Automated valuation models (AVMs) are increasingly being used on the secondary market, but some of their uses are causing anxiety.
  Industry observers are particularly concerned about cascading AVMs, platforms that combine different products. If the first model cannot deliver a valuation for a property, or a ‘hit,’ the platform automatically ‘cascades’ to a second AVM. If the second model doesn't deliver a hit, the platform turns to a third.
  In addition, some cascading platforms allow the user to pick one of several values, essentially allowing them to shop values, says Craig Leonard, vice president, portfolio and securitization solutions, for Veros Real Estate Solutions, an AVM vendor based in Santa Ana, Calif. If appraisals were more affordable, some originators would order three and use the highest value. AVMs present no cost obstacle, he notes.
  ‘It creates a very risky situation,’ Leonard says. ‘There are inherent risks in these platforms. If the first three can't find a value, the fourth could be a problem.’
  When examining cascading platforms, investors should find out who determines the value and why they pick particular values, advises Leonard, who gave a presentation at the Predictive Methods Conference sponsored by Veros and First American Real Estate Solutions last week in Newport Beach, Calif.
  Lenders that choose figures from cascading models are essentially value shopping, agrees Jacqueline (Jacquie) Doty, collateral policy director for Freddie Mac, who also spoke at the conference. ‘It could create a risky situation. That's exactly what we don't want. We want lenders to focus on accuracy above hit rate.’
  If values are inflated, loan to values are understated and loans could lack adequate private mortgage insurance, she points out. Loan performance can be worse that expected, and severity could be worse than forecasted because of a lack of mortgage insurance.
   While most give one value, some cascading platforms provide a range, with the high end from one AVM and the low end from another, says Leslie Albergo, a director for Standard & Poor's Rating Service. ‘My question for the originator would be, 'What number are they taking when they calculate the borrower's LTV?'’
  S&P should review the models whether or not they're used in cascading platforms, Albergo stresses. S&P analyzes how AVM vendors arrive at values and tests results against sale prices, finding variances by state metro areas and sale price groups. Performance results are input into the rating agency's loan-level credit enhancement model used to size and structure mortgage-backed securities.
Proper training needed
  Lenders and investors should beware of other possible AVM shortcomings, warns Freddie Mac's Doty. For instance, many lenders have the valuations ordered by AVM operators, rather than appraisers. But without proper training, those operators may not understand property valuations, how the models work, and how different property characteristics can impact values depending on property locations. Mortgage companies should also be aware of the lack of property inspections, she adds. For example, many homes in some areas were damaged by hurricanes.
  Appraisers should be trained in AVM-based valuations, and understand the model, how often it is updated, and if it's appropriate for that particular assignment. If they don't understand the model, they may only be rubberstamping its results.
  A lack of an industry standard for testing and measuring the models' performance is a hindrance, she adds. Many vendors want to keep their models proprietary, but without a standard, it's hard for Freddie Mac to assess risks of the automated valuations, other than its own AVM, HomeValue Explorer. Embedded in Freddie Mac's LoanProspector automated underwriting engine, the AVM can be used to bypass traditional appraisals, typically in purchase money mortgages with 80% or less LTV and good-quality borrowers.
Are you confident?
  Leonard urges AVM customers to ask about the vendor's confidence score, a measure of valuation accuracy. For instance, a 10% drop in the Veros confidence score equals a 5% variance in the score. Fewer hits with good confidence scores are better than more hits with low accuracy.
  ‘We strongly recommend large-scale testing and ongoing testing,’ Leonard stresses. ‘Understand what the confidence score is and how they're going to use the confidence score.’
  While only about 5% to 10% of securitized deals that S&P rates have AVMs and full appraisals still dominate the industry, automated valuation use has rapidly increased, especially for seconds and lines of credit.
  ‘On the secondary market, there's tremendous use of AVMs,’ Leonard says, noting that the valuations are frequently used as a quality control tool. ‘I don't know of anybody on Wall Street not using AVMs as a quality control tool.’
  Use of secured valuations, which involves an AVM with an insured value in lieu of an appraisal, is booming, especially for seconds and equity lines, according to Leonard. If the loan defaults, an appraisal would be completed to check the automated valuation. If it was wrong by 5% or more, the vendor will cover the loan holder's net loss.
  ‘No one is selling an insurance policy,’ he explains, saying it differs from an insured valuation. ‘The difference is we are warranting that the value is right. We're not selling an insurance policy. There's a tremendous amount of interest in it on Wall Street.’