In my years of contributing to this blog, I have written extensively about the long-standing debate about the current state of hurricane activity in the North Atlantic Basin. This debate has become no clearer following the 2017 hurricane season; one of the busiest and costliest seasons on record. 2017 followed a stretch of below- to near-average seasons that began in 2012 and it is unclear whether future seasons will remain active or return the recent level of relative quiet.
As is usual in the weeks following a hurricane’s impact on land, much of the focus surrounding Hurricane Maria has now shifted away from estimating losses with models to surveying the actual damage and claims incurred. With the collection of claims and losses, evaluating the array of loss estimates published by catastrophe model vendors in Maria’s immediate aftermath will begin. Included in this array is the RMS best estimate of insured loss, a range between US$15 billion and US$30 billion.
19:00 UTC Monday, September 11
Tom Sabbatelli, hurricane risk expert – RMS
Irma is a hurricane no more: following a Category 3 strength second Florida landfall near Marco Island, south of Naples, the storm’s intensity rapidly faded as it traveled to the north and west overnight. Irma now only maintains tropical storm strength as it crosses northern Florida. Expect to see further commentary on the storm’s rapid decay from my colleagues in the RMS HWind team later today.
Irma’s landfall and overnight track has allowed RMS to narrow yesterday’s stochastic event selection. A landfall eliminates offshore track scenarios that produce lower levels of onshore wind but higher levels of storm surge hazard along Florida’s west coast. Irma’s passage to the east of Tampa reduces the risk of significant storm surge levels and loss near Tampa Bay.
Initial damage reports indicate that damage may not be as severe as once feared, despite sizable roof damage in Naples and the Florida Keys.
As per the RMS Event Response process, our attention shifts from the regular, reactive stochastic event selections to a comprehensive interrogation of all causes of loss, both modeled and unmodeled. During Hurricane Harvey, my colleague Michael Young explained that one of the biggest challenges we face is collecting enough wind observations to create a complete picture of a storm’s wind field. Although our RMS HWind team have been collecting and processing data every three hours, we face these challenges with Irma as well. More data will become available in the coming days, which will enhance the accuracy of our wind field and surge modeling.
As part of our event response service, we have the following activities underway:
- Reconnaissance will help us narrow the uncertainty around what could be a potentially significant contribution to loss from storm surge and flooding. These efforts do not need to wait for boots to hit the ground, though; we have teams already scouring high-resolution satellite images to detail the exact extent of the flood waters and the underlying exposure at risk.
- Our RMS Field Recon team, fresh off their trips to Texas following Hurricane Harvey, are reactivating their reconnaissance plans. In the days ahead, their visits to cities across Florida will help reveal the depth of the flood waters and the extent of wind damage observed throughout the state.
You may have already seen that our colleague, Victor Roldan, has been documenting his experience riding out the storm from his home in Miami. Victor has reported that basements are flooded along Brickell Avenue, and area that was hard hit in Wilma, but wind damage is minimal.
Following a hurricane, power outages predominantly contribute to heightened business interruption and post-event loss amplification, which is possible in an event of this magnitude. As many as seven million customers in Florida may have been without power at one time, almost triple the peak outage observed during Hurricane Matthew.
Let’s not forget about the Caribbean islands left in Irma’s wake, still cleaning up and attempting to restore power and telecommunications several days after the storm’s initial impact. For instance, as many as one-quarter of customers in Puerto Rico remain without power four days after Irma’s passage. This prolonged restoration may prove to be a figure that could compound insured losses across the island. During their comprehensive review of the event’s lifecycle, RMS modelers will refine projections of the insured loss across all Caribbean islands, which is assumed will contribute materially to the total industry loss.
As the Event Response team now transitions from producing real-time event updates, they have many existing key data sources from which they can draw these critical observations. Ultimately, these insights will inform our official insured industry loss estimate, targeted for publication in approximately two weeks’ time.
14:30 UTC Sunday, September 10
Tom Sabbatelli, hurricane risk expert – RMS
As Hurricane Irma makes landfall in the Florida Keys as a Category 4 storm, the range of the storm’s possible future tracks in the latest RMS HWind forecast product is rapidly narrowing. It is now certain that Irma will track along Florida’s west coast and impact all major population centers from Naples to Tallahassee.
