RMS' techniques and those of the data-miners share three crucial similarities: They're only possible because of recent advances in computing power. They generate predictions about individual people and properties. And they have set off a mad scramble among insurers to slice their once-broad pools of policyholders into finer risk categories.
Insurance executives say that the rush to refine is producing positive results. It gives companies more detailed information about the risks they bear; allows them to offer lower rates to, for example, homeowners who live in safe places; and lets firms individualize policies to fit each policyholder's needs. "It gets us closer to our customers," said an Allstate spokesman.
But the ever-finer slicing appears to be having other effects as well, ones that worry a variety of regulators and insurance theorists.
States like California and Maryland have banned insurers from using credit scores, ZIP codes and other factors in deciding whether to cover someone, arguing that they unfairly discriminate against the poor and minorities. Washington state officials complain that the proliferation of categories and risk factors has so confused policyholders that the state now requires a company to provide customers written explanations whenever it gives them anything but its best rate.
"If I have a lot of house fires, [insurance companies] should charge me more," said Washington state Insurance Commissioner Mike Kreidler. "But when insurers reach and grab information like credit or occupation or education, people say, 'Wait a minute. I thought we were talking about insuring my home or auto. What does occupation or education have to do with it?' "
Some veteran observers wonder whether the intense focus on individual policyholders and properties is a recipe for business disaster.
"Insurers who look at each risk individually at the expense of broadly diversified pools are going to end up in the soup," predicted author Peter L. Bernstein, whose book "Against the Gods: The Remarkable Story of Risk" traces the mathematical origins of the insurance industry. "Diversification, not flyspecking one risk at a time, is insurers' optimal form of risk management."
Perhaps most broadly, the new techniques appear to be dismantling much of what insurance traditionally has been about.
Until now, insurance of almost every type has performed two key functions.
The first is pooling. Anyone buying an insurance policy is, in effect, kicking into a pot that covers the cost of future bad events befalling a few of their number. The second is providing cross-subsidies. Some buyers are more likely to get nailed by bad events because, for example, their genetic makeup leaves them prone to disease or their houses are not built to the latest code, and others are less likely.
But for the most part, insurers have not known which policyholders fall into which category, so they have charged generally uniform rates, which means that those in the "more likely" category get a subsidy by being able to pay the same as those in the "less likely" one even though they might end up costing more.
However, as disaster models such as RMS' and data-mining provide companies with increasingly detailed knowledge about individual policyholders, there are fewer and fewer pockets of such ignorance and therefore less and less room for cross-subsidies.
"Insurers are squeezing subsidies out of the system across the board, and they're going to carry it absolutely as far as they can," said Columbia University economist Bruce Greenwald.
On its face, the trend might seem a positive one. Among other things, it means that policyholders with good genes and safe houses can enjoy lower rates. But at least in some cases, Greenwald and others argue, the end of cross-subsidies spells trouble.
In the case of healthcare insurance, it would mean that a substantial fraction of the nation could no longer afford coverage. In the case of homeowners insurance, it ultimately might mean that large swaths of the nation's coasts become unaffordable for all but the wealthiest Americans who can bear unsubsidized rates.
And this may not be where the dismantling ends. Some analysts say that the same kind of modeling and data-mining that's helping companies squeeze out cross-subsidies could end up squeezing out much of the pooling in insurance as well.
As insurers use the new techniques to get ever-more-refined estimates of what individual policyholders are likely to cost in the future, they may be tempted to charge people closer and closer to full freight for treating an illness or rebuilding a fire-damaged home. Then even those who benefited from the end of cross-subsidies could see their rates go up as they effectively are asked to pay their own way, rather than share the cost by pooling with others.
Industry executives argue that competition among insurers will prevent such an eventuality. "I don't think you're ever going to get to the extreme of no pooling," said Greg Heidrich, senior vice president of policy with the Property Casualty Insurers Assn. of America, one of the industry's largest trade groups. But regulators are not as confident.