Credit Scores

One of the things I enjoyed about working with credit scoring in a prior job was the utter beauty of the interrelationships of the different variables coming together to make a prediction.  When you consider that beyond delinquency information, scoring relies on utilization, stability and breadth of credit, questions like “what happens to my score if I apply for a new credit card” become beautifully complex.

Similarly, one of the lessons I learned in that job was “watch out for the impact of economic changes on your credit model”.  While that’s a subject that has been reasonably well-hashed-out for banking and lending purposes, both domestically and internationally, American usage of credit data for insurance purposes is still comparatively new.  The current crunch is arguably the first test of the interrelationship between scoring and predictiveness.

So, it is with some interest that I notice a wire service story came out this weekend on how credit issues are starting to be reflected in (banking) credit scores:

At the same time, revolving credit usage — which includes credit cards — accelerated sharply to a year-over-year growth rate of about 8 percent in recent months. That’s the fastest rate in seven years and well ahead of the 2 to 3 percent rate of growth from 2004 through 2006 when home equity lines of credit were a bigger source of cash for consumers, according to Merrill.

But as credit cards are used more frequently, that often results in bigger balances left on the cards. What’s worrisome is that consumers who are faced with a number of ugly economic scenarios hitting at once — falling home prices, surging commodities costs and a weak job outlook — won’t be able to pay their bills.[…]

[C]ard companies including Washington Mutual, HSBC and Wells Fargo are lowering their credit limits, according to data from the consulting firm Institutional Risk Analytics.  Consumer advocates aren’t saying that is bad news — in fact, they believe it helps prevent cardholders from overextending themselves and is preferred to having a sudden surge in card interest rates. […]

Let’s say a cardholder has a credit limit of $10,000 and a balance on the card of $4,000. The card company worries that large balance may increase the prospects for default, so it lowers the credit line to $5,000. But in doing that, it completely changes what is known as the credit utilization rate, raising it from 40 percent to 80 percent. That is then factored into the calculation of one’s so-called FICO credit score, which measures creditworthiness, according to Craig Watts, a spokesman for FICO-creator Fair Isaac Corp.

When I did my stint in insurance credit scoring, part of the gospel I preached was that the actual numeric score coming out of the model was meaningless without context.

The main reason I said that was because of the abundance of scoring models in use among insurers.  Fico had at the time 18 different insurance scores at each of the three bureaus, Choicepoint had its own fleet of models which, while designed for portability, could still vary somewhat depending on which bureau’s data was pulled, and many if not most large and mid-sized insurers were dabbling with their own proprietary models.

And that was just within the insurance industry.  Over in the banking world, where every bank of note seemed to have its own flock of statisticians building proprietary models for various purposes, even as Fico and the individual bureaus were marketing their own industry solutions….   Telling me that you have a 750 score doesn’t exactly tell me a heckuva lot unless you can also tell me something about the model, the distribution of scores, the potential use, et cetera.

But people believe in the illusion of concrete fact embodied by a magical three digit number, and my message tended to fall on deaf ears.

I feel a little vindicated, however, by a story that ran in the Wall Street Journal (subscriber link) last week:

These days, hardly anyone questions the power of the almighty credit score. Lenders use it to determine who qualifies for a loan and what interest rate they get, insurers calculate premiums based on it, and employers use it to help make hiring decisions. As a result, sales of scores, reports and credit-monitoring services to consumers by the three major credit bureaus — Experian Group Ltd., TransUnion LLC and Equifax Inc. — generated $488 million in revenues in 2006 and are expected to reach $864 million by 2010, according to research firm TowerGroup.

Yet the scores that consumers buy from the credit bureaus or heavily promoted Web sites like FreeCreditReport.com or TrueCredit.com — owned by Experian and TransUnion, respectively — aren’t the same scores that are sold to lenders, landlords, insurers or employers.

Take FICO, the credit score developed by Minneapolis-based Fair Isaac Corp. that the majority of lenders use. Depending on the type of credit a consumer seeks — a mortgage, installment loan, auto loan or credit card, for example — lenders will use different “flavors” to determine a consumer’s default risk. Auto-loan payments, for example, weigh more heavily in the formula that calculates a FICO score for auto lenders, while credit-card payments matter more in the score used by credit-card companies.

Adding to the confusion: Each time Fair Isaac rolls out a new version of its scores, some lenders implement them while others stick to the old ones.

I can imagine the consumer advocates now, howling “why can’t we have The Score”.