What is less certain is the length of time Irma’s center will remain over water, with some scenarios projecting a landfall near Fort Myers and others delaying the landfall until Irma reaches the state’s Big Bend region.
15:00 UTC Saturday, September 9
Tom Sabbatelli, hurricane risk expert – RMS
The westward-moving trend of recent Hurricane Irma forecasts continues, with the Florida Keys, southwest Florida, and Tampa potentially within Irma’s sights. Although the National Hurricane Center forecast “cone of uncertainty” still covers much of south Florida, 83 percent of the individual forecasts analyzed by the RMS HWind forecast product bring the hurricane within 50 nautical miles of Key West, indicating that the Miami metropolitan region may be spared the worst of Irma’s winds. 75 percent of these forecasts also indicate a passage within 50 nautical miles of Tampa (see “Selected Probabilities” in the figure below).
17:00 UTC Friday, September 8
Tom Sabbatelli, hurricane risk expert, RMS
For the last 48 hours, all the forecasts have been consistent in indicating that Hurricane Irma will have a significant impact in Florida, most likely making landfall in the state. The latest RMS HWind forecast shows it tracking more westerly than before, and this reduces the potential for loss because of the relative concentrations of exposure in the southern end of the Florida peninsula.
Based on today’s long-range forecast, RMS calculates there is still a 10 percent chance of wind losses from Irma exceeding US$85 billion. This assumes a U.S. landfall, with the scenarios in RMS modeling showing almost all of that insurance loss to be in Florida.
But the loss range in this preliminary analysis could easily move higher or lower depending on shifts in the storm track and its intensity. Irma’s anticipated direction of travel has been changing continually through the week, oscillating between both coasts of Florida. It is likely that this changeability will continue, and so the modeling uncertainties remain significant.
RMS will continue to update its analysis of potential insurance losses as Hurricane Irma moves closer to the U.S. coast.
23:30 UTC Wednesday, September 6
Tom Sabbatelli, hurricane risk expert, RMS
Hot on the heels of Hurricane Harvey, Irma looks like it could be the second major landfall in the U.S. this season, as it currently moves towards the Caribbean as a category 5 hurricane, with sustained winds around 185 miles per hour (297 kilometers per hour).
As always, the RMS Event Response starts early in the life of tropical storms, to provide the latest commentary, following up with RMS HWind footprints as data becomes available and providing initial sets of stochastic event selections around 48 hours before landfall. RMS Event Response practices have been designed to best serve our clients and the industry as a whole, and speculation of industry losses whilst such uncertainty remains can be counterproductive. Clients can see full information on the RMS Event Response processes by reading the following document available on RMS Owl.
RMS has completed research on hurricane risk to single-family dwellings using an improved understanding of roof age, which can lead to more accurate loss projections using our models
Weak roofs mean losses during hurricanes. During reconnaissance trips to the southeast U.S. and the Bahamas following Hurricane Matthew last fall, RMS experts saw ample evidence of this simple fact. Their on-the-ground survey highlighted everything from shingle and tile damage to complete roof failures.
Roof weakness significantly influences RMS’ view of structural vulnerability in our North Atlantic Hurricane models, which can factor in a roof’s age, covering, and shape into calculations of potential loss. However, this valuable property data is not captured by many insurers, and this could represent a missed business opportunity to improve underwriting – whether it be pricing or risk selection.
Extending the Data, Refining the Insights
RMS already has a dataset of hurricane claims from over one million single-family dwelling (SFD) homes in Florida and the northeast U.S., representing $240 billion in total insured value. However, this dataset lacks roof characteristics for a majority of the homes, so we augmented it with roof age information obtained from BuildFax, which holds detailed building characteristics for over 90 million properties in over 10,000 U.S. cities and counties. From this enhanced dataset we found:
- About 70 percent of Florida homes (SFDs) had roofs aged 10 years or older at the time of the 2004-05 hurricanes
- Roughly half of the Northeast homes (SFDs) had roofs aged 20 years or older at the time of Superstorm Sandy (2012)
- Only 20% of all homes (SFDs) still had their original roofs, although this proportion was lower for coastal properties than for inland properties
So what was the relationship between roof age and losses? In the second stage of our research, our vulnerability modelers paired the exposure data with 182,000 hurricane claims, totaling $2.25 billion in paid losses, to look for patterns related to roof age.
