MythBusters: Are AVMs Stealing Your Data?

“Your magazine writes that ‘purveyors of AVMs have routinely lifted data from appraisal reports for years without asking permission or paying remuneration.’ I have been hearing this over and over again. However, I have never seen one piece of evidence that proves this assertion. Do you actually have proof?”
– Michael R. Cartwright, CMA/RPG

MythBusters: Are AVMs Stealing Your Data?

In response to a reader’s question (above), WRE asked the following five AVM experts whether “intelligent systems” (AVMs) are actually “mining” appraiser’s data. Here’s what they say:

>Darius Bozorgi, President & CEO of Veros, a leader in enterprise risk management and collateral valuation services
In general this concept has been one of the single largest misconceptions in the appraisal community. Unfortunately, a small number of appraisal industry members have perpetuated this myth either as unwitting participants or intentionally to serve some other agenda 

WRE: Does Veros mine appraiser’s data?  

Bozorgi: Veros does not mine or use appraiser’s data and has not done so in the past. I am not aware of any AVM that uses appraiser data or mines appraiser data. I have heard that there may be one or two AVMs that had/have an appraiser data database (in addition to other data sources), but I do not know to what extent they would rely on it or weight the data in their model(s). One of these providers has gone out of business. Any such references are rumor only.

WRE: With the advent of the new XML “standard,” do you think the mining of appraiser’s data is coming?
Bozorgi: I do not believe that the advent of XML in and of itself will or will not lead to mining of appraiser’s data. It’s a misleading issue as the use or non-use of appraiser data has little to do with its accessibility or lack thereof (and in what form). Instead, the viability of any data source (public record, economic, appraiser, broker, lender, MLS, government, association, or otherwise) is primarily dependent on its potential contribution to the accuracy or other performance consideration of an AVM model.

To date, the AVM industry has not seen sufficient “lift” in appraiser data to justify or consider its inclusion as part of the fuel for its models. That’s not to say it does not have value but only that in concert with what we already have appraisal data does not appear to compliment these data stores such that it would significantly increase AVM performance. Moreover, the future direction of AVM development does not appear to be pursuing appraisal data on any wide-spread or consensus basis. Instead, most developers seem to be pursuing technologies and data such as GIS solutions for example. Certainly, this may change at some point in the future, but I believe this is an accurate state of the current AVM market relative to its use (actually lack of use) of appraiser data. The advent of an XML standard adds nothing to the ability of interested parties to “mine” data. For example, the technology to extract text and other data from any electronic document, PDF included, has existed for some time (e.g. OCR).

Find Mr. Bozorgi’s Complete Bio at bottom.

>Brad Ellis, Chief Appraiser for Valuation Quality Assurance for Indymac Bank, America’s seventh largest mortgage lender
WRE: Do AVMs mine appraiser’s data?

Ellis: For the most part, no. Of all the AVMs available out in the market, the only two that have admitted actually “mining” data are Fannie Mae and Freddie Mac. The Fannie Mae AVM is used only internally and its output is not publicly available; that is to say, no one can actually buy one from them. Conversely, Freddie Mac’s product, Home Value Explorer, is sold publicly.

Important in this discussion is what data is actually even mined by these two AVMs. Frankly, it is doubtful that either firm collects data from an appraisal on any property except for the subject of the appraisal. However, this information is generally proprietary so little information on this is available. Both GSEs are able to do this due to the large number of loans that pass through their securitizations. 

Virtually all the rest of the AVM firms do no mining of appraisal data at all. Instead, they rely upon whatever data is available from published public records. Obviously, in areas where public records are not published, hedonic AVMs will not run. The one exception to this may be TransUnion’s Value Wizard that actually has some MLS data available and so it will sometimes provide a value estimate even in areas with limited public record data or in non-disclosure states.

Appraisers should bear in mind that the reliability of any computer generated value will always depend directly upon the quantity and quality of the data available. This is really no different from what an appraiser is faced with in any given assignment. Indexed AVMs can run wherever the sales prices are published and they are simply based upon changes in overall selling prices for that specific market.  

WRE: Will XML change things in the future?

