On Oregon Measure 42 and Insurance Credit Soring In General

I’ve been poking around online, looking for poll results on Oregon Measure 42, the anti-insurance scoring initiative on the ballot in Oregon this November.

I still haven’t found any publicly released polls online, annoyingly.

While I was searching for polling data, I did, however, find a ton of bad or incomplete information. This annoys me, as I believe, perhaps naively, that voters should be provided with factual information upon which to base their vote upon.

A prior job of mine was working closely with credit scoring for insurance — all the way from building the model, to incorporating it into our rating, to talking about it with agents, regulators, and consumers. In other words, I know a little something about the subject.

In my current job, I don’t work with personal lines, and I work at a company where PL is relatively minor and where we don’t do much business in Oregon. So, I don’t really have an immediate interest (other than professional curiosity) as to how the Measure 42 vote goes.

With that said, below the fold, I’ll offer my thoughts on some of the points being raised on Measure 42.

First, I need to begin with an obligatory disclaimer: What I’m about to write is my opinion. It is not necessarily shared by my employers past or present, the insurance industry, or the actuarial profession.

With that said, here are some of the points I’ve heard regularly raised when discussing credit scoring (and Measure 42), with my thoughts on each:

Credit scores can’t possibly predict insurance losses. That’s actually what I thought when I started working with scoring — it makes no sense; how can it possibly work? As hard as it may be to believe, but there is a correlation between score and future losses. Two of the better public studies I’ve seen were put out by Epic Consulting (report, appendices) and the University of Texas (report).

The EPIC study was an industry-funded study drawing countrywide data from auto insurers representing roughly one-third of the U.S. market. The UT study was commissioned by the Texas Department of Insurance and looks at the largest insurers in that state. In both cases, the message is the same: all other things being equal, score is predictive of loss. Here’s a chart from the Texas study:

In English, what the chart is saying is that the 10% of the population that has the lowest scores saw losses 53% higher than average. The best 10% saw losses 25% better than average.

But…that still doesn’t explain how credit causes insurance claims. That is correct. However, as someone who has designed insurance pricing systems, I have to say that I’m not overly concerned with what causes losses. Measuring attributes that cause a loss is difficult and expensive, especially when you consider that it’s virtually impossible to absolutely prove causality.

What I am concerned about is correlation. If I can show that some objectively measurable attribute is correlated with loss in a way that other attributes aren’t, then I can come up with a more accurate prediction about your odds for a loss, which I use in pricing.

While I don’t know why scoring works, I do have a theory.

When pricing homeowners or auto insurance, many of the variables used get at different types of attributes correlated to loss. I can get some idea of how much driving experience you have by looking at your age. I can get some idea of where you drive by looking at where you park your car at night. I have some idea of how prone your house is to windstorms, hurricanes, and earthquakes by looking at its zip code. And so forth.

What I lack is information about your behavior. How carefully do you drive? Do you tailgate? Are you prone to road rage? How diligent are you about performing home maintenance? How long am I likely to keep you as a customer (or over how long a period of time can I spread the up-front costs of writing your policy)?

Those are all questions that are intuitively related to your exposure to future loss (or my future profit). But they’re also questions that are difficult to answer, unless I want to be incredibly intrusive (installing a black box in your car, or having an underwriter visit you periodically) and/or unless I want to make the application and renewal process incredibly onerous and expensive.

Credit scores are a way to measure an aspect of your behavior. That aspect is, I believe, correlated to other behavioral traits you possess, which are in turn correlated to the risk that you will file an insurance claim, as evidenced by the data.

Yes, it’s a weak correlation…but as you saw in the chart above, it’s still enough to justify a pretty sizeable discount or surcharge at the extreme ends of the spectrum.

Consumers will pay more/less for insurance if credit scoring is banned. The correct answer to that claim is: it depends.

Let’s take a look at the Texas study chart again:

In this sample, the bottom 3 deciles (i.e., the worst-scoring 30%) have worse-than-average expected experience. They’re probably going to be charged more for insurance under insurance scoring than they would be otherwise.

The top 7 deciles, on the other hand, have better-than average experience. They’re probably going to be charged less.