As expected, we found that homes with older roofs generally corresponded with more claims, and claims of greater severity. This was most evident at the low wind speeds experienced in the Northeast U.S. during Superstorm Sandy, as well as at higher wind speeds experienced in the Florida hurricanes. These graphs show that buildings in Florida with a roof older than 20 years are associated with claims that are between 50-100% more severe, compared with buildings having a roof less than five years old. A similar trend appears in the Northeast, but is muted because of the smaller dataset.
That’s the picture from historical data. But what about modeling potential future events? To answer that question we analyzed the enlarged Florida dataset, focusing on how roof age at a particular location compares to the industry average for that region.
The change in modeled average annual loss (AAL) by county shows a patchwork of increased and decreased risk that corresponds to the average roof age of properties in each county.
So we can see that using roof age data leads to significant differences in modeled loss within regions.
That’s a valuable insight in itself. But we decided to drill down a little deeper.
From counties to ZIP codes to individual locations
Although the maximum change in AAL was less than 10% at the county level, changes of up to 20% were observed at the level of ZIP codes. These results show that improved understanding of predominant roof age could influence a company to change its regional underwriting strategy or refine its rating territories.
Going more granular still, within each county and ZIP code there is variation in the roof age of individual homes and this is critical to consider when writing new business. The scatter plot below shows the change in AAL at individual locations. Those homes with older roofs produce higher than average AAL and vice versa.
So when we go down to the level of individual locations the impact of roof age data leads to loss changes of up to 50%, demonstrating higher significance than at the regional level. For high hurricane risk locations in Florida with large baseline AALs, this change translates into substantial dollar amounts. That’s crucial to know, revealing key opportunities to improve underwriting practices. For instance, companies might choose to quote more competitively on price for properties with newer roofs.
Unsurprisingly, over time strengthened building codes and practices have led to stronger roofs that are more resilient to hurricane damage. But this research tells us much more – the sheer magnitude of modeled loss changes observed was significant, with clear implications for profitability, as explained by BuildFax CEO Holly Tachovsky:
“These results reveal key opportunities to improve underwriting practices, including pricing and risk selection. A focus on roof age can be the difference-maker for loss ratios in certain geographies. As a result, we see a growing level of sophistication among carriers that want to rate and select with a higher degree of accuracy.”
RMS remains committed to partnerships with industry experts like BuildFax to communicate the business benefits of emerging trends in the (re)insurance space.
For the meteorologist in me, hurricane and climate research is fascinating in its dynamism. The last two years have seen continuous scientific debate about the state of Atlantic basin hurricane activity, which we’ve reflected on thoroughly in the RMS blog.
But for the insurance industry, it’s more than just a fascinating debate: business decisions depend on clear insight. It’s more than just a number.
In April with the release of the RMS Version 17 North Atlantic Hurricane Model, we will include the latest biennial update to the industry’s long-term rates, in addition to the RMS medium-term rate forecast.
For the first time since its introduction, the RMS medium-term rate forecast has dipped slightly below the long-term rate.
For the U.S. as a whole, the new 2017-2021 medium-term rate forecast of hurricane landfall frequency is now one percent below the long-term rate for Category 1–5 storms, and six percent for major hurricanes (Category 3–5 storms).
Mind the Tail
The impact of the rate changes on the view of risk will vary from portfolio to portfolio. Measuring the new medium-term rate against the RMS Industry Exposure Database, we see a 16 percent decrease in the U.S. average annual loss (AAL) relative to the previous medium-term rate forecast – mainly driven by lower risk in Florida and the Gulf.
However, to focus solely on the headline AAL-based changes, or the national impacts, ignores the risk implications of the unique atmospheric conditions and key features of the new forecast.
At the 250-year return period, the decrease is more muted – at eight percent – which positions the medium-term rate slightly higher (one percent) than the long-term rate. Unlike with previous below-average periods, persisting warm sea surface temperatures in the Atlantic continue to indicate that the medium-term rate risk remains above the long-term rate in certain key U.S. regions, such as the Northeast.
