As a catastrophe event unfolds, reliable and timely loss estimates are crucial for insurers to plan loss mitigation and resource allocation strategies, effectively communicate the risk within their organizations, and set expectations for investors. However, there's always uncertainty in the actual parameters of the event, in the estimation of its intensity at each affected location, and in the calculation of damage and loss.
To explain why modeling firms should communicate real-time loss estimates as a range, Verisk Review recently spoke to Dr. Tim Doggett, senior principal scientist, and Kathryn Fobert, science and technical writer, at catastrophe modeling firm AIR Worldwide.
Verisk Review: Using 2012's Hurricane Isaac as an example, what are some sources of uncertainty in storm parameters that affect an insured loss estimate?
Dr. Tim Doggett: On August 27, 2012, Tropical Storm Isaac had already passed through the Florida Keys, where it delivered plenty of rain and wind but caused little damage. The Gulf Coast would not fare as well. Approaching from the southeast, the storm was expected to strengthen amid low wind shear conditions and warm sea surface temperatures.
When forecasters anticipate that a storm will reach landfall at hurricane strength within 48 hours, companies look to catastrophe modelers to provide real-time loss information as the event unfolds. As a tropical cyclone approaches landfall, the forecast becomes more and more precise. That means uncertainty in the storm parameters — storm track, central pressure, and range of maximum winds (Rmax) — decreases at landfall. However, there's also uncertainty pertaining to the local intensity calculation (local wind intensity and storm surge), damage estimation, and insured loss calculation to consider. Thus, the range of loss estimates typically narrows with landfall (and even more after landfall) but never diminishes to just a single number.
Kathryn Fobert: For example, the August 27 National Hurricane Center (NHC) forecast estimated that Isaac's intensity at landfall could range from a low Category 1 up to Category 3. Furthermore, the NHC cone of uncertainty included possible landfall locations that extended from Mobile, Alabama, westward to Lake Charles, Louisiana, with the most probable landfall occurring on the Mississippi Delta area of Louisiana and extending inland over New Orleans. There was additional uncertainty with regard to the storm's Rmax, a parameter critical to the loss estimation calculation because it determines the region of strongest wind speeds and the overall size of a hurricane's wind footprint. Uncertainty in Rmax translates to uncertainty in both the local intensity calculation and the extent of damage.
(Provided by AIR Worldwide and the final Property Claim Services® loss estimate)
Each loss estimate range captures the various sources of uncertainty. As reflected in the narrowing of the loss estimate range, the uncertainty decreased as the storm approached landfall. The exception was the increase in estimated losses from August 29 to August 30, which was a result of the unexpectedly slow movement of the storm, the length of time the storm maintained hurricane and then tropical storm status, and a more easterly track than anticipated.
Once Isaac came ashore on August 29, its status as a Category 1 hurricane was known, but uncertainty remained with respect to its actual surface-level wind speeds when it made landfall, first in Plaquemines Parish, Louisiana, and then in Port Fourchon. While the NHC reported 80 mph winds at both landfalls, those estimates were derived from flight-level measurements and did not reflect actual measurements from onshore stations, which can often be significantly lower. Furthermore, there was still some uncertainty regarding Isaac's actual Rmax at landfall, along with a range of possible storm surge depths.
Verisk Review: From a modeling perspective, what was AIR's response to the hurricane?
Doggett: We issued an insured loss estimate for Isaac every day for four days, beginning August 27. (See Figure 1.) Each loss estimate range, based on a full event set of potential realizations of the storm's progress, captured the various sources of uncertainty. As reflected in the narrowing of the loss estimate range, the uncertainty decreased as the storm approached landfall. The exception was the increase in estimated losses from August 29 to August 30, which was a result of the unexpectedly slow movement of the storm, the length of time the storm maintained hurricane and then tropical storm status, and a more easterly track than anticipated (which meant it passed closer to both New Orleans and Baton Rouge than previously forecast).
Verisk Review: How are the loss estimates produced, and why are they expressed as a range?
Fobert: AIR extracts its loss estimates from a full probability distribution of losses represented by an exceedance probability (EP) curve. To produce the curve, the hurricane modeling team gathers actual storm parameters from the NHC to generate a unique event set of dozens or even hundreds of events that best represent the storm's potential progress. Our researchers carefully collect, clean, and analyze data to produce the event set. Identifying "like events" from the U.S. hurricane model's stochastic catalog would certainly be a simpler approach. But those events are certain to be different from the actual event unfolding, and even slight inconsistencies in track, intensity, or Rmax can result in major differences in losses.
(Released by AIR Worldwide on August 27, 2012)
The graph shows the EP curve along with the select scenarios for AIR’s initial Hurricane Isaac loss estimate. The low end of the range is associated with a 95 percent probability of exceedance, and the high end of the range is associated with a 5 percent probability of exceedance.
We then run the U.S. hurricane model with this event set to produce a full probability distribution of losses. From the distribution, we select 11 scenarios that best represent the full range of potential industry losses by capturing the 90 percent confidence interval of the distribution. In other words, the low end of the range is associated with a 95 percent probability of exceedance, and the high end of the range is associated with a 5 percent probability of exceedance. Figure 2 shows the EP curve along with the select scenarios for AIR's initial Hurricane Isaac loss estimate.
In addition to a certain exceedance probability, each of the 11 scenarios is also associated with a storm track and a set of parameters. We make our storm tracks, along with the associated wind speed footprints and loss maps, available for clients to view. Figure 3 shows the loss maps corresponding with scenarios 1 and 11 for AIR's final Isaac loss estimate.
In the final loss estimate for Hurricane Isaac, shown here, scenarios 1 and 11 correspond with exceedance probabilities of 5 percent and 95 percent, respectively.
Verisk Review: How can a company tailor real-time loss estimates to its own portfolio?
Doggett: While reliable real-time loss estimates are of great value to the industry at large, insurers also want the ability to estimate losses for their own portfolios. For this reason, the select scenarios and the full event set are available for clients to download. With the goal of enhancing both the interpretation and communication of loss estimates, AIR recommends that insurers use the full event set to generate their own EP curve, thereby allowing them to present both the probability of loss and associated uncertainty.
The primary purpose of any catastrophe model is to prepare insurers for potential losses before they occur. Using a full event set, carriers are able to address specific business needs, such as determining if reinsurance is adequate, managing reserves, understanding where to suspend or continue to write business, and knowing how to deploy claims resources most effectively.