Ellis: Yes. In fact, this language is already being used to filter appraisal data so that certain client specific rules can be applied in the quality control process. However, mining of specific data from appraisals, if it ever happens, will probably be very limited, at least for the foreseeable future. The reason is the reliability of the data within appraisal reports. An appraiser may know that public records for a property are wrong since they will have inspected the subject in many cases or may have MLS data indicating additions, etc. that have not yet appeared in public records. But because it can not be confirmed within any sort of reasonable margin for error when compared to public record, it is a questionable piece of data within any database. Therefore, mining data that could be wrong will corrupt any sort of database that may be envisioned. Certainly, subjective areas such as condition, appeal, effective age, etc. can be so unreliable that most would simply ignore those ratings in an appraisal. Some of the data cannot be confirmed by any other source. Therefore, this is similar to “hearsay” evidence.

There also are many factors considered by appraisers in their value opinions that could never be measured by a computer. Topography, views, and similar factors would be very hard to quantify. While externalities like proximity to railroad tracks and similar influences may be measurable, how would a computer “judge” the difference between a panoramic vs. a partial ocean view, for example? The far more likely scenario will be for these data firms to aggregate data to provide some quality control testing.

There is a segment of the appraisal profession that has been very concerned and very vocal in their opposition to data mining. In my view, this concern is wildly over-stated. Will data systems ever replace appraisals for certain types of lending situations? Absolutely- they already have. HELOCs and rate/term refinances are already being done without appraisals all over the country. However, the need for professional work levels from localized firms will not disappear. I believe that good, knowledgeable appraisers who know how to provide proper professional service levels will really prosper. Those appraisers who know nothing more than how to fill out forms and hit numbers will ultimately perish.

Find Mr. Ellis’ complete Bio at bottom.

>Mark T. Simpson, Director Property Standards Single-Family Mortgage Business Fannie Mae
The goals of improving property valuation quality and reducing loan origination costs led to the development of Fannie Mae’s proprietary AVMs and minimum property fieldwork recommendations offered through Desktop Underwriter. We believe AVMs can be used as meaningful tools to enhance credit risk management and streamline the mortgage loan origination and servicing functions.  The use of AVMs in Fannie Mae’s technology for processing loans and recommending property valuation and inspection options has proven to be effective. However, AVMs have generally not evolved sufficiently to fully replace traditional appraisals and human judgment for the origination of mortgage loans.

Our AVMs rely on comparable sales that are confirmed through loan originations that are delivered to Fannie Mae and public record data sources, such as deed and tax assessment records.  And our database is proprietary; we do not sell or re-distribute the data. 

Fannie Mae’s Perspective on Automated Property Valuation Models

Fannie Mae’s mission is to tear down barriers, lower costs, and increase the opportunities for homeownership and affordable rental housing for all Americans.  To fulfill this mission, we are committed to working with our customers to develop solutions to their business needs that benefit consumers and reduce the time and costs of mortgage loan origination.  Fannie Mae’s automated underwriting system, Desktop Underwriter (DU), is one example of how we have helped lenders meet the borrowing needs of a wider spectrum of homebuyers, while reducing the costs of financing a home.  Similarly, Automated Valuation Models (AVMs) can be helpful tools to streamline the traditional property appraisal process, an important component of the mortgage loan origination process.

AVMs are statistically based computer programs that use real estate information such as comparable sales, property characteristics, tax assessments, and price trends to provide an estimate of value for a specific property.  The goals of improving property valuation quality and reducing loan origination costs led to the development of Fannie Mae’s proprietary AVMs and minimum property fieldwork recommendations.  Fannie Mae’s use of AVMs to recommend streamlined property valuation and inspection options benefits both lenders and borrowers by reducing the costs and time delays typically associated with the traditional property appraisal process, without sacrificing credit quality.

Fannie Mae believes AVMs can be used as a meaningful tool to enhance risk management and streamline the mortgage loan origination and servicing functions.  At present, we believe AVMs have generally not evolved sufficiently to fully replace traditional appraisals and human judgment for the origination of first lien mortgages.  However, the use of AVMs in Fannie Mae’s technology for processing loans and recommending property valuation and inspection options has proven to be effective.  In addition, Fannie Mae does not approve or endorse third party AVMs or insured or warranted property valuation products.

Overview of Fannie Mae’s Use of and Policies on AVMs
DU relies on our proprietary AVMs and credit risk management techniques to determine the minimum level of property valuation and inspection fieldwork required for loans processed through the system.  As part of its risk analysis, DU makes a property valuation and inspection recommendation based on a variety of factors, including the overall credit risk of the loan transaction and how robust our data are about the particular property.  DU assesses the reasonableness of the property sales price (or the lender’s estimated value for a refinance transaction) and recommends to lenders the minimum level of property fieldwork that must be performed for the loan to be delivered to Fannie Mae.  A lender may elect to obtain only the minimum documentation we require, or it may require additional documentation for any reason.

