Refi backlog spurring debate on quality-control sampling.

When the boom in home-loan refinancings started two years ago, lenders began to worry about how to produce high-quality mortgages amid heavy volume.

Today, the refinancing boom is over. But it has left in its wake a backlog of work that is still straining quality-control departments, making quality control a hotter issue than ever.

So far, the debate has focused on an obscure but crucial aspect of quality control: sampling techniques.

Droves of lenders, seeking to cut costs, are switching from the Federal National Mortgage Association and Federal Home Loan Mortgage Corp.'s suggested method -- simply checking 10% of all loans originated. Instead, they are choosing another agency option: statistical sampling that examines a smaller percentage of all loans but considers characteristics of the loans in the sampling technique.

While such sampling may well be less costly, it raises questions about how effective the procedure is in heading off longer-term problems, such as foreclosures and fraud losses.

Lenders perform quality control at several points during a new loan's processing. Even the first glance over an application can be called quality control.

But it is after the loan closes that lenders sample percentages of loans to check whether the lumps of mortgages are of "high quality." Lenders check for fraudulent reporting, inaccurate documentation, and creditworthiness, among other things.

10% Sample Assailed

Critics say the 10% sample that Fannie Mae and Freddie Mac permit doesn't find what lenders -- and investors -- are looking for.

"The loan sample of 10% is not based on risk; it is based on sheer numbers," said Edward E. Furash, president, Furash & Co., a Washington consultancy. "It's a fairly unsophisticated system."

A 10% sampling gives a picture of the quality of loans in a portfolio. But an accurate picture? Experts say no, because it is a rough sample and not truly random.

"I would not count on [10% sampling]," said Hakki Etem, principle of Cogent Real Estate Economics Inc., San Francisco, and an industry guru on sampling.

Mr. Etem estimates that three-fourths of all lenders still use the 10% sampling.

A portfolio of 1,000 loans may have as many as 300 originators, including mortgage brokers or correspondent lenders, he said. Each originator has different risk considerations.

A 10% sample produces a statistical picture accurate to plus or minus 1%, whereas the smaller random sample will yield plus or minus 2% accuracy. But Mr. Etem noted that a 10% sampling could cost five times more than a random sample, which checks about 3% of a portfolio.

"What have you really gained?" he asked.

Robert J. Engelstad, senior vice president for mortgage and lender standards at Fannie Mae, said the agency leaves it up to the lender to choose a method of sampling. Lenders that produce more than 7,000 loans a year can use statistical sampling. Fannie provides sampling guidance upon request.

Time for Change Seen

But that's not enough, said Mr. Furash. It's time the system was changed, he said. A more consolidated industry means that fewer mortgage banks are making more of the loans.

North American Mortgage Co., Santa Rosa, Calif., and Directors Mortgage Loan Corp., Riverside, Calif., are among the major lenders that have adopted random statistical sampling.

Mr. Engelstad said on-site checks by Fannie officials -- usually on a yearly basis -- find faulty sampling.

But Mr. Etem said random sampling also has its faults. He said many lenders who have switched to it do not use a true random sample. Some just pick every 10th loan for a check, he said. That's not random.

"If it's not truly random, you are up the creek," he said.

For reprint and licensing requests for this article, click here.
MORE FROM AMERICAN BANKER