Finally Catching On: Automated Valuation Models (AVMs) Have Been a Technology Looking for an Application in the Mortgage Business for Years. Only Recently Have Lenders and Secondary Market Investors Found Ways to Make Them Pay

Article excerpt

The debate about automated valuation model (AVM) technology continues to rile the mortgage banking industry, with the key talking points having to do with prime first mortgages and how much of the appraisal business would be or could be supplanted by AVMs. [??] It's an important but ironic debate, because beyond prime first mortgages, AVM technology has quietly become solidly implanted in the secondary market and is steadily gaining a toehold in the subprime mortgage business. All this has happened relatively recently. [??] In the last several years, AVMs have greatly penetrated the secondary market, notes Darius Bozorgi, president and chief executive officer of Santa Ana, California-based Veros Software Inc. The reason: improved technology in second-generation AVMs and related analytic products, and, finally, a track record of usage.

[ILLUSTRATION OMITTED]

"Second-generation AVM tools are much more accurate, provide a greater number of values with a much higher level of accuracy, and can value a wider array of properties such as townhouses and condominiums across greater price tiers and geography," Bozorgi says. "They have become much more robust and efficient. That leads to greater productivity for the mortgage industry."

AVMs have been used in the secondary and subprime markets for about five years, but the heaviest use has come in the past two years, concurs Paul Bohan, chairman and chief executive officer of The Bohan Group, San Francisco, a risk-analysis service company. "People are trying to get comfortable with the product. Plus, there are now a variety of AVM producers out there," Bohan says.

While AVM technology has been around for about three decades, the key to making it work on a national basis has been the availability of data, which are usually ascertained by local authorities where home sales are recorded. For the most part, that information has finally been automated so AVM technology companies can create large, up-to-date databases whether the property backing the loan is in Florida or Oregon.

"No matter where the loan is, an AVM can get you a value right away, and that data didn't exist in an aggregated form until recently," says Michael Sklarz, senior vice president and head of analytics for Fidelity National Financial, Jacksonville, Florida.

"The information technology allows the statistical models to be estimated almost in real time. If you put a request in today, the AVM draws upon the comps that are available today; if you put in a request a month ago, it would have drawn on data available at that time," says Sklarz.

Secondly, for those lenders that have been testing AVMs and using the tools in a production environment, a comfort level has been established along with a strong track record that extends for several years. "It is really an issue of understanding risk with a proven track record or history, and ultimately the performance of the loans backed by these products," says Bozorgi.

The secondary and subprime businesses have adopted AVMs to do similar tasks, but because of the mechanics of the technology that adoption has occurred for different reasons.

On a general level, the secondary market and subprime lenders incorporate AVMs as a tool to review (portfolios) for quality control and for independent validation of the (appraised) values of the properties. In the secondary market AVMs aid in the decision-making process to accept or reject the purchase of mortgages, either individually or in blocks. AVMs have been adopted by secondary-market users because they automate the process, allowing rapid review of thousands of mortgages at relatively modest cost.

Batch processing is one of the strengths of AVMs, because in batch tests submitted to Colchester, Vermont-based AVMax, the values returned closely follow a statistically "normal" curve with a mean error very close to zero. This means the total predicted value of the properties generated by AVMax is a highly accurate representation of the total actual value. …