The Science and Process Underpinning the Medium-Term Rates
Grounded in objective science, we follow a systematic process to develop the biennial medium-term rate each time we update it. We analyze 13 different statistical climate models, which all provide a five-year forecast of activity for the Atlantic basin.
The climate models reflect three main theories of hurricane variability in the Atlantic over recent decades:
- Shift models identify historical, multi-decadal periods of high or low hurricane activity, which are viewed as natural, inevitable oscillations
- Sea surface temperature (SST) models identify relationships between SSTs and hurricane landfalls in the past and use these to predict similar patterns in the future
- Active baseline models suggest that the low activity phase of the 1970s and 1980s was caused not by natural variability, but by high levels of atmospheric aerosols which are not expected to recur in the future
To provide a more reliable forecast, we take a weighted average across all 13 models – based on tests made of each model’s predictive “skill.” These tests compare how well the models predict hurricane activity in sample periods from the past, against what occurred. This rigorous testing process is revisited with each release of the medium-term rate.
The Latest Data – How Do the Climate Models Interpret It?
The new medium-term rate forecast uses updated information from the HURDAT2 hurricane dataset and the latest sea surface temperature data, including the 2014 – 2016 seasons.
The updated hurricane data reveals a four-year stretch of below-average Atlantic major hurricane activity between 2012 and 2015, leading to a five-year average trend that is decreasing.
On the other hand, sea surface temperatures over this same period have been rising. Energy derived from warm temperatures serves as an important driver for hurricanes – so you would expect to see an increasing rate of hurricanes, not fewer.
When the data is fed into the climate models they do not point in the same direction for future hurricane activity.
The shift models used in our medium-term rate forecast focus on the decrease in major hurricanes and identify the seasons since 2011 as statistically distinct from acknowledged active periods observed since 1950. This could be significant because it may indicate a transition to a quieter phase of hurricane activity, as discussed in Nature Geosciences.
But while the shift models indicate this transition, both the sea surface temperature and active baseline models do not identify a similar transition to a less active hurricane phase, in part based on the warmer Atlantic sea surface temperatures.
It’s Not Just a Factor
As I discussed earlier, the medium-term rate considers multiple drivers of hurricane activity, including sea surface temperatures. Peer-reviewed research highlights the influence of sea surface temperatures (SSTs) on hurricane tracks; thus, analysis of projected SSTs provides different forecasts not only of where along the coastline hurricanes are likely to occur, but at what strength hurricanes will make landfall. This is a process that RMS terms regionalization.
During higher sea surface temperature periods, the body of warm water over which hurricanes develop expands eastward towards Africa. This expansion increases the likelihood that hurricanes re-curve away from the eastern U.S. coast, towards the northeast and maritime Canada, following paths similar to hurricanes Irene and Sandy.
In the medium-term rate forecast it is regionalization that causes forecasted activity in the U.S. northeast and mid-Atlantic to be above the long-term average, despite a below-average forecast for the U.S. as a whole. This creates a pattern that differs from normal climatological expectations, which would typically be focused on the risk to Texas and Florida – although, obviously, in those southern states the risk does remain higher in absolute terms.
The forecast’s regionalization also produces slightly above long-term risk beyond the 100-year return period, on the industry U.S. exceedance probability curve. At the 250-year return period, for example, while the new medium-term rate has decreased risk by eight percent, the new forecast remains one percent above the long-term rate. Despite decreases in the forecasted frequency of large loss-causing, tail events in Florida and the northeast U.S., warm Atlantic sea surface temperatures continue to support the possibility of these events occurring at a rate above the long-term average.
Delivering the Model
Pre-release data sets for the new medium-term rate are now available in advance of the April release of the updated RiskLink® Version 17 North Atlantic Hurricane Models. These are accompanied by technical documentation describing the process’ methodology and its impact on risk. It will also be concurrently available within Risk Modeler on the RMS(one)® platform.
For further insights from RMS experts on the new forecast, as well as on model updates and the latest on the RMS(one) solutions platform, join us at Exceedance in New Orleans, March 20-23.