As a result of the DU risk analysis, the system makes one of three fieldwork recommendations: an appraisal with an interior and exterior inspection, an appraisal with an exterior-only inspection, or an exterior-only inspection.  When DU recommends an exterior-only inspection of the property, an appraisal is not required.  We are relying on our AVM to validate the lender-reported value.  In such cases, the traditional lender representations and warranties on property value for loans delivered to Fannie Mae are waived.  However, the lender continues to be responsible for the representations and warranties on property condition and marketability.  We also offer a waiver of the exterior-only property inspection for a fee.  In such cases, lenders are given the option to waive the property fieldwork and condition representations and warranties traditionally associated with our “no appraisal” offering.  Lenders must register for this offering to have access to that functionality in DU.

We have also implemented property valuation-related messaging in DU to identify transactions that appear to have excessively high sale prices or estimates of value.  Our intent in implementing this functionality is to provide practical property-related quality assurance information to the lender during loan origination. The use of this technology has enabled Fannie Mae to better manage property valuation credit risks. 

Strengths and Weaknesses of AVMs 
The strengths of AVMs relative to traditional real estate appraisals are speed, reduced costs, consistency, and objectivity.  This is not to suggest that conscientious, skilled appraisers lack consistency and objectivity.  However, an AVM can significantly reduce the time it takes to obtain an estimate of value and reduce the costs associated with the traditional property appraisal process, which can be particularly helpful for quality control applications and for the loan processing of lower risk mortgage loan transactions.

AVMs suffer from three principal limitations: First, they are dependent on the accuracy, comprehensiveness, and timeliness of the data they use.  Data issues can include inaccurate or incomplete sale (or transfer) transaction records, insufficient sales of properties with comparable features within a specified geographical area, and a lag between the time when the market data is current and the AVM uses the data to generate an estimate of value.  Second, AVMs generally cannot be used to determine the actual physical condition and relative marketability of a property.  And third, AVMs are dependent on the accuracy of the underlying system’s property valuation methodology.  AVMs are statistical models, which typically produce an average or mean estimate of value that does not necessarily reflect the market value of the property.  In addition, AVMs can never fully incorporate the breadth of knowledge and judgment of a skilled appraiser.

Consequently, AVMs tend to work best in circumstances when there is a relative abundance of accurate and current data, when properties in a given area are relatively homogenous, and when a property’s condition and marketability are relatively typical for the area.  They work less well when data are thin, in heterogeneous neighborhoods, and for properties that differ markedly from the average property condition or marketability. Therefore, it is critical that users of AVMs design an appropriate use and implementation strategy that considers the overall credit risk of the loan transaction and reflects the specific strengths and weaknesses of the particular AVMs and the property data supporting those products.

Fannie Mae believes AVMs can be used as a meaningful tool to enhance credit risk management. Indeed, they are becoming effective tools to support the mortgage loan origination and servicing functions.  Lenders use AVMs for a variety of applications, including quality control, loan origination, portfolio management, and marketing.  Although they are relatively new to the first trust mortgage arena, their use is expanding.  AVMs have many strengths, which must be viewed in balance with their weaknesses.  It is critical that users of AVMs understand the specific strengths and weaknesses of the different models and property data supporting those products in order to design an appropriate use and implementation strategy.

At present, we believe AVMs have generally not evolved sufficiently to fully replace traditional appraisals and human judgment for the origination of first lien mortgages. In addition, Fannie Mae does not approve or endorse third party AVMs or insured or warranted property valuation products.

We will continue to work with our customers to explore ways to further streamline these processes.  Our focus remains the same – to design solutions that serve the business and competitive needs of our lender partners and benefit consumers.  To do so helps us fulfill our mission to tear down barriers, lower costs, and increase the opportunities for homeownership and affordable rental housing for all Americans.

Find Mr. Simpson’s Bio. at bottom.  

