ATTOM Data Solutions has introduced a new automated valuation modeling (AVM) product that bridges the gap between “overpriced lender-grade AVMs that are only cost-effective for a select few giant corporations … and free AVMs available to the masses but diluted due to their inherent need to be all-inclusive.”
“The ATTOMIZED AVM targets the big middle between those two extremes, providing the marketplace with a lender-grade AVM delivered via bulk data delivery platforms that are flexible and cost-effective,” explains Rob Barber, CEO with ATTOM Data Solutions, in a release.
Valuations from the ATTOMIZED AVM are available for 2,194 U.S. counties, including more than 80 million homes.
It employs multiple valuation models to calculate the estimated value of a home and then applies a cascading model selection algorithm to choose the approach that is most accurate in the geographic area surrounding the individual property.
Accuracy rates are run regularly, comparing historic ATTOMIZED AVM values to actual sales prices, to ensure the model is continuing to cascade to the most accurate approach in every geography nationwide, ATTOM says.
The new AVM is available for delivery via the firm’s various data delivery platforms – bulk file, API, match-and-append, custom reports, and ATTOM Lists.
The model also intentionally excludes several types of properties that can have a negative impact on the predictive value of other homes and that require more situation-specific data and modeling outside of the methodologies that work best for standard residential properties of one to four units.
Some excluded property types are mobile homes, extremely high-end luxury homes, special use properties, homes on farm land, and multi-unit homes with five or more units.
“What we exclude is important,” says Clifford A. Lipscomb, vice chair and co-managing director with AVM Analytics. “Not only do we exclude certain anomalous property types, but we also exclude the subject property itself, making our monthly accuracy tests a true out-of-sample comparison.”
Each property’s AVM includes a confidence score that represents the precision of the AVM estimate and provides the basis for a range of values around the AVM, which is also included in the data delivered to clients.
For example, a property with an ATTOMIZED AVM value of $500,000 and a confidence score of 92 could be expected to sell within 8% of the AVM – a range of $460,000 to $540,000. The confidence score, generated via a re-sampling process, offers a conservative estimate of precision due to our consideration of the full distribution of prices in local markets.
“The ATTOMIZED AVM provides our customers with greater transparency on two levels,” says Andy Krause, principal data scientist with AVM Analytics, who manages the ongoing development of the product. “First, we deliver transparency on the individual valuation level by including greater details of the value estimate range and the process of selecting between our multiple models. Second, we offer transparency on the overall process we use for this AVM in the form of a detailed white paper available to download upon request. Access to this ‘black box’ used to develop an AVM is extremely rare in the industry, but we believe this level of transparency will arm our clients with more complete information to help them make better-informed decisions.”