Golden Bear Insuance Company

How a California insurer uses catastrophe models to determine real exposures

Despite the ever-present risk of earthquakes in California, many businesses and homeowners in the state still don’t have earthquake insurance, raising concerns about the potential for mass losses if a mega quake strikes. The future may be hard to predict – scientists have been forecasting “the big one” for many years now – but one insurance provider with expertise in the earthquake insurance space is turning to models to figure out the real degree of risk.

“We use cat models primarily for managing our earthquake exposure,” said Mike Brown, VP and property manager for Golden Bear Insurance. “We are, in the property department here, about 90% commercial earthquake, and so we use the models to evaluate each individual risk and to keep track of our portfolio, manage the overall accumulation, and get estimates of what we think is realistic as a potential loss scenario.”

An important aspect of employing models is recognizing their limitations. Brown calls on a quote from statistician George Box: “All models are wrong, but some are useful.”

“There’s never been an intent on the model makers’ part to sell their models as giving you the right answer. What they’re trying to do is show you a spread of theoretically possible outcomes based on different parameters, and then the user has to sort through it all and figure out, where in this spread do I fall? Am I going to assume that of the 10,000 simulated events the model analyzed, the statistical average loss of all 10,000 is what’s likely to happen to my book of business and that’s the number I’ll key in on?” explained Brown. “You may want to bump that number up 25% or down 25% – it really becomes the user’s responsibility to sort through the big pile of statistics and drill down to what’s most appropriate for my needs at this time.”

Not all natural catastrophe models were built equally – hurricanes and floods happen a lot more frequently, thus providing more data points from real life that model developers can use, which helps them get closer to accurate predictions. Not to say that there isn’t more work to be done in wind or flood modeling – Brown has only started to see robust US flood models in the last two or three years.

“My faith in their ability to accurately predict flood losses right now comes with a big grain of salt because they’ve got brilliant scientists working on it, but I don’t think they’ve analyzed enough data points yet to compare back against their own predictions and see where their methodology is not as tight as it could be,” he told Insurance Business, adding that he expects it’ll take modelers another five or 10 years before flood predictions “become reasonably close to reality.”

As for the earthquakes that regularly shake California, some models for the coming year will show worst-case scenarios with every building crumbling while at the other end of the spectrum, other models might predict relatively minor earthquakes that will leave all buildings intact – neither of those is accurate to what will actually happen.

Catastrophe models will inevitably improve in the future and, in turn, Brown said it is reasonable to assume that the insurance industry will develop greater confidence in their estimates.

“When we look at how much reinsurance do we need for the property catastrophe program, we look at what the model anticipates the likely expected losses would be and we factor in a pretty significant cushion for uncertainty because we know that’s ultimately the right answer,” he explained. “The more calibrated the models get against reality, the smaller those levels of uncertainty can be, which would mean maybe companies pay a little less on ‘sleep at night’ reinsurance, which would allow them to look at maybe lowering rates for certain areas.”

< Back