Media coverage of the denial of a Toys 'R' Us credit card to Federal Reserve Board Governor Lawrence Lindsey indicates that there are people in the industry who still do not understand either credit scoring or the basic nature of credit decisions.
A lender making a consumer credit decision is faced with predicting human behavior: Will this borrower repay this loan in a satisfactory manner, or not?
Many of the factors affecting credit performance are outside the control of the borrower: a recession, layoff, or serious illness can easily result in the best-intentioned borrower's becoming seriously delinquent. "Character" certainly plays a large part: many struggling blue-collar workers never miss a payment, while some wealthy individuals leave a trail of bad debts in their wake. But how do you measure "character" in advance of actual performance?
The simple fact is that there is no way to predict the credit performance of any individual - even Governor Lindsey - with absolute certainty. The crystal ball for creditors just doesn't exist. Any loan officer who believes it necessary to accurately predict individual behavior is doomed to frustration. By that standard, even one delinquency in a million-account portfolio represents failure.
What can be measured and predicted with reasonable accuracy is the level of risk, especially relative risk, presented by different groups of borrowers. Once lenders accept the idea of risk-based decisions, their lives become much less frustrating (although, perhaps, more complex).
For with the notion of risk comes resignation to the fact that some borrowers who are accepted will go delinquent, and that some of those rejected would have been great customers.
Being a successful credit risk manager doesn't mean getting every individual decision right; it means getting the right mix. The bad news is you can't accurately predict individual credit behavior; the good news is you don't have to. But you do have to do a reasonably good job of predicting the behavior of groups of applicants or existing account holders. That's where credit scoring comes in.
In developing scorecards, all of the information known about a borrower at the time a credit decision is made is carefully analyzed to determine not only which individual pieces of data are predictive of future performance, but which combinations and weightings result in the best overall predictor of credit performance.
The scores produced by applying the resulting scorecard to a subsequent applicant's credit bureau report or credit application permit the lender to determine the relative risk of different groups of borrowers.
For example, someone scoring 200 would be less risky than someone scoring 180 but more risky than someone scoring 220. Scores do not purport to say that every applicant scoring above a particular number will prove to be a satisfactory customer or that every applicant scoring below that number will become seriously delinquent. Nor does scoring determine what degree of risk is appropriate for a particular lender. That decision is up to the individual credit grantor.
What scoring developers claim - and what four decades of experience demonstrates - is that scoring can significantly outperform judgmental (subjective) decisions. How much is significant? In a typical consumer credit portfolio, switching from subjective decision making to scoring can be expected to reduce delinquencies by 20% to 30% if the acceptance rate is kept constant. Or it will allow a 20% to 30% increase in the acceptance rate while maintaining the same delinquency rate.
Why is this so? In the typical application scorecard development, 50 or 60 factors may be identified as having some stand-alone predictive value, and probably 10 or 12 variables will find their way into the final scorecard.
Even if each variable has an average of only three possible values, the potential combinations still run into the tens of thousands. Not many human minds can make a decision that accurately assesses that many factors.
Will scoring and judgment always produce the same decisions on the same applicant? Certainly not. If individual decisions were always the same, scoring couldn't produce the improvements just cited. Is each and every decision based on credit scoring "better" than a different judgmental decision? Of course not.
Returning to Governor Lindsey, would he have paid a Toys 'R' Us card satisfactorily? Probably. But that's not the question Bank of New York had to answer. For a consumer lender, the question must be, "If I extend credit to 100, or 1,000, or 10,000 applicants like this, will 99% or 95%, or only 90%, repay in a satisfactory manner?"
Even at 90%, the answer for any particular individual is that he or she would "probably" repay as agreed, but a 10% default rate would be staggering for most portfolios.
The scoring system used by Bank of New York in making its decision included as one of many factors reviewed "number of credit bureau inquiries" for the simple reason that this factor is demonstrably predictive of credit performance.
Does everyone with seven or eight inquiries go delinquent? Hardly, but Fair, Isaac's research shows that, as a group, applicants with that many inquiries are three times riskier than the average applicant, and six times riskier than those with no inquiries.
That is a powerful piece of information that no responsible lender ought to ignore. An advantage of statistically based scoring, however, is that no one factor is responsible for a ranking.
Contrary to the premise of the article by Douglas Austin which appeared in the Jan. 24, 1996, issue of American Banker, lenders, credit bureaus, and scoring developers are well aware of the difference between "voluntary" inquiries - those triggered by a consumer's request for credit - and "involuntary" inquiries - those generated by promotional mailings or the routine review of existing accounts. No scoring system developed by Fair, Isaac considers such "involuntary" inquiries in assessing credit risk.
In fact, the on-line credit bureau reports typically used in screening applications for new accounts don't display such inquiries, so they couldn't be considered even if the lender or scorecard developer so desired.
There are a couple of ironies to this story that should not be overlooked. If there is a topic on which Governor Lindsey has been more outspoken than the dangers of overreliance on statistics, it is the danger of consumers becoming overextended due to increasing credit card debt. Well, finding the folks who are likely to be getting overextended is precisely why scoring systems use factors such as the number of recent inquiries.
Moreover, Governor Lindsey and the Toys 'R' Us card is a story only because so many people are willing to embrace the tale's unspoken premise: A well-paid white male with no reported credit delinquencies must be the best possible credit risk. Judgmentally, that seems like a reasonable conclusion.
If a minority single mother, who makes a quarter of Governor Lindsey's salary and has a couple of minor delinquencies in her credit history, had been turned down, no one would consider that newsworthy. But the exhaustive data analysis that goes into developing a credit-scoring system just might show the latter applicant to be a better credit risk, even though she wouldn't have had a prayer in a judgmental decision process.
I hope that lesson is not lost on those charged with ensuring that credit is available to minorities, women, and low-income and moderate- income applicants.
Mr. McCorkell is senior vice president and general counsel at Fair, Isaac, which is based in San Rafael, Calif.