The answer, as much as no one likes it, is that there is no One True Score.  Users of credit data, be they banks, insurance companies, utility companies, cell phone carriers, etc. will tend to favor whichever model is most predictive for their purposes…or build a new model if they think they can get additional lift.

I can see where someone would argue that it sucks that consumers won’t always know The Answer beforehand, due to the abundance of scores in the world.  On the other hand, because of the difference in opinions among models, there’s always the room to shop around for a better answer, particularly if you have a so-so credit history.

Besides, the advice about what to do to optimize one’s score under any credit model is generally the same — if you pay your bills on time, have a reasonable number of credit accounts which you use sparingly, and if you don’t churn your credit needlessly…you’ll do OK.

Folks who have followed my posts here are aware that I am a fan of the use of credit scoring in insurance pricing/underwriting. In fact, one of my past jobs focused on the development and implementation of a scoring tool, and in the government- and public-relations exercise related to the implementation.

While credit scoring has been a landmark in the development of predictive analytics in insurance, and it is an incredibly powerful tool…it is just one piece of a larger puzzle, which users ignore at their peril.

While my opinions are based primarily on my insurance background, the same statements hold true for other financial services businesses. For example, consider this column, written by Fair Isaac’s CEO, Dr. Mark Greene, which appeared recently in Forbes:

The reaction in the financial services industry has been to slam the credit window shut. But that’s not a solution. The real question is how lenders can evaluate the individual loans in their portfolios in order to keep the credit crisis from turning into a credit collapse.

The first part of the answer is restoring the prudent application of the three Cs [credit score, capacity to pay, and collateral]. We’ve seen the results of moving away from these “old-fashioned” standards. It’s time to recognize they are necessary modern lending standards as well.

The second part of the answer is high-tech ingenuity, including reliance upon proven credit risk controls like credit scoring. The smart use of credit scoring remains essential, and will in fact benefit lenders who need to wisely balance risk and growth when operating in new or uncharted economic territory.

Looking at this problem, the Consumer Federation of America has proposed, among other things, a test of suitability before giving a consumer a loan. This includes, along with a prospective borrower’s FICO score, looking at their ability to handle added debt payments or interest rate increases. For that good idea to be widely adopted, it will also have to be automated. Such a tool is well within the ability of the industry to deliver.

Of course, I’ve got to believe that Dr. Greene will want shortly to announce such a tool…but that’s what he and his company are in the business of doing.
Just because we have a fancy black box spitting out numbers to which we assign meaning (sometimes too much meaning), it doesn’t mean that every other tool and our common sense must be thrown out the window.

Seen in the Wichita Eagle:

Credit bureaus are expected to adopt a new version of the widely used FICO credit score this year that will no longer benefit so-called “authorized users” on another person’s credit card account.[...]

The move is in response to the controversial practice of “piggybacking,” in which some Web sites allow consumers with poor credit scores to hitch on to someone else’s good credit record.[...]

Called FICO 08, the new formula also will be more forgiving of occasional slips by consumers but will take a harder line on those who are repeatedly late on their bills. The FICO score ranges from 300 to 850.[...]

TransUnion expects to have the new FICO scoring model “available for customer testing during the second quarter of 2008,” said spokesman Steven Katz.

Experian doesn’t have a date for adoption of FICO 08, said Rod Griffin, manager of consumer education. “We’re working with FICO on the technical issues and on implementation, but we don’t yet know what the date will be,” he said.

Equifax is apparently standing aside for now, due to an ongoing squabble with Fair Isaac.

Fico has been pushing for the credit industry to more modern models for a while now, since a bit before my prior life where I worked on credit scoring for an insurer. It’s a little amusing that it’s taken a scam to push through change at two of the three bureaus.

Of course, as the new models roll into production, I expect to see some consumers and consumer advocates become introduced to the notion of just how different a consumer’s scores can be under two different models, which should make for some “entertaining” debate. That, plus I half expect to see some additional fuel supplied to the fires of those who don’t understand and don’t like the nuances of differences in what data is collected by the individual bureaus.

(Note to the type A personality folks watching their scores—the actual 3 digit numbers aren’t necessarily significant. A 20-50 point swing isn’t terribly significant, as long as you’re still passing the hurdle rates. Besides non-mortgage-based lenders mostly use their own, proprietary models anyway. Fico scores these days are only really important in the mortgage world, where a standardized model is necessary to securitize the mortgages as well as to prescribe standards for Fannie Mae and Freddie Mac.)