Home / Commentary / AVMs: Then, Now, and in the Future 
Print This Post Print This Post

AVMs: Then, Now, and in the Future 

Editor's note: This feature appears in the August 2021 print issue of DS News magazine, available here

“Necessity is the mother of invention.” This old proverb best sums up the story of how automated valuation models (AVMs) have evolved from their humble beginnings to today’s industry-wide acceptance. 

The year was 1994, and I went to work for a small appraisal company in Southern California. For those who can remember that far back, appraisal processes—and especially the collection of prior comparable sales searches (comps)—were tedious and time consuming. MLS listings were almost totally paper-based and finding the right comps could take up to an hour searching microfiche and other sources. 

The process was broken and something better was needed, but what? Armed with a mission to invent a new solution, a couple of us within our own appraisal company set out to build a tool, not to value properties, per se, but to simplify identification and scoring of comps. What started out as an in-house project to improve efficiency eventually became one of the first AVMs, Valuate, which launched in 1995. That first AVM model required county-level property data to be accessed via telephone modem, connecting to mainframe computers for each individual comp search. That process could take five minutes for each search. The model would score the comps and the appraiser would pick and adjust the comps used in the final valuation. As poor of an experience as that might sound today, it was considered revolutionary at the time—perhaps too revolutionary. 

Early Challenges 

Most independent appraisers never adopted early models like Valuate because they didn’t want to spend the money, they were leery of the technology, or they simply felt more comfortable to keep doing it the traditional way. Our AVM did give us a competitive edge versus other appraisal companies in one area: responding to mortgage brokers who would call and say, “I’ve got a deal working and need to know whether the property can value at $200k?” While that doesn’t sound like a problem today, in those days there was no easy answer to that unless you were either a skilled local real estate agent or a certified residential appraiser. 

What started as an idea to assist the appraisal industry soon blossomed into a very real and compelling opportunity for the lending industry. Lenders needed a tool for their underwriters to quickly validate the appraisal value and give those underwriters confidence that the value was in the acceptable range. While the original AVM models let the appraiser choose and adjust comps, lenders did not want their own underwriters to be able to manipulate comps. As a result, that functionality was blocked. Lenders required an AVM model that provided more of a blind (or non-human influenced) comparison and value, and we were quick to meet those requirements. 

Soon after, our company introduced our first commercially available AVM, ValuePoint, which had a full graphical user interface and could perform the valuation without human assistance. This meant that non-appraisers could use and understand it. The ValuePoint model was licensed by First American as part of First American’s emerging property data solutions division. First American was so impressed they eventually bought it. By this time, our company had become 100% committed to AVM development and was hard at work creating another game-changing solution. In 1997, we released our newest model, the PASS AVM. We licensed the public record data in bulk and created a proprietary database that could leverage high-speed mainframes, cutting response time to less than a second. For the first time, an AVM was accessible to any user on a self-serve basis, on a website, in real-time with delivery in seconds, not minutes, and required no pre-installed software.  

In the following years, by adding additional data sources and sub-models, AVMs evolved into a more sophisticated, regression and rules-based technology which could emulate market prices and property characteristics and then apply retrogression analysis to track the history of home prices at the neighborhood level. By the early 2000s, several major companies were offering AVMs and the GSEs had even developed their own versions for use with their own automated underwriting systems.  

Even with those advancements, issues persisted. Some models would come up with more hits (a “hit” is the ability of an AVM to actually render a value result on a particular property) than others, and some AVMs performed better in different markets. To address this, the concept of an AVM cascade was created. An AVM cascade uses multiple AVM models that “rotate” from one model to another until one returns a valuation at a certain confidence score level. Users could tailor the AVM cascade models used in specific markets or situations for optimum accuracy and hit rates. These cascade models became increasingly popular, and AVMs became even more mainstream. 

