U.S. Hail in 2023
8th September 2023
We know U.S. Hail losses in 2023 have been very high, though where it ranks in history is less clear. Some say it may equal or even exceed record losses in 2011, such as AJG and Aon, while hazard analysis by BMS places 2011 in a different league, and 2023 very unlikely to match it. I use a novel hazard quantity with strong links to damage, and learn 2023 losses have a return period of around 10 to 20 years, and clearly below 2011 losses.
The analysis is based on Storm Prediction Center (SPC) preliminary data on hail occurrence and maximum stone size available here. There is a well known trend in numbers of SPC reports which could be confused with activity levels, but it can be minimised in a couple of steps. First, the use of hail days over a specific area has smaller trends than the total number of reports, as shown in Figure 3 of Allen and Tippett (2015). Second, decent-sized cities will have smaller reporting trends in hail days over recent decades. These two facts are combined into a new quantity called city hail days to provide a more homogeneous record of hail activity from SPC reports. If hail above a threshold size is reported in a city on a given day, then this is recorded as a city hail day. The number of city hail days can be found for a year, and this can be repeated for many cities over the U.S. to give the total number of city hail days per year. If the cities have good population density over the past few decades, then the method provides a long and reasonably homogeneous historical record of US hail activity.
In addition to better temporal homogeneity, the city hail day metric provides good information on longer-term loss histories, since most insured damage occurs in these cities.
Data on U.S. city locations and population were downloaded from SimpleMaps and those with a population above 50,000 and more than 0.5 degrees distant from neighbouring larger cities are shown in Figure 1. These 313 cities form the core of the method to measure U.S. hail hazard: count all cities with at least one hail report within its area in a day, then sum them to get the annual total of city hail days. The city area has been defined to be larger for those with greater population, and vice-versa.
Figure 1: map of the 313 cities used in this study, and their population (in 1000s).
SPC storm data contain a bias by hailstone size too, which is well described in Schaefer et al. (2004). In summary, estimates of stone size from the general public are clustered around the sizes of more common objects such as quarters, golf balls and baseballs. Further, Allen and Tippett (2015) discuss how the type of reporter is changing over time, which might introduce trends per size bin. These observations suggest getting the fuller picture by looking at historical timeseries for hail sizes exceeding a range of thresholds.
SPC provide preliminary data for 2023, and daily files from 1st January to 31st August have been downloaded and analysed. While the data are not final, a test revealed preliminary data were within 1% of the final data for 2011. For comparison purposes, the same January-August period is analysed in all years.
Figures 2 and 3 contain timeseries of the annual numbers of city days with hail exceeding a variety of size thresholds. Figure 2 concerns significantly damaging hail, and Figure 3 displays numbers of more severe hail reports. As a rule-of-thumb, the total losses are roughly a 50:50 combination of the significant and severe hail in Figures 2 and 3 (large hail may be fewer in number but they are much more damaging per occurrence).
The key features are:
- 2023 is in the top 5 most active years for significant hail (Figure 2) and clearly below 2011
- 2023 severe hail is on a par with 2011 (Figure 3 - its deficit in 2” hail is roughly balanced by more 3”+ hail)
- There was a distinct lull from 2013 to 2022 in significant hail (Figure 2)
Figure 2: timeseries of total number of city hail days in January to August, for significantly damaging hail categories.
Figure 3: timeseries of total number of city hail days in January to August, for severely damaging hail categories.
Two more points are worth highlighting. First, the use of city data here is unlikely to completely remove under-reporting in the 1980s and early 1990s, hence values in the early period may be slightly underestimated. Second, the city-based metric will be strongly connected to hail losses, since most insured damage is caused by these hailstones hitting the set of studied cities.
The picture for hail is quite clear: 2023 is in the top 3 or 4 over the past 40 years, and 2011 is the only year in the past 40 which we can be sure was more damaging. Further, we know this year’s tornado damage is well below that seen in 2011, when it was a significant fraction of the total SCS loss. It seems highly likely that a repeat of the events of 2011 would cause tens of percent more losses than occurred in 2023. Yet some in the industry say 2023 is on a par, or likely to exceed 2011 losses. What’s causing such different views on 2011?
The key is the trending of 2011 losses. If we use trending to address the question ‘what would be the losses if past events occurred today?’ then the sharp trends in SCS claims must be accounted for. There exists a roughly 7.5% per year trend in mean claim severity, which is independent of any exposure considerations. This is much steeper than CPI of about 2.5% per year over the past couple of decades. Combining the severity trend with a cautious 1% per year growth in the total number of industry risks, then 2011 losses would grow by 166% if they were to repeat in 2023. If we underestimate the true trend then we might place 2023 losses above 2011. On a related note, some have commented on record numbers of events with losses above a threshold, such as one billion. The main driver is the remarkably high claims inflation in SCS over the past couple of decades, though an extra 1 or 2% exposure growth in major urban centres would also increase the occurrence of severe loss events.
In summary:
- the 2023 hail hazard is amongst the very highest in the past 40 years, and its annual losses are likely to have a return period of around 10 to 20 years
- 2011 is likely to have suffered greater hail hazard in cities, and much more tornado damage than 2023
