Next generation AVMs wait in the wings

From the acquisition of new data sources to integration of the latest property location tools, the next wave of automated valuation models promises to enable smarter decision making.

Automated valuation models (AVMs) have been in use for more than a decade in the mortgage world and, like other areas of technology, are due for an overhaul, according to some experts.

What will “version 2.0” of AVM technology look like, and how will next-generation tools benefits lenders, investors, appraisers and others? In recent years, high-volume lending was the driver for AVM adoption, as there weren’t enough appraisers to satisfy valuation needs. The technology’s use in home equity lending helped it gain a foothold in origination.

“Now that it’s slowed, that’s not going to be a driver. I think the unbiased nature of them is going to be a driver,” said Lee Kennedy, founder and managing director of
AVMetrics, which works with lenders and investors on their AVM strategy and model testing.

On the other hand, users will continue to explore appraiser-AVM hybrid valuation methods.

“There’s been a push for a long time to involve the appraiser in the process,” Kennedy said.

The biggest reason is to generate more relevant data for underwriting. For example, there are two key areas where AVMs haven’t yet been able to provide good analysis — property condition and property location.

“Of those two, property location is probably going to be the next generation of AVM,” Kennedy said.

That will translate into integration of geo-coding, spatial analysis and other emerging tools.

“(AVMs) are going to be able to develop a relationship, mathematically and statistically, between the location of the subject property and the comparables and the relative location of that property within that defined neighborhood,” Kennedy said.

That means smarter market segmentation and smarter analysis of how a subject property compares to surrounding properties in terms of positive and negative external influences. “If you’re in the flight zone of a major airport, is that a positive or a negative?” Kennedy said. “If you’re sitting on a golf course, positive or negative?

A lot of that is already available through the assessment districts. And obviously, appraisers have that knowledge. Now you’re seeing companies like Zaio capture that information even better.”

AVM technology has now matured to the point where vast quantities of such data can be transferred to models. “As geo-coding, spatial analysis and other components
become available in the larger MSAs, you’re going to be able to tell the house that has the view and the house on the golf course versus the one next to the railroad tracks,” Kennedy said. “The models are going to get much more accurate about defining location long before they’ll ever be able to define condition.

“The next generation of models? It’s all going to be based on location.” What constitutes ‘next generation’? “Intelligent segmentation” is one of the keys behind IntelliReal’s vision of what the next wave of AVM developments will bring.

The company provides the iAVM platform to real estate brokerages and franchises, lenders, mortgage service companies, Wall Street professionals and others. It’s in a technology partnership with Integrated Assets Services (IAS), which uses IntelliReal’s iAVM as a platform for its own risk management valuation solutions.

How areas and neighborhoods are grouped and analyzed by valuation models is evolving, according to IntelliReal CEO Eric Miles. For example, models already are migrating from working at the ZIP Code level.

“Within a given ZIP Code, depending on what neighborhood you’re in or what side of the tracks you’re on, you will have dramatically different valuation criteria and market dynamics,” Miles said.

Cutting-edge models are segmenting markets into “intelligent segments,” according to Miles. iAVM, for example, has 17,000 market segments, or “neighborhood clusters,” with a unique model for each.

Intelligent segmentation will have huge repercussions for secondary-market businesses that rely on AVMs to conduct instant batch valuations on loan pools containing thousands of loans.

Next-generation AVMs will enable users to track price volatility in a given segment. If a particular neighborhood is declining, they can see it.

Several other factors will define emerging AVMs, according to Miles, including the ability to incorporate new data sources and codify parts of the appraiser’s decision-making process.

A next-generation AVM will automatically take advantage of ever-expanding data sources. Data is filtered through the valuation model, which has to take into account data from sources such as public records, CraigsList, census data and the MLS and “mash it up” to integrate with publicly available technology such as mapping systems, geo-spatial data and images from Zaio, the MLS or an assessor’s office.

“That’s the mark of a next-generation AVM: taking advantage of broad and heterogeneous data sources,” Miles said.

AVM accuracy is only as good as the analysis the model performs and the quality of the data that goes into that analysis. Assessor data can be spotty at times, Miles said, and states such as Indiana and Texas don’t even disclose property sales figures.

The need for data sources beyond public data resources therefore has AVM providers thinking creatively.

Many are eager to tap into MLS, Google Base (FSBO) and other sources. MLS data, however, can usually only be accessed by an authorized MLS associate. IntelliReal has worked out an agreement —this part of its business model remains a key piece of the company’s “secret sauce” —that gives it data feeds from 250 MLSs. The company can turn on those feeds based on having a broker in that area who’s authorized.

Artificial intelligence can also play a role in taking AVMs to the next level. Data such as comments in an MLS listing isn’t structured, and AVMs will have to get smarter about weighing and codifying such information.

Still, Miles said, the evolution of AVM technology won’t require any more “big pieces.” The tweaks to models will be of a finer grade. Small pockets of data related to interest rates or housing starts will help AVMs value property and predict future directions, for example.

“An interesting feature of a next-generation AVM is the ability to set up a date,” Miles added. “I can go back in time and tell you what it was worth six years ago, and I can tell you what it’s going to be worth three years from now.”

IntelliReal is putting some of those theories to practice with its iAVM offering, which has been in development for three years and in production for one. The platform consolidates different valuation models in a “mini cascade.” For any property, iAVM runs up to four different AVMs, including:

• A classic trended model that looks at most recent sales and movements in the market;
• An appraiser-type model that uses artificial intelligence and determines values based on sales comparables;
• A hedonic model, where regression analysis weighs different property attributes; and
  
• Assessor value, what an assessor believes the property to be worth. If an assessor is always 15 percent under value, the model can compensate for it.
  

Each model offers a confidence score, and a final algorithm selects the best.