US Hurricane Clustering: A New Reality?

Alex Ntelekos, Dimitris Papachristou and Juan Duan

The 2017 Atlantic hurricane season was the fifth most active in 168 years.  It was also one of only six seasons to see multiple Cat 5 hurricanes (Irma & Maria).  These two hurricanes, followed similar tracks and, together with Hurricane Harvey, occurred close together.  This situation can hinder relief efforts.  For insurers it may also lead to resource strain, disputes and unhedged risks, if insurers do not have enough ‘sideways’ reinsurance cover.  Our post asks whether three major hurricanes occurring in the US in close succession really was exceptional or, as our analysis of recent data suggests, it might happen more often in future.  Is the insurance industry underestimating the likely ‘clustering’ of major hurricanes?

Introduction

Natural weather events such as hurricanes can cause significant loss of life, disruption to homes and communities, property damage and other economic losses.  In addition, many insurance companies are heavily exposed to these events as they will insure property and other risks within areas commonly affected.  The US Eastern Coast is one area where hurricanes are relatively common and where insured values are particularly high.  As a result, it is important for insurers to consider the potential occurrence of hurricanes in this region, and the damage they might cause.

The estimation of hurricane losses depend on several uncertain parameters including their frequency, size and intensity, the likelihood of making landfall, the accuracy of data on insurers’ exposures and historical meteorological records, property values in the affected areas, and the amount of rain and flooding.  We focus in this paper on the observed frequencies of major Atlantic hurricanes.

Our analysis suggests the frequency of the most intense Atlantic hurricanes with significant insurance implications could double in the future if recent observed changes in hurricane frequency persist. We estimate a 15% (a 1-in-7) chance of three or more major hurricanes making US landfall every year and a 0.6% (a 1-in-180) chance that these all impact the US East Coast (leading to potentially very high insurance losses). Insurance solvency regimes typically expect insurers to be able to withstand a 1-in-200 year shock.  So a season with three or more hurricanes is within the spectrum of possibilities that insurers should consider.

Hurricanes can be classified into different groups

Kossin et al (2010) identified four distinct regional groups of Atlantic hurricanes. Their paper analysed all storms included in the National Hurricane Centre’s HURDAT database from 1950-2007. We have extended their analysis to include all storms up to 2017, using updated data provided by Suzana Camargo of Columbia University.  The four groups, that pack together hurricanes with similar tracks, are shown in Figure 1.

Figure 1: Major hurricane tracks from 1950-2017 separated by regional group

The groups reflect different genesis locations, mechanisms and varying impacts of the main climatological indices (e.g. El Nino-Southern Oscillation – ENSO):

  • Group 1 storms form further north, over typically less warm water, and usually curve northwards over the Atlantic, thus have minimal impact on the US.
  • Group 2 storms form mainly in the Gulf of Mexico. Although these often make landfall, they typically have limited time to develop into major hurricanes. Despite that, Group 2 includes hurricanes that led to significant insurance losses such as Katrina (2005), Sandy (2012) and Harvey (2017).
  • Group 3 storms are the classic Cape Verde, high intensity deep-tropical storms. These develop over warm tropical waters off the coast of West Africa and have time to gather energy to develop into major hurricanes. A significant proportion curve to the north and miss the US. However, Group 3 spawns some of the most powerful hurricanes with the potential of severely impact the US East Coast. Examples include Gloria (1985), Hugo (1989) and Fran (1996).
  • Group 4 storms are like Group 3, but originate further west and tend to maintain a westerly track, through the Gulf. Recent destructive examples include Andrew (1992), Ike (2008), Irma (2017) and Maria (2017).

The table below summarises the split of major hurricanes between the different groups from 1950-2017 (the ‘analysis period’). It also includes the number of Continental US landfalls, where the largest insurance losses typically occur.

Table 1: Split of major hurricanes by regional group

Regional Group Number of major hurricanes Continental US Landfalls (%)
Group 1 23 5 (22%)
Group 2 24 21 (87%)
Group 3 76 16 (21%)
Group 4 53 27 (51%)
All Groups 176 69 (39%)

 

Although we have included all major hurricanes when analysing past and future hurricane frequencies, for insurance-related statistics, we have presented the results for land-falling hurricanes only.

Tendency for major hurricanes to occur close to each other within a hurricane season

We have analysed the tendency of major hurricanes within a season to occur close to each other. In Figure 2 we show the distribution of genesis dates of major hurricanes split by regional group for the analysis period.

The results point to two main conclusions:

  • Maximum major hurricane activity occurs in early September, and;
  • The frequency of Group 3 major hurricanes significantly higher than other groups. Furthermore, they tend to occur in close succession.

Figure 2: Distribution of genesis dates of major hurricanes by regional group

In a year when two or more Group 3 major hurricanes occur, we estimate there is around 50% chance that they will occur within two weeks of each other. If the same geographic areas are impacted – likely given the similar paths – it may hamper recovery and be difficult to attribute losses between the different storms. This can lead to litigation and reputational damage for insurers.

