Automated valuation models are gaining credibility in the residential mortgage-backed securities market. And one ratings agency has some advice about how to manage the risks associated with using values from less than full appraisals.
Fitch Ratings, which previously discounted AVM, drive-by appraisal or broker price opinion values by 10 to 15 percent in soft or weak markets, announced it will now limit the reductions to 5 percent or more. The discounts apply to lenders with usage processes and controls that do not adequately mitigate over valuation risk and lenders that do not disclose their AVM policies to the ratings agency.
The updates were outlined in the New Treatment of AVMs in U.S. RMBS report which affects pools with newly originated first liens.
Fitch said its research determined that data supporting AVMs do not significantly lag market conditions.
"Data providers that cull property information from public sources have become much faster at collecting and entering the data," Fitch said in the report. "Also, because home prices do not fall precipitously within very short time intervals, comparable sales data a few weeks or months old are unlikely to generate overinflated values in a soft market.
"The risk of over valuation is low if a lender has a solid understanding of local conditions and adheres to established benchmarks and risk parameters for its AVM or nonfull appraisal usage."
Fitch recognized AVM limitations, including the inability to observe the property's condition, the inability to access data for properties located in areas where available data are limited or unavailable, and the possibility that lenders can tap into multiple AVMs to "shop" for a value.
But "the benefits of AVMs outweigh the risks," the report said.
In evaluating a lender's AVM usage procedures, Fitch said it will consider the lender's AVM selection process, business rules and testing procedures, as well as policies for addressing risk layering and program restrictions to ensure that AVM usage is selectively applied.
Lenders should use a reasonable sample size of its portfolio for determining an acceptable percentage by which the AVM results can exceed the sales price or value. Applying unjustifiably high thresholds for selecting an AVM would concern Fitch, the report said.
Also because using several models to obtain a value is necessary but can be an indication the lender is shopping for the highest value, lenders should stick to a cascade of three, as this "is reasonable and appears to be the standard," the report suggests.
Preferably, the sample of loans used for testing should be representative of the lender's loan production pattern, volume, and product type for which the AVM will be used. Ideally, the sample would also include purchase money mortgages since the sales price is the best indicator of value.
Lenders should also include in their testing recent or pending sales transactions that are not yet in the public record. This, because values that are a few months old are likely to be in the public data records AVMs draw on to obtain a value, and the accuracy of the results is skewed by the inclusion of its sales price as a comparable.