Who We Are
How You Can Help
Insurance Credit Scoring
1701A S. 2nd Street
Austin TX 78704
(512) 912 1327
(Fax) 912 1375
Statement of Vickie Benitez
My name is Vickie Benitez. I am the Executive Director for the Center for Economic Justice, a Texas-based non-profit organizations dedicated to advocating on behalf of low-income consumers on insurance, credit and utility issues. Thank you for the opportunity to discuss the issues of credit history and insurance underwriting and rating today.
I am also a funded consumer representative. If not for the NAIC funded consumer representative program, I would not have been able to participate before you today. We want to thank the NAIC for the funded consumer representative program.
My discussion today will focus on the following:
Insurance redlining is a reality – consumers in poor and minority neighborhoods are systematically denied coverage in lower-cost standard companies and forced to purchase higher-cost policies from non-standard companies or residual markets.
The use of credit scoring, and other computer underwriting tools, contributes to the redlining problem.
We are simply amazed and bewildered by state insurance regulators’ failure to meaningfully examine the use of credit histories by insurers and, particularly the use of the credit scoring models.
Since the NAIC white paper was issued two years ago, the only new information about credit scoring models reinforces our concern about the use of those models.
We urge regulators to carry out their regulatory responsibilities and we want to praise Commissioner Larsen for taking the lead on these issues in his state of Maryland and trying to shed light on the credit scoring black box.
Let’s start with some basic facts about insurance redlining. The accompanying charts show that the poorest consumers are systematically denied coverage by lower-cost standard insurers and forced to purchase much higher-cost policies from non-standard insurers or the residual market.
The chart groups ZIP codes by percentage of all insured vehicles that are insured by non-standard companies or the assigned risk plan. Stated differently, the share of insured vehicles in non-standard companies and the assigned risk plan represents the rate of rejection by lower-cost standard insurers. The table shows the auto rejection rate increases as the minority population of a neighborhood increases and as the median household income decreases. Even holding income constant, we found that consumers in high-minority areas were three to four times more likely to be rejected by the standard market insures than consumers in low-minority areas.
How can these results occur? Do bad drivers decide to live in the same neighborhoods? Are minority drivers worse drivers than white drivers are? Are poor drivers worse drivers than affluent drivers are? Common sense, as well as data, say no. Accident data from the Texas Departmen tof Public Saf3ty show that minority drivers are not involved in a higher propoertion of accidents than their share of the population. Then how can these disparate results occur when insurers are simply using "objective" measures to evaluate drivers? Why would the poorest consumers choose to purchase insurance from the highest-cost insurers?
One of the answers, surely, is insurers greater use of a variety of detailed demographic, economic and geographic data and cheaper computing in the relentless quest for the "best" risk. The use of credit histories and credit scoring models in underwriting and rating is just the tip of the iceberg.
Consider the recent announcement by Fair, Isaac and the insurance industry about the "exciting" opportunities to pre-screen consumers based upon demographic characteristics and credit histories.
A June 1998 Fair Isaac press release triumphantly says:
TargetScore differs from other insurance marketing sources in that it uses more comprehensive demographic data and provides two important profit measures, rather than just one. It permits marketers to target consumers based on two key indicators of policyholder performance – their predicted loss ratio and likelihood of retention. Using these scores, insurers can identify and target those prospects with the highest potential life-time value. They can also structure their offers to appropriately match products and pricing with each prospect’s anticipated level of profitability."
That’s right – match products and pricing with each prospect’s anticipated level of profitability. I wonder what products and pricing are matched with lower-income consumers?
A June 1998 NAII press release, entitled, "Credit Prescreening Offers Insurers Exciting Possibilities" says:
For insurers that want the underwriting benefits of credit reports, but don’t want to deal with the regulatory hassle that sometimes accompanies their use, prescreening may be the answer. . .. Sorich noted the difficulties insurance companies have when using credit reports. "The states have broad authority to develop a variety of mandates, requirements and prohibitions on how insurance companies can use credit reports," said Sorich. "However, this broad authority does not apply to prescreening.
