U.S. Severe Convective Storm Claims Going Through the Roof

During the development of the current RMS U.S. Severe Convective Storm (SCS) model, we found that claims for U.S. Personal lines were growing much faster than general economic inflation. To update SCS claims trends and to try and understand what could be driving this hyper-inflation, we analyzed the new five-year dataset from 2013 onwards, and also a longer duration 17-year period from 2001 to 2017 when observation datasets are of best quality.

Trends in SCS Event Costs

We gathered SCS losses due to hail, tornado and straight-line wind sub-perils from all the information we have on U.S. client claims, which amounts to over one million claims and several billions of U.S. Dollars in total loss. Figure One below shows the time-series of annual SCS loss totals and the decomposition into claim frequency and severity for the period 2001 to 2017. The 7.5 percent per annum trend in claim severity and 3.3 percent per annum rise in frequency combine to produce a growth of total loss, or SCS claims inflation of 11 percent per annum over the 2001-2017 period.

Figure 1: Time-series of annual U.S. Personal claims for SCS events: total loss values are standardized to 300, mean claim severity to 100 and claim frequency to 50 in 2001. The y-axis uses a log scale to suit compound growth.

This SCS claims inflation dwarfs the 3.8 percent per annum growth of non-inflation-adjusted GDP in the U.S., the average CPI growth of 2.1 percent per annum, and the five percent per annum growth of insured values from RMS analysis for Industry Exposure development (excluding growth in number of risks). The relative boost to claims growth over insured values implies loss costs from 17 years ago need inflated by a staggering 150 percent to relate to present-day risk.

Independent studies contain similar trends in SCS claims. For example, the Insurance Research Council (IRC) studied U.S. home claims for all cat events between 1997 and 2013, and found a seven percent growth of mean claim severity and 1.3 percent per annum trend in claim frequency per fixed number of buildings — or approximately 8.3 percent per annum claims inflation. If we re-analyze our available data over the same 1998-2013 period we obtain 9.5 percent per annum growth, consistent with IRC results.

In summary, SCS losses have grown much faster than general economic indicators over the past 17 years. But what is causing this growth? We now look to identify and measure the drivers of these troubling trends.

Hazard Changes

The first suspect is a hazard trend. The number of (E)F2 and stronger tornadoes is a useful measure because it is a good indicator of severe thunderstorm weather, and reports have been homogeneous for at least 20 years. Figure Two shows the annual number of F2 and stronger tornadoes hitting the U.S. using finalized data from the Storm Prediction Center (SPC).

A trend of minus 0.2 percent per annum between 2001 to 2017 does not show an upward trend in significant SCS weather events over the whole period. Next, we defined a Tornado Damage Index as the sum of the damage expected from each tornado based on its area (SPC-observed length and width), and its (E)F-rating, as documented in the RMS North America SCS Hazard validation white paper. At a national level, the index is trending upwards by three percent per annum. Further analysis revealed this was caused by spatially larger tornadoes in more recent times and known trends in observational practices cast doubt on whether this three per cent growth is a real meteorological trend.

Figure 2: Annual number of F2 and stronger tornadoes in U.S. 1998-2017. Source: SPC.

Trends in hail hazard are more important here because hail drives the majority of U.S. SCS insured losses, and our client claims contain a steeper claim severity trend for hail than tornado. Allen and Tippett (2015) remarked how the SPC database of hail reports is dominated by reporting trends throughout their 1955 to 2014 study period and obscuring meteorological trends.

We conducted two analyses to bracket the true hail hazard trend. The upper limit assumes the growing number of hail reports is entirely due to changing hazard. The lower limit assumes any linear trend from 1990 to 2017 is entirely due to reporting trends from Allen and Tippett, then we computed the annualized trend from 2001 to 2017 of this detrended sequence. In both analyses, we count the annual number of SPC reports of hail of two inches (five centimeters) or bigger. These two extreme assumptions produce a range of 0.6 percent and 2.3 percent per annum trend in the period 2001-2017.

We lean towards the lower end of this range because the modern-day trend in hail reports is a continuation of older trends known to be due to changing observational practices. Further, Allen and Tippett find little trend for the largest hail sizes over this period, suggesting little change to severity.

Our best estimate is that trends in hazard climate, weighted towards the dominant hail sub-peril, could add around one percent per annum to SCS claim frequency in the period 2001-2017.

