The criticism is being leveled at computerized real-estate appraisals, which depend on models that use prices from home sales and other data to determine the value of a house. Because of the volatility in the housing market, they are underestimating prices, some homeowners, real-estate agents and fee appraisers say.
Lenders use computerized appraisals primarily for home-equity loans, preapprovals for mortgage refinancing, loan modifications and mortgage originations of less than $250,000. Automated appraisals are cheaper and faster than in-person appraisals. They run as little as $20, whereas appraisals done by people can cost hundreds of dollars.
The computerized models are used as a check on in-person appraisals, which often were too generous during the housing boom, according to federal banking regulators and state attorneys general. The regulators said banks often held sway over appraisers, encouraging them to value homes at certain prices in exchange for future business. In the wake of the housing bust, regulators imposed tough new rules, prohibiting banks from picking individual appraisers for individual properties.
"The selling point was that [computerized appraisals] were faster and not prone to bank pressure," says Steven Kane, a Colorado commercial and residential appraiser who is the author of two books on how to apply automated valuation models.
Computerized appraisals calculate a home's value by using an index derived from historical repeat-sales data, or sales records of homes with similar property characteristics, such as square footage and the number of bedrooms and baths. In-person appraisals don't incorporate as much transactional data as a computer model.
Gary Cohen, an advertising-sales manager in West Los Angeles, Calif., says Citibank suspended his $510,000 home-equity line of credit based on a drop in his home's estimated value.
A computer model used by the bank showed his home had dropped to just over $1 million in 2009 from the $1.65 million it was appraised at four years earlier.
So, Mr. Cohen, 65 years old, paid $750 for an in-person appraisal from a firm designated by the bank. It estimated his home was valued at $1.3 million, but Citibank still wouldn't reinstate his credit line.
"The discrepancy is so great that you have to know whatever method they are using is not accurate," Mr. Cohen says.
Mr. Cohen sued Citibank, a unit of Citigroup Inc., over the appraisal. In court documents, Citibank said that even if his home is worth the higher figure, the bank has a legal right to suspend the credit line.
Citibank continues to believe the suit has no merit and intends to defend its position vigorously," said a spokesman.
Borrowers also have sued J.P. Morgan Chase & Co., Wells Fargo & Co. and other big lenders, claiming that banks are misusing automated valuation models in order to cut home-equity lines of credit. J.P. Morgan Chase and Wells Fargo declined to comment.
Automated valuation models were pioneered by Yale economist Robert Shiller, who developed the first systems in the early 1990s. While arguing that these appraisals are more objective than human appraisers, Mr. Shiller and others say that in some situations the models may be providing unrealistically low values, prompting lenders to reject loan applications or lend less money on particular properties.
Some models weigh past sales of a particular property over time against a historical home-price index, and they are running into problems with properties that have been bought only once. That is the situation in places such as Nevada and Southern California, where new subdivisions sprouted during the housing boom but many homes never sold or entered foreclosure before ever being sold in a nondistressed transaction.
"The main difficulty is that I need two or more sales prices for a property, and if I'm not able to find it, it doesn't fit into the sample used to calculate the index," says David Stiff, chief economist at Fiserv, one of the largest providers of automated appraisals using this methodology.
Prof. Shiller concedes there can be problems with these appraisals if a too-short period of historical data is programmed into models.
"In a slow market, it might suggest that prices are going to be falling for a while," he says.
Other computerized models break down the particular characteristics of a property—number of bedrooms and bathrooms—as well as sales of comparable homes, to arrive at a value estimate. They often are hampered by a lack of accurate or comprehensive data in county and municipal records. Improvements, for example, are recorded by building permits, so if homeowners don't file permits, the records won't be accurate.
These models can "change a lot, depending on which variables you include or exclude, so there can be a bias," says Prof. Shiller.