NAII claims to be a staunch supporter of state regulation of insurance. This press release shows otherwise.
Free Pass to Use Credit Scoring
Given that redlining is a reality, we again ask the question, why have state insurance regulators given insurers a free pass on the use of credit histories and, in particular credit scoring models?
If an insurer submitted a rate filing that relied upon some loss costs developed by ISO, regulators would look at the ISO loss costs to either see if they were reasonable or if they had been previously approved for use by insurers. What would happen if the regulator were told by ISO that the analysis supporting the loss costs was a trade secret and ISO would not reveal the underlying data or analysis to the regulator? Instead of providing a loss cost filing for the regulator to evaluate, ISO simply provided its own study showing the correlation between its loss cost factors and insurer loss ratios. By now, you are thinking, this is an absurd hypothetical situation – no regulator charged with approving or reviewing rate filings would allow insurers to use an advisory organization’s prospective loss costs, if the loss costs were presented as a black box model. And no regulator with responsibility for approving or reviewing rate filings would allow such a thing.
Yet, this bizarre "hypothetical" scenario is precisely what insurers do with the Fair, Isaac scoring model and, for at least six years, insurers have gotten away with it.
What about the use of computer catastrophe models. Florida set up a special commission to study computer catastrophe models before the models could be used by insurers as support for rate filings. The commission looked at the guts of the models – data, assumptions, and results. In California, intervenor parties in hearings on rate filings are able to obtain access to detailed descriptions of computer catastrophe models as part of the process of evaluating rate filings. The Texas Department of Insurance has simply prohibited insurers from using computer catastrophe models as support for insurer rate filings because; "The uncertainties associated with models and the lack of publicly available detailed information concerning existing models necessitate this action. The Department is required to ensure that the use of catastrophe models will not produce rates that violate the standards . . . that rates must be just, reasonable, adequate and not excessive for the risks to which they apply."
We are pleased that the State of Texas is not willing to allow insurers to use black box computer catastrophe models without examining those models in detail.
Why hasn’t his happened for the credit scoring models? Why have insurers been given a free pass on credit scoring?
It is obvious to us that Fair, Isaac is an advisory organization, providing exactly the same type of services as ISO. Why is ISO subject to state regulation, but not Fair, Isaac? Why haven’t states addressed credit scoring by regulating Fair, Isaac as the advisory organization that it so clearly is?
What Have We Learned Since 1996
Two years ago, the NAIC issued a white paper on the use of credit histories in insurance underwriting. The white paper expressed concern over regulator’s inability to examine the credit scoring models. It is now two years after the paper was issued, three years after most of the work was done. To this day, have insurance regulators examined the Fair, Isaac scoring model? No.
We do have some additional information. We know that the credit scoring models have unfairly discriminated against consumers. Earlier this year, we learned that credit scoring models were giving lower scores to consumers for a particularly egregious behavior by those consumers – the consumers were shopping around for the lowest rate. A February 1998 news article reported the following:
"Though Fair, Isaac and Co. has never publicly revealed technical details about how its statistical models work, it confirms that inquiries do count as a "risk factor." In general, the more inquiries you have, the greater the probability that you are seeking to increase your access to more credit, and possibly increase your total indebtedness. That, in turn, can depress your FICO score."
This may sound reasonable, but there are significant problems with this process.
"Some mortgage applicants have claimed that their scores were unfairly lowered by credit-check inquiries they never knew about, or that simply represented intensive shopping for a car, furniture, or a mortgage."
The article reports that, as a result of the issue being raised, Fair, Isaac changed their scoring model to prevent consumers from being penalized for shopping around.