Changes in Exposure and Vulnerability

Perhaps the biggest change in home exposure over the past two decades has been the rise in median age of housing from 30 years in 2001 to 37 years in 2013. Considering ongoing roof replacement, the age of roofs is expected to have increased by around three or four years over the same period. Older roofs are more fragile, hence expected to suffer more SCS damage. The idea that aging homes could drive trends in losses has been discussed elsewhere and we now review studies quantifying the impacts of roof age on SCS damage and claims.

An IBHS study investigated the dependence of damage ratio on the year a house was built, in hurricane winds under 90 miles per hour (145 kilometers per hour), while Alduse et al. 2015 did a modeling study of the effects of aging on failure of asphalt shingles, and both found significantly raised damage due to aging roofs.

Here, we place more weight on an RMS study on the impact of roof age on the mean severity of claims because it is based on 182,000 claims rather than the much smaller sample of under 1,000 used by IBHS, and based on settled claims rather than fragility modeling. The study highlighted raised claim severity with older roofs, and also found raised claim frequency too. Figure Three shows the combined frequency/severity impact on claims by wind speed, with each bin standardized to 1.0 for roofs aged three years or under.

Roof age has a massive influence on damage: roofs older than 12 years will suffer three times more damage than roofs up to three years old. BuildFax suggests median roof age is around 10 to 15 years, hence an increase by three years since 2001 implies a 50 percent rise in damage from Figure Three, or a mean boost of 2.6 percent per annum to claims inflation.

Figure 3: Homeowners’ claims (normalized to 1.0 for roofs of three years and younger), by roof age and selected wind speed bands.

Changes in Repair Costs

Trends in costs of labor and materials also affect claims. The cost of residential roofing materials has inflated at 5.1 percent per annum over the past 17 years while roofing labor costs have grown at slightly above 5 percent per annum from 2005 to 2014. Both are around 1.5 percent per annum higher than annual GDP growth.

Figure Four contains SCS loss trends per U.S. state, and a clear spatial signal of higher claims inflation in the highest-risk Great Plains area can be seen. The drivers mentioned previously show no signs of a locally enhanced trend in the Great Plains: hazard seems the likeliest explanation but analysis of hail reports and TDI reveals they contain no such spatial signal. Instead, there is evidence of changes in the roofing industry, with unscrupulous behavior among some of the roaming contractors leading to inflated costs, as well as poorer-quality roofs more vulnerable to subsequent damage. This would tend to affect the highest-risk states more than others, consistent with the spatial signal in Figure Four.

To get an idea of the size of the issue, we note there is an eight percent per annum countrywide trend in total losses excluding Great Plains states, which grows to 11 percent per annum when including the Great Plains states. Assuming local increases in the Great Plains are due to changes in how roofs are fixed (we have no other credible explanation) then it could be injecting up to three percent per annum into SCS countrywide claims inflation.

Figure 4: Trend (per annum) in total losses for personal lines, per U.S. state.


RMS research shows SCS losses increasing sharply at 11 percent per annum from 2001 to 2017, far exceeding CPI inflation and GDP growth. We are confident such hyper-inflation exists in SCS losses, but much less certain about its drivers and their relative roles. We know the costs of roofing labor and materials are running a little over five percent per annum and are much more relevant than CPI or GDP measures. Our best estimates of other factors are:

  • Changing SCS hazard lifting losses by around one percent per annum
  • Aging housing stock and roofs boosting a further 2.5 percent per annum
  • Changes in roof contractor practices injecting up to three percent per annum

Knowing event losses in the past is just one half of the problem for pricing and validation. Knowing how to trend them to present-day conditions is vital too.

Director, Model Development

Stephen is a hazard specialist who leads the development of climate hazard models from our London office. After joining RMS in 2009, most of Stephen’s focus has been on developing the Europe Windstorm (EUWS) hazard module; working with station data to calibrate RiskLink 11 EUWS event set hazard, then various hazard improvements to the RiskLink 15 EUWS version, and the RiskLink 16 EUWS clustering model. Stephen also spent 15 months leading the recalibration of the U.S. and Canada Severe Convective Storm model, released in January 2014. Before RMS, Stephen worked in various research and development posts over a period of 13 years at the U.K. Meteorological Office, including the development of short-range weather forecasting; designing and building new seasonal and decadal climate prediction systems; and the development of radiation and cloud physics parametrizations.

Leave a Reply

Your email address will not be published. Required fields are marked *