The Cracks in Credit Scoring

As loan defaults rise, worries grow about how creditors pick borrowers

Ever since the tech bust in 2000, consumer spending has been keeping the economy afloat. And consumer lending, rather than pay raises or job growth, has been fueling that spending. But now, the techniques that allow profit-hungry lenders to make bigger loans faster, are coming under intense scrutiny. First among them the credit scores that banks, credit-card issuers, and mortgage companies plug into complex mathematical models to figure out how likely borrowers are to repay their loans.

Under the stress of weak growth and rising unemployment, some models are breaking down. Consider Metris Cos. (MXT ), the 10th-largest credit-card issuer in the U.S. On Oct. 28, rating agency Fitch downgraded $4.2 billion in securities backed by Metris credit-card loans, because charge-offs had increased 30% in the past 12 months, far above expectations. Metris is just one of a half-dozen credit-card companies that announced big jumps in bad loans this year. "These issuers put a lot of trust in their models, and at the end of the day no one really knew how this type of portfolio would perform in a stressed environment," says Fitch analyst Rui Pereira. "Depending on whether or not there is another shoe to fall economically, things could get worse."

Credit-scoring models have become an integral part of the financial system and have been used extensively since the 1990s. About 70% of the home loans issued since then and nearly all of the $1.7 trillion in credit-card, auto, and personal loans outstanding were made using a customer's credit score to determine how much to lend and at what interest rate. Credit scoring is also an important tool for investors who buy pools of these loans, because it allows them to evaluate hundreds of loans in minutes.

So far, the problems have been concentrated in credit-card loans to customers with poor credit histories, so-called subprime borrowers. The models underestimated the impact of an economic slowdown on them. "Subprime borrowers are affected disproportionately by economic factors," says Al Brothers, executive vice-president of distressed-debt specialist Cavalry Investments LLC. "They tend to have the highest debt-to-income ratios, and they tend to be the last to get hired and the first to get fired."

Economists fear the problem could spread to $450 billion of subprime mortgages. As much as 10% of them could go bad, says Douglas G. Duncan, chief economist for the Mortgage Bankers Assn. As a result, investors who buy securities backed by consumer loans are shying away. "There's been a real overreliance on statistics when you're dealing with loans to people," says Gary E. Wendlandt, chairman and CEO of New York Life Investment Management LLC.

Regulators are getting worried about credit scoring, especially the way it has been used to pump up lending. The amount of subprime loans outstanding has doubled to $500 billion in the past three years as some lenders chased high fees and interest rates that can yield profits five times those of conventional loans. "Even before we got into an economic slowdown, these lenders were pushing rapid growth through ambitious business models," says George French, a deputy director at the Federal Deposit Insurance Corp. To get a clearer picture, bank regulators may require consumer lenders to break out subprime loans by next March.

Questions are growing about the numbers backing the scores. In an Oct. 7 speech, Federal Reserve Chairman Alan Greenspan voiced his concerns that problems may arise from an "insufficiently long data series." Most credit-scoring models rely on two years of data. That meant the models produced rosy forecasts in the long economic boom of the 1990s. But as debt mounted, and the recession hit, losses exceeded levels suggested by the models. "When you place too much hope on past experiences, you're setting yourself up for trouble," says the FDIC's French. Adds New York Life's Wendlandt: "We didn't even have credit-card lending 30 years ago."

Consumers who increasingly game the system further undermine the reliability of the scores. With more riding on credit scores than before, a cottage industry has sprung up dedicated to helping would-be borrowers improve their ratings to get lower interest rates.

There are few checks on the accuracy of the data. Three credit bureaus, Experian Information Solutions, Equifax (EFX ), and Trans Union, each keep accounts on 190 million customers and collect about 16 billion pieces of information annually in the U.S. They then sell it to lenders or third parties, who use it to create a credit score. But the information may be incomplete because lenders aren't required to report consumer data to the bureaus. Furthermore, the bureaus don't check the information unless consumers complain. The industry hasn't monitored what percentage of data overall may be flawed, says Norm Magnuson, vice-president at the Consumer Data Industry Assn., a trade group. But a recent study found that fewer than 1% of customers were denied credit as a result of errors, he says.

The biggest problem, according to some critics, is that the credit models are racially biased and unfairly drive minority borrowers into high-rate loans. A suit filed in U.S. District Court for the District of Columbia on Sept. 13 accuses Fannie Mae (FNM ), the government-sponsored housing agency, of discriminatory lending practices. Safiyyah Rahmaan, an African American, allegedly was denied a 6.5% mortgage loan after Fannie Mae's automated underwriting process said her credit score was too low. The suit says Fannie's scoring process is biased and cites a Freddie Mac (FRE ) study that found that while 27% of Caucasians had "bad" credit records, 48% of African Americans, and 34% of Hispanics did. The Housing & Urban Development Dept., which regulates fair lending, is reviewing Freddie's and Fannie's underwriting practices, a spokeswoman said, but results are not yet available. A Fannie Mae spokeswoman says that the company plans to formally respond to the suit on Nov. 18, adding that it is in full compliance of all fair lending laws, and that its programs have expanded minority homeownership.

Until recently, the scores drew the most criticism for being opaque. But under pressure from consumer advocates, Fair, Isaac & Co. of San Rafael, Calif., which produces the most widely used scores, called FICO for short, released some details about its secret formula in 2000. About 35% of a FICO is based on a borrower's history of paying back debt; 30% on how much of the credit available a borrower has used; 15% on the length of the borrower's credit history; and 10% each on type of credit and pattern of credit use. FICO scores generally fall between 550 and 800, but nearly 20% of the U.S. population has a credit score under 620, generally the cutoff for a prime-rate loan.

Founded by engineer William Fair and mathematician Earl Isaac in 1954 to provide credit advice to lenders, the company stoutly defends its role as a key player in the industry. Lenders should be using many other factors to make decisions, says Michael W. Rapaport, vice-president at Fair Isaac. For instance, they should be monitoring their portfolios regularly and readjusting rates and loan size as a borrower's situation changes. "If you're going downmarket, it's even more important to track and monitor your portfolio closely," he says.

Loans have been made to customers with poor credit histories or low incomes for decades. Hardscrabble finance companies such as United Cos. and the Money Store did so successfully by knowing their customers well. All have either been bought out or closed in the past five years as competition increased from banks that can borrow money more cheaply. At least the old-timers weren't tripped up by fuzzy math.

By Heather Timmons in New York

    Before it's here, it's on the Bloomberg Terminal.