. Whenever a storm makes landfall on a coastal stretch with a high value concentration, there will inevitably be widespread economic and insured losses.
The Atlantic Multidecadal Oscillation (AMO) is a pattern of natural climate variability which is correlated to the major hurricane frequency in the North Atlantic. The AMO is characterized by fluctuations in ocean temperature and changes phase approximately every 25-40 years. Since 1995, North Atlantic sea surface temperatures have been warmer than average, contributing to very active hurricane seasons of 1995, 2004, 2005, 2008, 2010, 2017, and 2020.
The effect of climate change on North Atlantic tropical cyclone frequencies is at this time uncertain, but increasing sea surface temperatures are expected to allow hurricanes to intensify more. Higher atmospheric temperatures increase the moisture in the atmosphere, making hurricanes with extreme rainfall such as Hurricane Harvey (2017) more likely to occur.
Post 2004 and 2005 active seasons, for many years, risk modellers and the industry considered the impact of the AMO on major hurricane frequency in the Atlantic and adjusted risk views accordingly. However, for the last several years, some of the commonly used catastrophe models have started to disregard the influence of the AMO on Atlantic hurricane activity and to revert to base their predictions on 100+ year averages.
At Swiss Re we incorporate a "near-term view" of tropical cyclone activity, including the AMO phase, warmer sea surface temperatures, and observed near-term tropical cyclone frequencies, into our risk modelling.
Hurricane risk today is, in fact, higher than a long-term average would indicate, also underpinned by storm observations. Relying on data from the distant past creates a disconnect between modelling and actual risk.
Sustainability at its core
In 2020, losses linked to hurricanes totalled around USD 20 billion. But in 2017, that figure was USD 92 billion – equivalent to 0.5% of US GDP. Given the patterns of extreme weather that we are seeing, it is surely only a matter of time before another heightened storm season leads to staggering levels of damage.
Furthermore, alongside near-term climate signals, underlying – but hard to measure – climate change, macro risk factors like rapidly growing populations and property values in exposed areas have been and will contribute to continuously increasing losses resulting from natural catastrophes globally, making past experience a less definite predictor for future losses. Not only picking the right observation window of past records, but also implementing forward-looking model perspectives becomes paramount.
With another active hurricane season likely for 2021, on the back of the current elevated AMO status and several worrying years, it’s clearer than ever that a significant change is needed to a common practice of hurricane risk modelling. It may be that there is a greater upfront cost to include recent climate signals and observations compared to using a less hazardous long-term record, but we cannot allow competitive pressures to bias risk assessment.
The systematic underestimation of hurricane probability serves no one well. For an industry with sustainability and resilience at its core, it is key to manage the impact of hurricanes, in frequency and severity, on the back of a sound, factual and quantitative-based risk model.