>Lewis Allen, VP, Chief Appraiser Option One Mortgage
Lewis Allen: I think the assumption is correct that AVMs are not mining appraiser’s data, mainly because of the inconsistencies in reporting and that most of the data in the appraisal was acquired elsewhere. Many of the fields on the report allow for variations in the data inputted. This is not a bad thing and since no two properties are alike the appraiser needs some latitude in how the information is reported. However, this would cause problems when mining the data.

AVMS use public records data and/or Multiple Listing Service (MLS). The only item of interest is the subject property because the appraiser actually inspects it and it’s the only piece of first-hand information that is not available elsewhere. But this data, as noted, is not reported in a consistent enough way to be used in these intelligent systems. Also, it can be difficult to verify.

WRE: What should appraisers know about the future and AVMs?

Allen: Appraisers should be embracing AVMS for use in the valuation process to give their reports more credibility, especially as the issues of fraud and pressure on appraisers become more intense. At Option One we’re testing a hybrid system where, ultimately, we’d like to use the intelligent systems to provide an analysis of relevant data and then send an appraiser out to use their expertise to evaluate things like location, condition and marketability- how the property fits into the market. These are the things that an AVM can’t provide and what appraisers are trained for. The appraiser might come back and confirm that the value provided is acceptable for the subject property, or they might indicate an adjustment is necessary because of a condition or influence that cannot be captured or analyzed by an automated system. We want an honest and independent assessment of the property.

Allen said the fees for these types of services are often at an hourly rate many times higher than traditional appraisals. “Appraisers are still needed to examine and analyze the market and the subject property’s position in that market. Marketability is becoming increasingly more important,” Allen said.

According to Allen, the business of data gathering has changed but appraising has not. Appraisers still use a system that is a vestige of an era that is decades old. “Thirty years ago when data was difficult to get, gathering it was a valuable service,” said Allen. “With all the information available today in seconds on the Internet, pulling three comps does not have the value it once had. Today, we need to know how the property fits into the market. It’s all about risk. We do not lend on value, we lend on the property. After all, if we foreclose, it’s not the value we are taking back,” Allen said.

Find Mr. Allen’s Bio at bottom.

>Bill Rayburn, MAI, SRA is Chief Executive Officer, Chairman of the Board and FNC Co-Founder
WRE: Does FNC/Appraisal Port mine (take/reuse) appraiser’s data? 

Rayburn: AppraisalPort does not take any data. Here’s what we do: we open up the appraisal report on the lender’s side, convert that document to data and stream the data to the lender so that they can take it, review it, and archive it in order to be regulatory compliant.

This is the same basic function that has always existed—long before AppraisalPort—except that the lending institutions using our solutions store the appraisal reports on the loans they fund or service as data sets instead of a big, thick paper files in a filing cabinet. Lenders have always had access to the information in an appraisal report. AppraisalPort and FNC have only modernized that access. 

WRE: Will XML change anything?

Rayburn: An XML standard is nothing new—FNC, through our relationship with the Appraisal Institute, has been using the AI Ready XML standard for several years. We’re also happy to translate reports in MISMO XML for our lenders, depending on their preferences. Our clients need data, not documents. An XML standard allows lenders to streamline the appraisal review process and makes review totally consistent within a lender. Receiving appraisals as XML data allows for automated decisioning of the appraisal report, something that benefits everyone in the process by speeding turn times and lowering costs.

XML standards and the lenders who use them are not the enemy. If anything, this increased efficiency helps appraisers get more work and become more important to the lending process as appraisals become more standardized.

Please find Mr. Rayburn’s Bio at bottom.  

>Darius H. Bozorgi is President & CEO of Veros, a leader in enterprise risk management and collateral valuation services. Mr. Bozorgi has several years experience in the world of automated collateral assessment products and other predictive technology based decision-support applications for the financial services, financial investments, aerospace and engineering industries. He serves on the Board of Directors of the Real Estate Information Professionals Association (REIPA) – an organization of many of the largest real estate data and technology companies in the country. Bozorgi is the Chairman of the recently formed Collateral Assessment & Technologies Committee (CATC) of REIPA representing most of the nation’s leading collateral assessment technology companies. Bozorgi co-founded Veros after spending most of his early career practicing law in Chicago, where he specialized in civil litigation for one of Illinois’ largest litigation firms. He has also been involved in the successful capitalization and operation of several technology and finance ventures. He has served on the board of directors of several companies and currently serves on the board of Credit One Corp., located in Santa Ana, California. Bozorgi received his J.D. from Chicago-Kent College of Law and his undergraduate business degree from the University of Michigan, Ann Arbor. Bozorgi is frequently asked to lecture and speak on practical applications of predictive technologies in the financial services industry. He lives in Southern California with his wife and three children. Mr. Bozorgi’s email: Telephone: (714) 415-6390