By the mid-2000s, the key to the growing acceptance of AVMs was their increasing accuracy and performance. This was being achieved through expanding access to multiple data sources, database management, and high-speed machine computing and learning. The ability to update the property information on a weekly, and then daily, basis moved AVMs even closer to real-time. AVM providers were internally testing literally billions of valuations produced by these AVMs. Today, AVM testing has expanded ten-fold to build better confidence scores. Accurate confidence scores have become vital to understanding the valuation model’s reliability. 

As a result of the mortgage crisis from 2008-2010, the use of hybrid valuation methods that combined technology, modeling, and human review with photos of the property was introduced. The human element in hybrid AVMs addressed risk concerns that a fully automated solution might present. Hybrids, in general, were cheaper than a full appraisal but more expensive than a traditional AVM. These hybrid models were also driven by the shrinking population of certified residential appraisers and competition among lenders to secure deals and to give borrowers a cheaper and faster alternative to determine values. Hybrids enabled lenders and the GSEs to select what combination of appraisal and automated tool that best fit the risk profile of the transaction. Combinations included just an AVM, or an AVM with an inspection, a drive-by review, a desk review, a broker price opinion (BPO), or even a full appraisal.   

Throughout this time, these solutions were in place primarily for high-equity refinance transactions, servicing solutions, and distressed asset disposition. Today, the use of hybrid valuation models is increasing, and they are even being used to support certain types of lower-risk purchase transactions. Appraisal waivers and the resulting ability to use hybrid solutions have seen a sharp increase over the last few years, even before COVID-19, but accelerating sharply due to the pandemic. It would appear that hybrid valuation solutions are here to stay. 

 Where Are We Today and Where Are We Going? 

AVM modelers are now looking at more ways to access even more data points that can add nuance to existing models beyond recorded sale prices. This data includes local price volatility, list-to-sale price ratios, time on market, geospatial information, the impact of natural disasters (e.g., fire, flood, storm surge), and probably the most important—property condition. New technology is also being considered and implemented in AVM modeling, such as text mining on MLS listings, image recognition, artificial intelligence, and more.   

Cloud computing platforms are being used to model, run, and test AVMs and provide lenders, servicers, and investors with instant access to the valuation outcomes. Leading AVM companies are also looking into the ability to input even more technology-driven sources of data, including drone inspections of a property’s outer condition in the front and backyard, and even borrower-supplied data from mobile apps. New geospatial data technology is also being leveraged to determine true neighborhood boundaries.  

As data sources, technology, and testing methodologies evolve, the AVM industry is moving towards significantly advanced AVM testing based on artificial intelligence and machine learning, allowing users to know exactly what performance to expect in real world applications. The industry is also evolving and creating different versions of AVMs to address the different use cases needed. For example, one version of AVMs for general consumers who are looking at values on a real estate website, another for white-label values for tools like mortgage calculators, a premium level for lenders, and yet another for portfolio management and investors. 

From their humble beginnings to help appraisers save time determining comps, AVMs are now universally used by the entire housing industry. AVMs have saved consumers, lenders, servicers, investors, and appraisal management companies (AMCs) hundreds of millions of dollars in cost and time and have helped mitigate potential risk of over- or under-valuing a property. While the valuable contributions and services that are provided by certified residential appraisers will never cease, AVMs will continue to grow in supporting the needs of the valuation industry.  

For this AVM geek, on a personal level, it’s been a great source of pride to have been part of the AVM’s past and bright future. 

About Author: Jon Wierks

Jon Wierks is VP, Analytics, for First American Data & Analytics, a division of First American Financial Corporation (NYSE: FAF). In this position, he leads the development and support of First American’s valuation solutions. Wierks has over 25 years’ experience in the real estate and valuation space as a creator/co-creator of several of the industry’s highly regarded AVMs and has earned six valuation-related pending or issued patents.

Check Also

Studio Apartment Prices Lead Rent Charge

As prices on everything increase across the board, more and more are turning to renting—either ...