Credit Scoring

Insurance Journal has been posting the testimony of Dr. Powell of UALR made before the House Financial Services Oversight & Investigation subcommittee last May.  Within the discussion posted is the following observation, which merits highlighting:

We are not in this hearing because everyone likes insurance scoring. I have heard critics of insurance scoring describe potential or anecdotal unfair outcomes associated with its use. I do not dispute the fact that some consumers have encountered individual rating scenarios that seem to lack intuition. For example, I know of a consumer in Arkansas who received an increase in his premium because his wife cancelled a credit card they were not using. However, he called a few competing insurance companies and found one that offered him the same coverage at a significant discount from what he was paying before the change in his credit. This is an example of competitive markets reaching an optimal outcome.

I suppose this is the wrong week to note that argument could be applied in support of rate deregulation in general….

Sometimes, the day job interfering with my extracurricular online reading can be annoying.  For example, I would have been very interested to have read this blog entry from the Palm Beach Post  yesterday:

Florida Insurance Commissioner Kevin McCarty is taking his campaign to stop auto and property insurers from denying coverage or charging higher rates to consumers with bad credit to Capitol Hill.

McCarty will testify before the U.S. House of Representatives Financial Services Subcommittee on Oversight and Investigations at 10 a.m. [Wednesday, 21 May] in Washington about the negative effects of insurance scoring on poor and minority groups.

I would have love to have caught the webcast of the hearing.   The Subcommittee has a web page specifically for this hearing, providing prepared written testimony.  Presumably a transcript of the hearing will be posted there when available.

Looking through the description of the hearing, and skimming through some of the comments, it sounds like the hearing is somewhat better-balanced than the blog post would suggest.  It’s a hearing on HR 5633, a bill to impose a federal ban on insurer’s use of credit information if it proves discriminatory or a proxy for race, ethnicity, etc.

On the surface, it seems like a silly bill, since it prohibits something that’s already prohibited…but it seems that some legislators, regulators, consumer advocates, and attorneys would choose to view the bill as a virtual ban on scoring, prescribed from the federal level.

(If we’re going to start having federal prohibitions on underwriting activities, could we go ahead and start having the option for federal regulation across the board, in lieu of the current patchwork of 57 state regulatory agencies?)

Despite what the blog post indicates, it seems that there are folks testifying on both sides of the issue.  While I may disagree with the pro-prohibition crowd…well the testimony looked like a pretty reasonable depiction of the views on the subject.

This late in an election year, I doubt that it has legs.  However, wacky things can happen, and with the insurance industry almost as unpopular as the President and Congress…well, it’s easy to imagine some federal politicians engaging in silliness to win points with constituents through “protecting” them from the evil insurance industry.

Conning is making some interesting-sounding noise about their latest paper on the nonstandard auto market. From a Conning press release:

“As predictive modeling has become more prevalent in auto insurance underwriting, the standard auto market has expanded to include and price risks that would once have been thought of as nonstandard,” said Alan Dobbins, analyst at Conning Research & Consulting. “As a result, the new nonstandard market is undergoing a dramatic shift in risk profile.”

The Conning Research study, “The Nonstandard Auto Insurance Market: Evolutionary Challenges” reviews the recent history and performance of this market, and explores the forces shaping change in the segment.

A few of us have been expecting this for a while.

Back in the day, when I worked credit scoring, I was routinely asked why insurers love credit scoring.

Well, yes, there are the very basic goals of being able to increasingly differentiate among risks, and of having a new predictive rating/underwriting variable which is rather uncorrelated with other metrics we’ve traditionally relied upon.

However, I think the transition to credit-scoring based rating/underwriting during the late 90’s and the early part of this decade is more significant as a landmark highlighting actuaries and insurers starting to look at data in different ways.

With a lot of cheap computing power at our fingertips, potent statistical packages easily available to undertake the arcane mathematical incantations, it’s an exciting time to be working, if you like playing with data.

Credit scoring is a rather prominent milestone on the road to more sophisticated multivariate analysis of insurance data. Once we get beyond the mental/psychological roadblock of certain data not being intuitively related to insurance results…some pretty exciting things become possible.

