A View From The Press Box

This whole controversy reflects a difference of philosophy about the purpose of insurance. Is it supposed to be a risk- sharing mechanism or a wealth redistribution mechanism?

The first position holds that every policyholder should pay a premium that reflects his risk of loss as closely as it can be determined.

Since an individual's risk of loss cannot be determined precisely (if it could, it would not be a risk any more, it would be a certainty that simply has to be funded for, not insured), it can only be estimated through statistical analysis of large groups of individuals to determine the characteristics that correlate with losses, and to set premium relativities accordingly.

The other opinion holds basically that richer policyholders should subsidize poorer policyholders, without a lot of regard for relative risk as estimated by statistical analysis. They want to use insurance as a wealth redistribution scheme.

Both objectives may be good objectives, but mixing too much of the latter into an insurance system will destroy it.

If what you say, MH, is true (that one group--rich policyholders--shouldn't have to subsidize another--poorer policyholders), then why should good drivers, who for a myriad of reasons might have lower FICO scores than hoped for, have to pay for bad drivers, who might have higher than expected (by the carriers) scores?

What was wrong with the old reliable rating system--in my opinion--where a driver who exhibited poor or reckless driving skills by accumulating points/violations and accidents, which resulted in higher than average claims experience, paid more than a good driver?

It wasn't perfect, but pretty accurately equated bad driving experience with bad loss experience.

Not only is this new system--insurance scoring--extremely unpopular with the consumer, it is based on a fundamentally flawed base--the credit report.

That fact that one of five FTC commissioners dissented from the commission’s report is hardly cause “to wonder what credibility the report has left.”

In the first place, Commissioner Harbour has something of a penchant for filing dissenting statements, having done so more often than any other commissioner since she was appointed in 2003. What is more, even a cursory reading of her dissent reveals that Ms. Harbour--a career antitrust lawyer--has little understanding of advanced statistical research methods.

Far from sounding “hollow,” the majority’s point-by-point refutation of Ms. Harbour’s gratuitous critique suggests that her “grave methodological concerns” reflect either ignorance of standard research methods, or a personal bias against credit-based insurance scoring.

Mr. Daroff asks, “What was wrong with the old reliable rating system…?” The answer is that while it may have been reliable, it is not nearly as reliable as a rating system that utilizes credit reports.

That is the incontrovertible conclusion reached by the FTC study and all previous studies, including one by the University of Texas and another by the Texas Department of Insurance (which Commissioner Harbour inexplicably endorses as a “template” for how she believes the FTC study should have been conducted).

Bob Detlefsen responds to Steve Daroff:

'“What was wrong with the old reliable rating system…?” The answer is that while it may have been reliable, it is not nearly as reliable as a rating system that utilizes credit reports.'

YES, exactly. People really don't get this, but it's key. Credit scoring accurately predicts risk classes DESPITE the errors. This is how a robust risk measure ought to behave. The paper I cited in the previous thread may shed some light as to why this is.

Driving record is important and a good variable, but it's not as objective as you think.

Say I have a pool of 100 drivers, and somehow we know that all of them have an intrisic probability of 5 percent of having an accident (only one per year). We look at their record over 10 years.

Because we "know" that the probability of an accident is the same for each driver, each should be charged the same "rate"--0.5 accidents over 10 years. But through nothing but dumb luck, the most likely distribution of outcomes is that 60 people are accident free, 32 have one accident, seven have two, and one unlucky person has three.

Eight of these 100 people are likely going to be charged more than the other 92, even though their risk is the same.

Of course, we don't know the intrinsic probability of an accident--we use driving record to measure that! But as I show, this method is not perfect.

Driving record plus credit score together is hugely powerful, as that paper I cited before shows.

Todd R. Bault, FCAS
Sanford C. Bernstein, LLC
1345 Avenue of the Americas
15th Floor
New York, NY 10105
Ph: 212 756-1857
Fx: 212 848-2370
todd.bault@bernstein.com

The comments herein are part of a larger body of investment analysis. For our research reports, which contain information that may be used to support investment decisions and appropriate disclosures, please see our website at http://www.bernsteinresearch.com

I openly admit that my grasp of advanced statistical research methods is tenuous, at best. But there is an old axiom that I am aware of regarding statistics which I will not waste precious space on, and I am sure, we all know what that is.

Correct me if I am wrong, but if you look at an individual who has a low credit score, for whatever reason, but has a good driving record over a period of time (let's say at least 5 years), you are predicting that he/she will file an above average number of claims in the future?

Common sense tells me that, statistical probabiliity or not, that it isn't true.

Until I see real life scenarios where credit issues have changed a person's driving experience dramatically for the better or worse, I will remain entirely skeptical.

If statistical research is so predictive, how come political polling is not that accurate? Does the name Harry Truman come to mind?

"If statistical research is so predictive, how come political polling is not that accurate? Does the name Harry Truman come to mind?"

Political polling tends to be extremely accurate. The "Dewey-Defeats-Truman" headline to which you allude is an outlier that occurred more than 50 years ago, when polling techniques were far less sophisticated than they are today.

What's truly amazing, though, is your categorical dismissal of "statistical research." Don't you know that all insurance underwriting is based on statistical research and analysis? Have you not heard of the "law of large numbers"?

The alternative to objective statistical analysis is to rely on the subjective impressions of individual underwriters (what you refer to as "common sense").

Reasonable people can argue, from a normative standpoint, about which variables ought to be used to analyze risk. But to suggest that risk shouldn't be assessed through statistical analysis is ludicrous.

To Bob:

I didn't say that "risk shouldn't be assessed through statistical analysis." I said that I thought driving record and claims experience analysis is more accurate than credit score analysis in determining what a policyholder's future claims experience will be.

Until someone does a side-by-side analysis of the two methodologies, I continue to hold to my opinion.

If you know of one, please let me know and I will be glad to look at it and possibly change my opinion.

I think the industry, which I have been a part of for over 30 years, has rushed to judgement on this approach to underwriting risk, and I think the industry's approval rating has declined precipitiously due to this issue, Katrina claims, and contingency fees.

If we have proof of the above, we need to get it out to the public, and quickly. Our PR people have not sold the consumers on this concept, as well as a few P&C carriers, who are not sold as well.

To Steve:

It's not really a question of which variable--driving record and claim experience or credit score--is "more accurate." No insurer relies solely on credit scores to assess risk.

The relevant question is whether using credit scores in conjunction with more traditional rating variables provides greater accuracy than if credit scores aren't used.

Both the FTC report and the TDI report are unequivocal in concluding that credit scores provide additional predictive information distinct from other variables.

If you don't trust the FTC study, you should look at the TDI report (http://www.tdi.state.tx.us/reports/documents/credit05sup.pdf).