What this story illustrates is that the credit scoring models -- whether for mortgage lending or insurance – contain a number of assumptions that may result in unfair discrimination against consumers. The penalty for being a good consumer and shopping around is such an example. Once the problem was identified, Fair, Isaac allegedly responded by modifying the credit-scoring model. Given this example of a glaring problem with the credit-scoring model, it is certainly possible – if not likely – that other problems exist in insurance credit scoring models. Consumers certainly will not know – they are simply provided a higher rate quote. Consumers must rely upon insurance regulators to examine the credit scoring models. Is this model reliable enough to justify that examination as a routine expense? Moreover, even if problems are identified, how will consumers know that problem has been fixed by Fair, Isaac unless a regulator can review the credit-scoring model?
To illustrate this problem with auto insurance, consider the following story told us by a Texas insurance agent. A consumer started shopping around for auto insurance. The first quote of $1100 quote was from a Progressive agent. Progressive is known for being an active user of credit histories in auto insurance rating. The consumer contacted several other agents and companies for quotes. The sixth quote was from another Progressive agent. But now the quote was $1700. It seems that in the process of shopping around, each agent or company did a credit inquiry and by the time the consumer had gotten around to the second Progressive agent, the additional credit inquiries had lowered the credit score, resulting in a higher premium.
What else do we know today that we did not know two years ago? In March 1998 the United States Public Interest Research Group released a study regarding the accuracy of information in consumers’ credit history files. One of the findings of the study, entitled "Mistakes Do Happen: Credit Report Errors Mean Consumers Lose," is:
Twenty-nine percent (29%) of the credit reports contained serious errors - false delinquencies or accounts that did not belong to the consumer - that could result in the denial of credit.
The prospect of almost 30% of consumers being unfairly denied insurance because of mistakes in credit reports is staggering and clearly requires action by the insurance regulators stop and prevent such unfair discrimination. The executive summary of the study is attached for your review. The entire study is available on the Internet at http://www.pirg.org/consumer/credit/mistakes/index.htm.
What about the "studies" produced by Fair, Isaac and the insurance industry purportedly showing the "correlation" between credit scoring and loss ratios? Well, we know that the Tillinghast study commissioned by Fair, Isaac in 1996 was not responsive to the NAIC’s requests and that the white paper working group did not find the study helpful or conclusive. Incredibly, the author of the Tillinghast study, Wayne Holdredge agreed with the NAIC!
In a March 9, 1998 issue of BestWeek: Property Casualty Edition, Mr. Holdredge writes:
Regulators wold like the industry to conduct additional multivariate analysis that would "neutralize" the difference in losses that might be accounted for by the usual rating variables.
Tillinghast produced the analysis for Fair, Isaac, the firm that provides credit scores to the industry. We understand quite well the NAIC’s desire to know more. We believe the arguments for going ahead with additional research – regardless of who conducts the further analysis – are compelling.
Rather than fighting the request, the industry ought to act on it as quickly as possible.
What else do we know now that we didn’t know in 1996? We also know that new models are being developed to screen potential consumers. There is no need to deny coverage -- just don't even offer it in the first place. These models are even bolder --
prescreening based upon demographic characteristics. Again we ask, how can the insurance industry claim to be regulated by the states when activities like these go unscrutinized by state regulators?
We conclude by expressing our frustration that credit-scoring models have gotten a free pass by state regulators. Could insurers use a black box for prospective loss costs? Would any regulator accept ISO loss costs if ISO said its analysis was a trade secret and not available to regulators? Would regulators accept computer catastrophe models without an analysis of the structure and assumptions of those models? The answers to these questions are clearly "No". Why do credit-scoring models get the free pass?
Why isn't Fair, Isaac being regulated like an advisory organization? It takes historical experience from insurers, analyzes the experience and provides rating information back to insurers. These activities are no different from those performed by ISO in developing loss costs and ISO is regulated as an advisory organization in every state.
We call on regulators to follow Maryland's lead and require the credit scoring models to be revealed to regulators, that the models be tested for fairness and if the modelers won't participate, prohibit insurers’ use of the models.