To be honest, with the arsenal of data and modeling tools now available….well there are a few actuaries in the industry who probably wouldn’t mind seeing credit scoring go away, now that we have freedom to look at and manipulate data in pretty snazzy ways.

And that doesn’t just apply to auto, or even to just personal lines. Some of the models being built to evaluate commercial lines business are pretty darned cool, at least to those of us who enjoy this sort of thing.

Seen in the New York Times:

IF you get speeding tickets, watch out: The chances are good that you will also engage in possibly dangerous investing behavior, too. That is the implication of a new study that found that individuals who receive more speeding tickets tend to churn their portfolios.[...]

These rich data sets enabled the professors to eliminate from consideration other possible causes of trading activity and focus on the distinct influence of speeding tickets. They found that, other things being equal, an investor’s portfolio turnover rate rose 11 percent after each additional speeding ticket he received. That is a surprisingly strong correlation, and is highly significant from a statistical point of view, according to the professors.

Now, you consider that in the U.S., there is a large body of people who refuse to believe in the correlation between a consumer’s credit history and expectations of future insurance losses. Yet, we have here a study that documents a correlation between financial risk-taking and driving behavior.

Hmmmm….kinda makes you think, don’t it?

Since I’m simmering over experiencing first-hand what the practical implications are of medical TPA’s being safe from bad-faith or other consumer grievance claims under ERISA (see this post and the one I’ll write once my temper has cooled a bit more), something tells me that I should restrain myself from commenting too much on this:

MSNBC’s Red Tape Chronicles has a post discussing one of Fair, Isaac’s new products:

The project, dubbed “MedFICO” in some early press reports, will aid hospitals in assessing a patient’s ability to pay their medical bills. But privacy advocates are worried that the notorious errors that have caused frequent criticism of the credit system will also cause trouble with any attempt to create a health-related risk score. They also fear that a low score might impact the quality of the health care that patients receive.[...]

Several published reports have described Healthcare Analytics product as a MedFICO score, computed in a way that would be familiar to those who’ve used credit scores. The firm is gathering payment history information from large hospitals around the country, according to a magazine called Inside ARM, aimed at “accounts receivable management” professionals. It will then analyze that data to predict how likely patients will be to pay future medical bills. As with credit reports and scores, patients who’ve failed to pay past bills will be deemed less likely to pay future bills.

And at this point, I’ll stop, since my “professional and recreational stats-geek” side and my “inner consumer advocate” are in serious conflict as a result of events of the past week. The pros and cons of such a scheme, I’ll leave as an exercise for the reader, for now.

Three articles in my reading pile today struck me as being someone interestingly related.

First, RiskProf points to a comment by Bob Hunter in the Christian Science Monitor on a new direction of attack in his war on credit scoring in insurance rating and underwriting.

“A lot of people don’t think about how this classification system works,” Hunter says. “I think most people say, ‘Oh, the rate is $50 less, I’ll take it.’ They don’t [realize], ‘The reason I’m paying $50 less is because some poor people over there have to pay $100 more.’”

Next, seen in a Toll Roads News post on the Big Dig HOV lanes:

High Occupancy Vehicle lanes (HOVLs) in the Big Dig tunnels in Boston costing around $250m are largely unused two years after opening, the Boston Globe reports. They counted 181 vehicles in a recent peak hour northbound and only 112 southbound. The Turnpike’s traffic counts show between 59 and 167 vehicles/hour are using the lanes through a broader survey period. Traffic lanes in expressway conditions like the Big Dig have a capacity of 1,600 to 2,400 vehicles per hour so the HOVLs are 90% to 97.5% empty.[...]

When the Big Dig HOVLs were opened two years ago (2005-09-23) the then Massachusetts Turnpike chief Matt Amorello said in a press statement: “One of the goals of the Central Artery/Tunnel Project was to help the environment by promoting carpooling and the use of mass transit. These ramps, as part of the Big Dig’s extensive HOV network, provide a strong incentive toward that end. With gas prices hovering around $3 a gallon, we’re confident that this change will get more people carpooling or taking the bus into Boston.”