That’s roughly comparable to other stats I’ve seen. In my experience, about 2/3rds of insurance buyers pay less because of credit scoring, and 1/3rd pays more.

If credit scoring were banned, that would mean (all other things being equal) that 2/3rds of all insurance customers would see a rate hike because of a ban, and 1/3rd would see a rate cut.

That’s more than just an academic statement. That is exactly what happened when credit scoring was banned in Maryland for homeowners insurance a few years ago. Some customers were not happy when I had to tell them that we were going charge them more for insurance because we had to remove the discount they had been getting because of a change in state law.

I have seen the proponents for Measure 42 respond to this concern by claiming (essentially), “look at California; they aren’t paying more and credit scoring isn’t allowed there”. That claim is a red herring. Credit scoring has never been permitted in California, and therefore has never seen the widespread changes caused by adopting or removing scoring.

Arguably, the reason that personal insurance prices are dropping now in California is because tough rate regulation has made insurers reluctant to take rate cuts when experience has permitted, for fear of having problems getting the rate back when claims inflation arises in the future. Insurers are only now cutting rates due to political pressure; Commissioner Garamendi needs votes to be promoted to Lieutenant Governor, after all.

Insurers are using credit scores to screw consumers out of more money. While I can understand how it’s tempting to see insurers as a dark, evil aspect of corporate capitalism…scoring isn’t a tool to screw consumers. If we’re trying to screw anybody with credit scores, it’s other insurers.

The introduction of credit scoring to insurance in the U.S. has come at a time when many insurers have developed increasingly sophisticated insurance pricing algorithms. Most of the big insurers believe that they have found a way to build a better pricing mousetrap, and they try to use their new tools to attract profitable customers from other insurers. Credit is in most cases a vital part of those mousetraps, and the increased competition helps keep prices lower than they would otherwise be.

Even for “less profitable” customers, the competitive situation has improved, I think. Some insurers have made a mark on the market by going after the customers that the traditional “preferred business” insurers find less attractive. Here, because of competition among the new players, prices are kept in check.

Among all these new pricing mousetraps is something that I think too many people don’t realize — there is generally no monolithic school of thought among insurers about how a particular risk “should” be priced. If one insurer doesn’t find you to be an attractive risk, another probably will.

If you think that your insurance company is asking for too steep a premium, it is in your best interests to shop around. There are plenty of websites now where you can go to get auto and homeowners insurance quotes. If you don’t want to do the work yourself online, your friendly local independent agent should have the contacts necessary to get quotes from several different companies, to find you a lower price.

Isn’t credit scoring a way to discriminate against poor people and minorities? I’ll happily and in good conscience swear on a stack of Bibles that I’m aware of no evil plot afoot to intentionally engage in illegal discrimination. If there were one, I probably would have heard about it, and someone would have blown the whistle on it a long time ago.

The more interesting question is whether there’s a subtle, unintentional form of discrimination at work. There have been studies that purport to show one way or the other…but frankly, all of those studies have been lacking.

The closest I’ve seen anyone come is to suggest a correlation between score and locations with certain demographic traits, but such a correlation isn’t necessarily damning, and it probably exists with other well-accepted rating elements (e.g., the good student discount in auto insurance).

A better question to ask is this: Does [a protected group] pay on average more or less for their insurance in a rating system that has credit scoring than in a system that doesn’t?

That is a question that is, as far as I know, unanswered. Investigating that question would require data that insurers are currently barred from collecting (i.e. details on ethnicity and income), and it should span a broad segment of the market.

(It’s possible that the FTC is working on such a study as part of its mandate from the FCRA reauthorization of a few years ago, but that happened about the time I changed jobs and I’ve been a little out of the loop since then.)

My gut feeling is that credit scoring would pass such a test, provided insurers were exercising care in their use of scoring. State Departments of Insurance can verify that that care is being exercised as a part of their routine rate regulation.

Aren’t credit reports filled with inaccurate information? Well, it’s true that the PIRGs and Consumers Union like to tout a claim that an astoundingly large number of credit reports have at least one error in them. And, while I can’t opine on the accuracy of the exact claim, I can confirm that from my own experience, there is a lot of crappy data in credit bureau files.

