The increasing risk of hurricanes – were we lucky in 2020?

A hurricane season will always be viewed with some trepidation; the potential for big-hitting losses is always there even in the quieter years: 1992 and Hurricane Andrew being the prime example. However, currently – as science and industry tries to grapple with understanding whether climate change is having an impact or not on hurricane risk – the interest in the forthcoming season is heightened further, especially given the capriciousness and at times fortunate nature of previous seasons.

How Lucky Were We In 2020?

The six landfalling hurricanes in 2020 served up a reminder, soon after 2004 and 2005, that a hyperactive season will always hold the potential for sizeable losses. But how lucky were we last year when insured losses were only around $21bn, given that six storms made landfall?

A quick look at the tracks of the systems in 2020 shows how they typically made landfall, largely, away from key population centres, aside maybe from Hurricane Zeta that was weakening as it passed close to New Orleans. The tracks shown here are the lifecycles of the 2020’s storms whilst they were above tropical storm status.

But just how lucky were we, monetarily? To assess this, your friendly local catastrophe model is hugely useful. Inigo licenses the AIR US Hurricane catastrophe model (amongst their suite of models) and we’ve used this to do some very simple – but hopefully informative – counterfactual analysis. We’re essentially wondering what could have happened if things had turned out slightly differently. All we have done here is randomly pick from the stochastic event set similar storms that made landfall in 2020.

We’ve then randomly drawn 100 times from each of the many alternative landfalls that exist in AIR’s stochastic hurricane set to see what the losses could have been in 100 “alternative 2020s”. The only thing that we control here is the landfall intensity and landfall state to be the same as the six storms that made landfall in 2020. Everything else is left up to what we pick at random, but we ensure that the storms don’t make a second landfall anywhere else.

The chart here shows that for six storms, alternative tracks into each state give a broad range of possible losses from about $15bn to $90bn but our “actual” losses sit quite a way down the distribution of 100 “alternative” 2020s.

This suggests that the hurricane’s direction, size and exact location of landfall in each of the counties led to a fortuitously low return of around $20bn or so across the six storms. The range here serves to remind us how changes in the landfall location – that often come down to subtleties in the atmospheric flow in the last 2-3 days before landfall – can lead to markedly different outcomes.

What about 2021 – and how it sits in the “bigger picture”?

It’s not the aim of this blog post to update you with the latest hurricane forecasts for next season – but you can find them here at the excellent Barcelona Supercomputer Centre website. All that is needed to be said is that things are looking above average in terms of forecasts, but the forecasts aren’t looking maybe as active as last year.

But what exactly is average these days? And what does it say about modelling hurricanes currently?

We can try to understand this by taking a look at the sea surface temperatures (SSTs) in the “Main Development Region” (MDR) that straddles 10-20N, 20-80W across the southern North Atlantic Ocean, over which a fair proportion of developing hurricanes will travel:

We show on the chart the August-October mean sea surface temperature. These three months take up the lion’s share of the hurricane activity. The thin lines show the yearly average SST across these three months, the thick yellow line shows the average SST of the prior 30 years’ worth of hurricane seasons.

The Colorado State University team recently shifted their “base” average season to encompass the 1991-2020 from their previous base of 1981-2010. This has pushed up the number of hurricanes in the basin for their “average” from 6 to 7. If we look at the change in warmth of the seas between 1981-2010 and 1991-2020 shown by the white lines in the chart above, the seas in the latter period are 0.2c warmer, which might in some way explain the increase in the underlying activity in the basin.

We can also see from the thin line that there is an oscillatory behaviour of the sea surface temperatures whereby there are “warm periods” of the sea surface temperature. We are currently in one, and the last warmer period was from around the late-1920s to the mid-1960s.

Scientific wisdom around this up until recently has pointed to the see-sawing of the sea temperatures being down to the “Atlantic Multidecadal Oscillation” (AMO) – a variability of the Atlantic sea temperatures driven by ocean circulations that leads to a couple of decades of cooler seas followed by a similar length of time of warmer seas.

However, there is now discussion in academic literature where it is felt that the AMO up-and-down signal could be driven via other sources than the ocean itself, through volcanic activity or potentially – in the case of the cooler phase of the 1960s – through human aerosol emissions. The fact that this science now doubts whether the oscillations exist is important to how we understand current hurricane risk.

In recent years, we’ve been waiting for the seas to swing back again into a colder phase of the Atlantic: but we’re still waiting. There are hints in the thick yellow line (that represents the prior 30 years’ average SSTs) that we’re potentially coming to the top of the current warm spell. BUT: if we compare this current warm spell with the peak of the last warm spell (the other black line on this chart labelled 1935-1965), the seas are on average around 0.3c warmer. These two “warm spells” are not the same: we are potentially seeing the impact here of a warming globe on the sea surface temperatures in our most recent “warm spell”.

The concern here is that there is interesting evidence in the sea surface temperatures and in the state of the science that maybe we need to understand risk very much in the present day rather than looking to the past to inform how we price risk for insurance purposes. It is, however, equally important to recognise that the science is not settled on this topic: as much as it’s not settled as to what the future looks like for hurricanes.

It’s also worth noting that it’s not just sea-surface temperatures in their entirety that control our hurricane seasons. However, this recent warm phase is an important aspect of our thinking at Inigo as we build out the “Inigo View of Risk” to be relevant to the present day. We are assessing output from global climate models to understand how the magnitude of changes in the tropical Atlantic sea surface temperatures we have seen recently might impact hurricane activity not just in the basin but also at landfall.

Dr Dixon brilliantly illustrates through his original research for Inigo the crucial role of fortuity in any one hurricane season. Almost any year could easily have been so much better for customers and insurers alike than the observed reality, or so much worse. At Inigo we will aim to be there for our insurance and reinsurance clients in the tough years, and will pay close attention to the latest science in order to make sure that we are charging appropriate rates, and managing our exposures thoughtfully, so that we can deliver solid returns to our shareholders in the long-term.

Russell Merrett

In recent years, through luck or otherwise, we’ve seen category 5 hurricanes later in the season (Hurricane Iota in 2020) and hurricanes further north and east in the Atlantic (e.g. Hurricane Ophelia in 2018) than experienced in the historical record. We must be cognizant to the fact that warmer seas potentially widen the area and length of time over which hurricanes occur, which may have a knock-on impact on the risk we underwrite.

All eyes will be on this year’s hurricane season to see if this behaviour continues.

We are very grateful to AIR Worldwide for allowing us to use their data in this blog post.