And third is an article that ran over the weekend in the Concord Monitor:

Edwards, a former Democratic senator from North Carolina, says the federal government should underwrite universal pre-kindergarten, create matching savings accounts for low-income people, mandate a minimum wage of $9.50 and provide a million new Section 8 housing vouchers for the poor. He also pledged to start a government-funded public higher education program called “College for Everyone.”

“It is central to what I want to do as president to do something about economic inequality. I do not believe it is okay for the United States of America to have 37 million people living in poverty,” he said in a meeting with Monitor reporters and editors this week. “And I think we need, desperately need, a president who will say that to America and call on Americans to show their character.”

At every stop, Edwards said, he tells voters he’ll ask them to sacrifice. Asked to describe what he means, he described his plan for increases in capital gains taxes, saying taxes on “wealth income” should be in line with those on work income.

The thread linking these three snippets in my mind is the notion that people will voluntarily engage in self-sacrifice for the greater good.   It’s something that I’d like to believe happens…and it certainly does with some folks…but across society as a whole?  I don’t buy it.

In the first passage, Bob Hunter seems to be picking up on something a few of us have been saying for a while — that when credit scoring is implemented correctly, you end up with a majority of folks seeing reduced rates, and a minority seeing increased rates.  There’s been observed a little correlation between score and certain demographic attributes, and thus Bob seems to make the argument that if you go for lower rates at the expense of the huddled masses, you’re a Bad Person.

(I don’t entirely buy the notion within the context of insurance…or at least I think if there’s a perceived societal need to subsidize the huddled masses’ car insurance, it ought to be handled through government mechanisms, rather than by regulatory imposition of inefficiency on the free market…but I digress.)

In the second passage, I’m reminded about the apparent attitude among many (most?) Americans when it comes to conservation issues.  Maybe many folks will today voluntarily engage in the warm and fuzzy feeling of changing a single incandescent light bulb over to a CFL, or seek to buy a trendy Prius as their new car of the year…but when it comes to real conservation, including carpooling, the use of mass transit, etc., I don’t see many folks being able to be bothered.

The third quoted article reminds us that there is a horde of President-wanna-bes who perceive (arguably correctly) several tragic issues in America today — health care, increasing pressures on lower-income families, etc. — that ought to be fixed.  At least Edwards appears to be up front in some circles that such fixes will likely be expensive, ad that funding will have to come from somewhere.

Perhaps I’m just extra-cynical today, but it seems to me that in America, we love to talk the good talk and portray the image of being kind to one another and doing the right thing for society as a whole.   However, when folks are called upon to really sacrifice….well, many (most?) folks can’t be bothered to be inconvenienced.

I know that there are many examples here and there, quiet, in onesies and twosies, that challenge my gross overgeneralization….and I agree that by and large people are generally good at heart.  However, that doesn’t stop me from being…interested in our society’s tendency to present a kinder, gentler image…as long as it doesn’t cost or inconvenience us in any way.

Seen in a wire service story:

The Massachusetts insurance commissioner on Friday banned the use of drivers’ credit scores in auto insurance coverage decisions under a final set of rules opening the state’s auto insurance market to greater competition.

Commissioner Nonnie Burnes expanded restrictions on insurers’ use of credit histories after consumer advocates, the state’s attorney general and some insurers said at a Sept. 20 public hearing that Burnes’ draft regulations weren’t strong enough.

Burnes’ Aug. 28 draft proposal forbid insurers from using data from credit reports in setting individuals’ rates during a one-year transition to a new market starting next April. But her earlier proposal would not have barred insurers from considering a driver’s credit history in underwriting, or deciding whether to insure someone.

Considering the political climate in Massachusetts, this isn’t too surprising…and I have to admit that permitting scoring for underwriting but not rating is the exact opposite of the compromise struck in most other states, where rating based on credit is OK, but refusing to write due to credit is not.

From my perspective as a “credit lover”… I still don’t think this is a bad thing. To the extent that insurers are being permitted to determine their own rates, not having access to credit is fine given that access is being blocked to all competitors. Also, given that one of the concerns in Massachusetts is of massive rate swings as deregulation is phased in…perhaps having one less variable at work is not a bad thing.