(The data isn’t as crappy as, say, DMV files, but I digress.)

However, most of those errors are known shortcomings within the credit reporting systems. Birth dates, employment information, even some address information contained in credit reports can be very wrong…which is why that data isn’t used in credit models.

Other issues can arise in how data is reported into the credit bureaus. For example, following a mortgage through a person’s credit file as it has been sold and re-sold or financed and re-financed can be a very messy experience. However, developers of credit models are aware of these headaches and design the models to compensate.

And yes, there are some blatant, outright errors in the credit data. Sometimes the error is to your benefit, and sometimes it’s not. That’s why users of credit data are supposed to inform consumers when an adverse decision is made; if the decision is based on bad data, corrections can be made.

The fact that scoring works in spite of the data issues is a testament to its power. I can find meaning within the noise.

If credit file accuracy were the only issue with scoring, a better response would be to seek to compel the entities that report data to the bureaus to do a better job of reporting. Those issues cause more impact on consumers through how they manifest in credit card and mortgage interest rates, for example, than they do through insurance rates.

Besides, cleaner data would make it possible for actuaries and statisticians to develop better scoring models, which in turn opens the door to possibly lower rates for some.

The wording on Measure 42 is flawed. Yes, I agree that the measure is sloppily worded. For example, as it is written, it could be construed to extend to commercial insurance, where credit risk can be more intuitively related to future insurance claims, and where most of the anti-scoring regulators seem to accept that credit has some justification.

However, if I were an Oregon voter, and if political environments operated in a rational manner, the sloppy wording aspect wouldn’t necessarily influence my opinion. I suspect that that issue could be quickly and neatly resolved with a directive from the insurance commissioner clarifying that the measure is understood to be limited to personal lines insurance.

Sizemore is a yutz. Why should we want to vote for his measure? Not being a scholar of the Oregon political scene, I’m not qualified to opine on the alleged yutziness of Sizemore.

From this side of the continent, he strikes me as having some characteristics in common with other vocal consumer advocates — it’s not easy to tell whether he’s making noise because of a desire to help consumers, or to serve his own goals. I admire folks who seek to spotlight and correct problems, but I am annoyed by hypocrites and society’s rules do not permit me to wield a clue-by-four as often as I am tempted to.

Exactly where Sizemore falls into that spectrum of traits, I cannot say.

I must say, however, that I am a little disappointed that the opponents of Measure 42 are stooping to attempting to turn the initiative into a test of Sizemore’s (un)popularity. Admittedly, it seems that’s the level that American politics has degenerated to…but it doesn’t mean that I have to like it.

There are reasons that Measure 42 is not the best idea in the world. Sadly, those reasons don’t translate well into easy soundbites.

Insurers say that the world will end if credit scoring is banned. OK, I’m exaggerating here; after having written this much, who can blame me for being a little loopy? Besides, from some of the media snippets I have seen, the more vocal opponents of Measure 42 have seemed a little hysterical.

If Measure 42 passes, there will be market disruption. People’s rates will change seemingly randomly, and complaints will be made. Insurers will have to adjust their rating algorithms, and the marketplace will be reshaped into something different. My colleagues who work on Oregon insurance products will probably have a tough few months due to all those changes, if Measure 42 comes to pass.

However, it won’t be the end of the world. The market will be different. Some people will pay more, others will pay less. And insurers will find a way to make a profit and recoup the expense of the transition. Life will go on.

In conclusion, I’d like to thank you for being almost as masochistic in reading this as I was in writing it.

I have no vested interest in how Measure 42 turns out, but I do know a bit about credit scoring in insurance.

Personally, I think Measure 42 is not a good idea, and I would be voting “no” if I lived in Oregon. Scoring is adequately regulated today, and I think it’s more of a benefit than a detriment to consumers.

However, I can also respect that Oregonians may wish to provide direction to their leaders on this matter. If consumers in the state understand the market disruption that could arise from the measure passing, but want to ban scoring anyway on principle…more power to them.

If you’re an Oregon voter and you want to do me a favor, there is one thing I would like: Please base your vote on facts, and not on soundbites and campaign propaganda.

If you have questions or comments, please feel free to respond to this post, or to email or IM me.