The past as a predictor of the future

It is frequently assumed that the observed historical variability will continue. We use simple statistical modelling of the average properties of each group to estimate hurricane activity. Table 2 shows the estimated probabilities of major land-falling hurricanes by group.

We note that “major land-falling hurricane”, in this context, means a hurricane that attained major status at some point in its lifetime and subsequently made landfall (not necessarily as a major hurricane at that stage). Intense hurricanes can cause significant damage even if they land as a lower category – Superstorm Sandy, the second costliest hurricane in history, made landfall on the US East Coast just as an extra-tropical storm, having previously attained Category 5 status.

Historically about 85% of major land-falling hurricanes were Category 1 or above at the time of landfall for the analysis period.

Table 2: Probabilities of future major land-falling hurricanes based on the analysis period

Regional Group Probability of 2 or more major land-falling hurricanes in a year Probability of 3 or more major land-falling hurricanes in a year
Group 1 1-in-400 < 1-in-500
Group 2 1-in-25 1-in-250
Group 3 1-in-40 1-in-450
Group 4 1-in-15 1-in-60
All Groups 1-in-4 1-in-12

 

Based on the average properties of each regional group between 1950 and 2017, we estimate a 1-in-40 chance that two or more major land-falling Group 3 hurricanes occur in a year. The probability for three or more major Group 4 hurricanes is 1-in-60. Given the destructive potential of deep-tropical hurricanes and the tendency of hurricanes to occur close together, these relatively high-probability events could have significant insurance implications.

Recent changes in hurricane frequency

The impact of climate change on hurricane frequency and severity is a widely debated issue with some uncertainty still remaining (e.g. IPCC, 5th Assessment Report). But data show there is a detectable increase in frequency and severity of intense hurricanes during the last few decades (Emanuel 2005, Holland 2013). This implies the (long term) past is no longer a good predictor of the future (Milly et al. 2008).

Figure 3 shows the average frequency of major hurricanes. The results show a clear upwards trend of Groups 3 and 4 during the last 30 years. The annual frequency of major hurricanes in these groups has increased considerably during this period compared to the 1950-2017 average.

The frequency of all major hurricanes taken together has also increased significantly during the last 30-years. Despite the cyclicality (Atlantic Multi-decadal Oscillation) of major hurricanes (Figure 3) the last two decades are showing a sustained increase of frequency that is above the historical maximum activity. Whatever the cause, the frequency of deep-tropical cyclones (Groups 3 and 4) is exhibiting a pronounced increase that is evident in the data.

Figure 3: 10-year rolling average frequency of major hurricanes, 1950-2017

Analysis of Group 4 major hurricanes (not shown here) suggests an increasing volatility in annual frequency. This suggests the probability of several major hurricanes occurring in a year may be higher than predicted by the Poisson distribution commonly used in insurance (Mumby et al. 2005).

In Table 3 we re-estimate the probabilities of major land-falling hurricanes shown in Table 2 reflecting the trends observed during the last few decades.

Table 3: Updated probabilities of future major land-falling hurricanes using last 30-years data

Regional Group Probability of 2 or more land-falling major hurricanes in a year Probability of 3 or more land-falling major hurricanes in a year
Group 1 1-in-400 < 1-in-500
Group 2 1-in-25 1 in 250
Group 3 1-in-20 1-in-180
Group 4 1-in-7 1-in-30
All Groups 1-in-2.5 1-in-7

 

We highlight the following:

  1. The probability of three or more major hurricanes impacting the US in any given year is estimated to be 1-in-7 (~15%). This is almost double the ‘historical average record’ scenario shown in Table 2.
  2. The estimated probability of three or more major hurricanes impacting the US Eastern coast (Group 3) in a given year (1-in-180) falls within the spectrum of risks that insurance firms typically model and expect to be able to withstand.
  3. The estimated probability of four or more major land-falling Group 4 hurricanes – not shown above – is also less than 1-in-200 years.

Conclusion

In summary, we have shown that if the last 30-years frequencies persist, the recent experience of the 2017 hurricane season is probably not an outlier but closer to the norm.  This is an important topic for insurers and others alike, and one where further research is needed to improve our collective understanding of hurricane clustering and its potential impact.  Given these trends, insurers, in particular, need to consider the operational, economic and risk management impact of the assumptions made when modelling hurricanes.

Alex Ntelekos, Dimitris Papachristou and Juan Duan work in the Bank’s Insurance Division.

If you want to get in touch, please email us at bankunderground@bankofengland.co.uk or leave a comment below.

Comments will only appear once approved by a moderator, and are only published where a full name is supplied.

Bank Underground is a blog for Bank of England staff to share views that challenge – or support – prevailing policy orthodoxies. The views expressed here are those of the authors, and are not necessarily those of the Bank of England, or its policy committees.

Comments Off on US Hurricane Clustering: A New Reality?

Filed under Insurance, Microprudential Regulation

Comments are closed.