As Hurricane Irma struck Florida this weekend with intense winds, commercial and residential property insurers were doing their best to prepare with the use of computer-generated catastrophe models.
“Compared with the events in 1992 surrounding Hurricane Andrew,” reports Werner Kruck, COO of Security First Managers, Florida’s second largest homeowner insurance carrier, “insurers are much better equipped to respond.” A key reason, he notes, is the widespread adoption of computer-based storm modeling.
But all modeling is not the same. Security First relies on six different models, which it runs at least twice a year to help manage its risk exposure and pricing. Of those, only one can be run more frequently during an actual event.
“The only model that we can run on demand, that actually projects claims and not just financial damage, is RiskInsight from Karen Clark,” says Kruck. Last year, when Hurricane Matthew was developing, “We were running the model every four hours. That gave us the ability to adjust our expectations and prepare our claims people,” he explains.
It also allowed the insurer to issue advance warnings to its policyholders, along with recommendations on how best to prepare for the storm, followed by early damage assessments and claim filing instructions in its immediate aftermath.
“The two things that are different about the Karen Clark model is that it projects actual claims and provides actual wind speeds at each of our insured locations,” explains Kruck. “This allows us to adjust our expectations on the fly.”
The Clark model achieves this by updating the hurricane’s signature every four hours. That’s in contrast to other computer models that use the latest stochastic storm in their catalog to approximate the hurricane’s outcome, which Kruck says simply “is not as accurate.”
The insurance industry first began making extensive use of cat modeling in the wake of Hurricane Andrew, when insurers grossly underestimated the $15 billion in damage wreaked by the storm. But the original process, a big advance at the time, hasn’t changed very much and doesn’t take advantage of the latest advances in data collection.
A throwback to Andrew
“Today, the industry is in a similar position as it was before Andrew, when cat modeling was still largely unproven and had yet to be widely adopted,” observes Karen Clark, CEO of the eponymous Karen Clark & Co. “Today’s models provide extensive information, but give the industry only one or two numbers to rely on.”
The key number in question is an insurer’s PML, or probable maximum loss, due to a major weather event. Clark argues that while useful, the PML alone can mask an insurer’s exposure and provide a false sense of security.
To complement it, KCC launched a new methodology for estimating and managing catastrophe losses, which it calls the Characteristic Event approach. With CE cat modeling, hazard probabilities are quantified and losses estimated for different event parameters. This is the reverse of the now standard Exceedance Probability, or EP, modeling approach—which provides probabilities for the hazard as opposed to the losses. Using CE, Clark says, insurers can readily determine their exposure concentrations and identify any hot spots.
To help prove her point, Clark asserts that KCC is the only modeling firm to date that has released detailed insured loss figures for Hurricane Harvey, including separate estimates for wind, storm surge and inland flooding damages. The company will release similar figures for Irma, she says, within a few days of the storm’s passing.
In the meantime, KCC’s clients are making use of RiskInsight, KCC's CE implementation, to try and get on top of the events surrounding Irma. “As you can imagine, there’s a great deal of urgency,” Clark says.