About Veros
Veros Real Estate Solutions, a proven leader in enterprise risk management and collateral valuation services, uniquely combines the power of predictive technology, data analytics and industry expertise to deliver advanced automated decisioning solutions. Veros products and services, integrated into industry leading companies, are now optimizing millions of profitable decisions throughout the mortgage industry from loan origination through servicing and securitization. Veros provides solutions to control risk and increase profits including automated and secured valuations, fraud and disaster risk detection, portfolio analysis, forecasting and next-generation collateral risk management platforms.

>Brad Ellis, a frequent contributor of WRE, has been a full time appraiser for 20 years, serving as past Director of the National Association of Independent Fee Appraisers (NAIFA), former panel manager and quality control manager for HUD’s AQA program via the REAC and an early Appraiser Qualification Board-certified instructor. He is currently Chief Appraiser for Valuation Quality Assurance for Indymac Bank, America’s 7th largest mortgage lender.

> Mark Simpson is Fannie Mae’s Director of Property Standards in Fannie Mae’s Single-Family Mortgage Business department in Washington DC.  He has held various positions at Fannie Mae since joining the company in 1986.  Mark is responsible for the development and implementation of Fannie Mae’s single-family property valuation standards, the property component of Desktop Underwriter, appraisal report forms, property underwriting guidelines and loan eligibility requirements based on market and industry developments and trends.  He is a graduate from the University of Maryland College of Business and Management.

> Lewis Allen’s career in real estate and mortgage banking has spanned over 25 years and includes Title Insurance, Appraisal, Underwriting, Marketing, Affordable Housing and product development. He is an educator and instructor, having taught and developed courses in Appraisal, Underwriting, Banking Regulations, Fraud and Real Estate. He has played a leading role in the development of automation techniques as they apply to the valuation process within mortgage banking.

With a focus on the integration of new technology with business policies and applications, he has been instrumental in the development and implementation of Automated Valuation Models (AVMs) for the mortgage process. He has been a pioneer in the deployment and understanding of AVMs in the collateral valuation function, which included the acceptance and adoption by major institutions, investors, secondary market, and Wall Street.

Most recent activities include the development of the first Collateral Risk Scoring system that merges credit criteria with collateral risk. This process provides an analysis on each individual loan to determine the best course of action and the corresponding risk parameters. It also reduces costs to the borrower, saves loan production time, and eliminates a considerable amount of fraud potential while making decisions in a more consistent manner.

Mr. Allen’s activities have included working with some of the largest banks, mortgage lenders, data providers and software developers’ focuses on the integration of technology and alternative business concepts into common sense solutions for today’s mortgage banking needs. This includes point of sale origination, marketing, underwriting operations, quality control and portfolio analysis. Mr. Allen has helped his clients make sense out of the menagerie of technology vendors, combined with outside of the box processes and techniques to create opportunities that best fit each institution’s business model. The ultimate goal is to reduce costs while improving overall efficiencies with more acute risk identification. This includes analysis and assessment of current protocols, process re-engineering, marketing strategies, policy and regulatory compliance, training and education, and adoption and implementation of technology.

> Bill Rayburn, Chief Executive Officer, Chairman of the Board, FNC Co-Founder
Bill Rayburn has served as FNC’s CEO since the company began. A finance expert and visionary, Bill intimately understands how collateral underpins the mortgage industry and has a special talent for imparting that knowledge to others. Bill often speaks or sits on panels at major industry conferences and led a consulting firm specializing in seminars for bank regulatory agencies, financial institutions, and appraisal firms before founding FNC. At FNC, he is responsible for steering the company’s mission and vision through the Age of Collateral. 

A former business professor at the University of Mississippi, Bill earned his Ph.D. in Business Finance, Chartered Financial Analyst designation, and Appraisal Institute MAI and SRA designations. 

Along with FNC co-founder Dennis Tosh, Bill quite literally wrote the book on collateral – actually, a few of them – including Sheshunoff’s “Bankers Guide to Real Estate Appraisal Compliance” and “Uniform Standards of Professional Appraisal Practice: Applying the Standards

Tags: , , , , , , , , , , , ,