I’d love to say yes.
And Karen Clark’s study of cat modeling, released this week, points that way. Fort Lauderdale’s Sun-Sentinel:
Hurricane risk prediction methods that property insurers use to help calculate premiums were off by $34.8 billion to $53 billion the past five years, according to a new report by the founder of the method.
The report is here. And here is the report’s money chart, which shows just how far off the modeling firms have been:
The first table and chart show how much the major modeling firms overshot the number of storms reaching the U.S. The second table and chart show how much they overestimated losses. From the latter table, you can see that all three modeling firms predicted losses above $60 billion over the period. Actual losses were less than a fourth that, or $15 billion.
The implication: Modelers projected too high. The rates insurers set – based on the models – were too high.
Update: RMS responds here (pdf).
The actual story, and Clark’s message, is quite a bit more nuanced.
Karen Clark (the company bears the founder’s name) doesn’t have a beef with cat modeling. Karen Clark (the person, not the company) helped create the cat modeling industry in the wake of Hurricane Andrew in 1992.
She does have a problem with the modelers’ short-term projections. Those projections indicated that 2006-2010 would sustain 35% more hurricanes than average. Of course, that didn’t happen.
Clark suggests that modeling isn’t sophisticated enough to make near-term frequency projections. We still aren’t certain whether the recent increase in hurricanes is a real trend or a byproduct of more sophisticated weather monitors at sea. (Decades ago, if a hurricane didn’t reach landfall, we might well never have known it existed. Today, satellites find the hurricanes forming before they’ve cleared the East Atlantic.)
A roundup of responses from cat modeling firms is here.
Even as the tussle ensues, it’s a reach to say that Florida homeowners were overcharged.
Clark’s comparison starts in 2006. The chart below shows loss and ALAE ratios by calendar year for Florida homeowners’ business.
Clark’s study looks at five years where Florida dodged the hurricane bullet. Now, she’s not cherry-picking the data. The cat modelers didn’t develop five-year projections before 2005, and the five-year projections came about, in part, because of the enormous hurricane losses of 2004 and 2005 (seven major storms came to shore).
So Clark looked at the five years that the modeling companies picked out, and they were off the mark. But in her paper, Clark seems to prefer the long term average. If you scroll up to the first image of this post, you’ll see a skinny yellow line – the long-term average – and it projects $50 billion in losses.
So the cat modelers projected $60 billion, the long-term average projected $50 billion and the actual losses were $15 billion. The long-term average and the cat modelers were closer to each other than they were to what actually happened.
That’s not too surprising. Cat models are, in Clark’s words, “blunt tools,” trying to put a point estimate on a highly variable process. I’d add that if their five-year forecast were correct it would be far more remarkable than just about any amount, in any direction, that they might miss by.
Clark is trying to emphasize what the models do well, and what they don’t:
Catastrophe models are powerful, broad-based tools that are very good at:
- Providing a framework for tying together the principal components of catastrophe risk: hazard, engineering, and exposure
- Providing numerous scenarios that produce estimates of losses from different types of events
- Determining approximate estimates of losses associated with events of different magnitudes
- Providing a general indication of relative risk
However, catastrophe models are characterized by high uncertainty, and thus cannot:
- Produce accurate point estimates of infrequent events, such as the 1 in 100-year loss
- Produce credible, robust estimates of losses at specific locations
- Predict near term catastrophic losses
I’m sympathetic to the argument, but what’s an insurer to do? To develop a rate, a company has to estimate expected losses from catastrophes. Flawed as they are, cat models are an improvement over the old methods, which got blown away with Hurricane Andrew, 19 year ago. And Clark is emphasizing the fact that the models overshot their estimates. They often come in too low, which is one reason Florida’s HO insurance is, if anything, too cheap.
As evidence of the, believe it or not, bargain Floridians get, I point to the next chart, which compares loss and ALAE ratios from Florida HO business to that of the rest of the country:
It’s a little hard to scale this chart properly, which is kind of the point. Most years, Florida’s loss ratio is lower than that of the rest of the country. But when the hurricanes hit in 2005, Florida’s loss ratio was double the rest of the country, and a year earlier, the ratio was six times the national average.
Over the 10 year period, Florida posted a 76% loss ratio, while the rest of the country posted 65%. Just looking at those two ratios – ignoring the other elements in an actuarial pricing – tells you Florida rates should be 17% higher.
(That assumes that HO rates are adequate nationwide. They probably aren’t, but that’s a topic for another day.)
But the price you pay for insurance depends not just on the losses and expenses that the insurer will incur. It also depends on the amount of capital the insurer needs on hand in case a disaster strikes. The best measure of that is variability. The next chart compares the variability (measured by the standard deviation) in Florida’s loss ratio with that of the rest of the country.
Yeah, Florida’s results are eight times more variable than the rest of the country. The more variable the results are, the more capital an insurer needs, and the riskier the investment. And the more risk the insurer takes, the more it will require in profits as reward.
So without a lot of fumbling, one can legitimately justify charging Floridians 25% more than they currently pay. And that paradox – that Floridians pay a lot for homeowners’ insurance but still get a bargain – drives the entire, bizarre market.