After a relatively quiet start, the 2016 Atlantic hurricane season grabbed the attention of the insurance industry during September and October. On October 6, all eyes fixed on Hurricane Matthew, a Category 4 storm barreling towards Florida and presenting the greatest threat to the U.S. insurance industry since Sandy in 2012. Despite Matthew’s high winds and floods, caused by both storm surge and rainfall, the storm’s offshore track certainly spared Florida and the southeast U.S. from a potential worst-case scenario.
Matthew was preceded by Hurricane Hermine, a Category 1 storm that made landfall along Florida’s Panhandle on September 2. These storms ended a nearly 11-year period where no hurricanes affected the state of Florida.
While the soft insurance market is expected to weather the effects of Hermine and Matthew, modelers at RMS are investigating whether these events provide valuable clues about future, near-term Atlantic hurricane frequency. Did either of these storms finally end the United States’ well-publicized hurricane drought?
It’s The Major Storms That Matter
The term “hurricane drought” first appeared in Geophysical Research Letters in research by Tim Hall and Kelly Hereid, and is defined as a lack of major hurricanes, ranking as Category 3 or greater on the Saffir-Simpson Hurricane Wind Scale, making landfall on the U.S.
Based on this definition, which itself ignited academic debate and drew ire from some meteorologists – the drought continues. Although a Category 4 hurricane on its approach to Florida, Matthew did not officially make landfall in the U.S. until striking South Carolina at Category 1 intensity.
Scientists agree that the Atlantic Basin entered a prolonged period of above-average hurricane frequency in 1995. To insurers, this translates into a period of increased likelihood of elevated damages and loss. As the drought stretched into record territory, scientists and insurers alike wondered whether this era had come to a close. Colorado State University’s Phil Klotzbach points out that it’s the major hurricanes, those driving the drought, that could provide us with the first clues.
An average of 2.7 major hurricanes have formed each year in the Atlantic Basin since 1950. With three major hurricanes – Gaston, Matthew, and Nicole – the 2016 season was the first since 2011 to exceed this average. But the four preceding seasons featured below-average major hurricane activity. You have to go back to the last low period of hurricane activity, around a quarter of a century ago, to see such a run of quiet years.
The Most Intriguing Statistic
This four year period is important to RMS modelers monitoring the medium-term rate (MTR), our scientific reference view of hurricane landfall frequency, looking ahead five years. It is the product of 13 individual forecast models; the contribution of each of those models is weighted according to its ability to predict the historical fluctuations in activity.
These forecast models include “shift” models that support the theory of cyclical Atlantic hurricane frequency. These shift models identify the seasons since 2011 as statistically distinct from periods observed since 1950 which are acknowledged as more active, based on the lack of recent major hurricanes.
It Ain’t Over ‘Til It’s Over
One month remains in the current season, so it is possible that more hurricanes could inform our view of the bigger picture.
The final month of the hurricane season, November has produced some notable Atlantic hurricanes, proving that the season’s latter stage requires close observation. Our attention to cyclogenesis turns primarily to the Gulf of Mexico and Caribbean, as evidenced by these past events:
- Hurricane Kate, the latest landfalling U.S. hurricane in modern history, peaked in intensity as a Category 3 storm before weakening ahead of its landfall near Mexico Beach, Florida on November 21, 1985.
- Hurricane Lenny, a Category 4 hurricane forming in 1999, is best known for its “wrong way” track that crossed the Caribbean Sea from west to east.
- Hurricane Michelle, the most intense hurricane of the 2001 season, made landfall in western Cuba as a Category 4 storm on November 4, causing over $2 billion in economic damage.
- Hurricane Paloma, following a track just to the east of Michelle, reached Category 4 strength at the height of its lifecycle, later impacting the Cayman Islands and Cuba in November 2008.
Shift models inform only one driver of activity considered by the MTR methodology. RMS plans to publish the findings of its annual medium-term rate forecast review in early 2017, after considering the most recent activity and other drivers of near-term hurricane behavior. This forecast will contribute to the updated RMS view of hurricane risk, forthcoming in spring 2017 as part of the version 17 software release.