I’m trying to get caught up with my reading pile which went fairly neglected in the crunch of the past few weeks. One item in that pile is this Insurance Journal article:

A ruling by the Alaska State Supreme Court requiring insurers to re-rate all policies at the very first renewal as if credit information was never considered will have the unintended result of forcing good insurance risks to pay more – to subsidize higher risk drivers and homeowners, according to the Property Casualty Insurers Association of America (PCI).[...]

Alaska law currently allows insurers to consider credit information in rating and underwriting decisions on new policies, but forbids insurers from failing to renew, or at renewal, underwriting or rating a personal insurance policy, based in whole or in part on a consumer’s credit history or insurance score.

The article continues, but the gist is that the Alaska State Supreme Court sided with the credit-unfriendly DoI in interpreting Alaska’s scoring law to mean that credit can only be used to tier or rate in the first policy term; future renewals have to assume average credit for everybody.

In such an environment, it doesn’t make sense to use credit for insurance rating or underwriting. Thus, that interpretation amounts to a backdoor ban.

I would have argued that the fact that the legislature passed the law clearly indicates that there was legislative intent to permit some use of scoring…and the interpretation supported by the Court seems to go against that clear intent.

I wonder how legislators will enjoy 2/3rds of their insurance-covered constituents getting a hike in their renewal bills as a result of the ruling.

When I started to get involved in researching and working with credit scoring at a prior employer, a couple of my reasons for doing so were a result of pressure from my inner activist—given the sensitivity of the information, I preferred to trust myself rather than relying on corporate privacy protocols (which were comparatively in their formative stages at the time), and because I had trouble believing that credit was as predictive as claimed and therefore I wanted to prove it for myself.

Well, perhaps there was some justification for that first motivation, beyond simple paranoia. Although I didn’t work for the company mentioned in this Insurance Journal story, this is exactly the sort of thing that gave me nightmares when considering data security in my prior job:

A woman who works for the company that processes Alabama Medicaid claims has been charged with stealing the identities of nearly 500 Alabamians, Attorney General Troy King said.

Kwantrice Thornton, 24, was charged with stealing the information from Electronic Data Systems Inc. as part of a Medicaid and tax fraud scheme that included selling 50 of the identities to other people, King said.

It’s times like this that even though I may rail against corporate IT security paranoia, there is some justification for some of the seemingly silly and draconian rules we get to play by.

As if there wasn’t already enough bubbling in Florida’s insurance cauldron in the form of property insurance and PIP-related issues, it seems that there’s a desire to re-agitate Florida’s “love” of credit scoring. From the Daytona Beach News-Journal:

Worried about discrimination against minorities and poor people, state regulators are considering proposals that could place new restrictions on insurance companies that use credit scores to help set rates.[...]

[R]egulators, who held a hearing Thursday on two credit-score proposals, say they are concerned the practice disproportionately hurts groups such as blacks and Hispanics. Overall, those groups are more likely to have lower incomes than whites, which can lead to financial issues that affect credit.[...]

Among other things, one of the proposals would require insurers to keep statistical information to show credit scoring doesn’t disproportionately affect people based on race, ethnicity, income and zip code.

One of the problems with that idea, pointed out in the hearing, is that insurers don’t collect information on ethnicity and income. It’s taboo information, which insurers tend to be paranoid about coming into any contact with over the potential for accusations of overt pricing or underwriting accusation. Neither insurers nor consumer advocates really want insurers to even touch such knowledge with a 10-foot pole.

The other big problem with the idea is more subtle. As far as I know, terms like “disparate impact” or “disproportionately impact”, while they may have convoluted legal definitions, haven’t been cleanly translated into a sensible, measurable statistic.

I would argue that the test for a particular variable is to address the question of whether a particular group is paying more or less in a rating scheme that uses a particular variable than they would if a hypothetical rating system that didn’t rely on that information were used. It’s certainly a doable test, but building an unbiased data set to run that sort of a test would be challenging.

However, that’s just my opinion.

Without a clear definition of what’s to be measured and tested, there is a terrible risk that data collected could be biased, and/or could be abused and misinterpreted in ways unfair to one side or the other in this debate.

Expressing a desire to confirm claims (and my personal belief) that scoring isn’t unfairly discriminatory is admirable…but without guidelines being defined and agreed to, a simple legislative expression of that desire could be a nightmarish mess.