To keep up on what is happening in appraisal businesses, mortgage lending, USPAP, etc. , Plus humor and strange homes, sign up for my FREE weekly appraisal email newsletter, sent since June 1994. Go to Home on the left side of the menu at the top of this page or go to
Sign up in the Big Yellow Boxes

I regularly write about hot topics in appraising and appraisal business management issues
in my paid Appraisal Today monthly newsletter.
$99 per year or (credit card only) $8.25 per month, $24.75 per quarter, or $89 per year.
For more info, go to


Automated Valuation Models for Appraisals

Automated Valuation Models (AVMs) are always a hot topic for many appraisers. As with AMCs, AVMs have been around for at least 20 years in appraising, starting with multiple regression programs used for mass assessments by assessors’ offices.

My first appraisal job, in the mid-1970s, was field checking property characteristics for a California Assessor’s office that was converting to automated valuations.

The key to an accurate AVM valuation is data. Although many lenders thought AVMs could provide values for all their loans, many areas don’t have adequate data. Some states, such as Texas, are non-disclosure on sales price. Even California, which supposedly has some of the “best” data, has many counties (including urban counties) that don’t provide data on property characteristics or only limited data. Other counties sell their data.

Fannie Mae’s Collateral Underwriter: lots of data from appraisal reports starting in 2011 to be used for their AVM and appraisal report reviews

Fannie has always taken data from appraisals submitted by lenders. But, the data was not standardized. In the past, they scanned paper appraisals, which was not very accurate.

Starting September 1, 2011 the new Uniform Mortgage Data Program (UMDP) from Freddie Mac and Fannie Mae. Specific fields in their appraisal report formats had specified entries for specific fields. For example roof choices: composition shingle, tar and gravel, etc.

In January, 2015, Fannie started using Collateral Underwriter, software that evaluated appraisal accuracy, especially for adjustments such as Gross Living Area. They used statistical analysis. They also looked at their values vs. appraiser values.

Fannie’s data is not available to anyone else: appraisers, lenders, etc. 

Appraisers using MRA to obtain adjustments since 2015

In 2015, when Collateral Underwriter started, Fannie looked closely at the accuracy of appraisal adjustments.

Many MRAs were developed and marketed for appraisers to use.

Appraisals and the bell shaped curve
Appraisers have never been as accurate as mortgage lenders thought.

As appraisers know, values for very similar homes actually fall along a bell shaped. For lending purposes, we appraise the probable sales price along a curve or range of possible values. When two (or more) appraisers value the same home, seldom are the values the same. If you do appraisals for relocation purposes, you learn this very fast. If two values are within 5%, the relocation company doesn’t usually have to order a third appraisal.

That’s one of the reasons lenders like to use AVMs. They don’t have the ‘human” element. But, of course, AVMs can’t get any more accurate than the bell shaped curve.

So what does this mean for appraisals? Maybe collateral risk-based pricing, where more accurate values (less “variance”), with a lower range, will get a lower interest rate. Appraisers need to rethink the service they provide, separating the factual data and trends from the final subjective opinion of value.

Just like underwriting, the borrower’s credit can’t always be done by an automated underwriting program and is referred to a human underwriter as an “exception,” AVMs don’t always work and have to be referred to an appraiser.

Appraisers can provide a range of value, say $200,000 to $225,000, and the lender can pick where along that value range they want to lend. If the value required for the loan is $225,000 they could offer a 3.25% interest rate. For example, the value is $200,000, they could offer 3%, for example.

Risk based pricing for the borrower’s credit is a controversial topic now, but may be the future of lending for both the credit and the collateral.

What is multiple regression analysis?
Before the term “AVM” became widely used, multiple regression analysis (MRA) was the most used term for computer based valuations.

MRA has been around for many years, in many types of applications, for determining which variables are most important in determining an outcome. In valuation applications, an MRA is used to determine if square footage, lot size, age, quality, etc. affect property values, and by how much. College professors (and some appraisers) use MRA in other valuation related applications, which are frequently published in appraisal journals. When I was in graduate business school 40 years ago, MRA was frequently used in research.

When you do a valuation using an adjustment grid, you are doing what an MRA or AVM does to determine value. The MRA or AVM uses mathematical calculations to determine the dollar amount of the adjustments.

In its simplest form, a regression analysis is a calculation that takes a property characteristic such as square footage and determines a value, similar to a square footage adjustment. For example, a 2,000 sq.ft. house X $100 per sq.ft. yields a value of $200,000. In this example, the equation would be Y=100X, where Y is the value and X is the square footage of the home.

MRA is widely underutilized in appraising. We are taught to use matched paired sales to determine adjustments, but how often do you have paired sales available?

Appraisers can use MRA features in spreadsheet programs or statistical programs.

What is an AVM?
AVMs are computer programs that use real estate information, such as demographics, property characteristics, sales prices, and price trends to calculate a value for a specific property.

The basis of AVMs is adjustments, whether used as an MRA, neural network (multiple MRA equations), appraiser emulation, or other method.

Some AVMs use more than one method of calculating a value. Typically an MRA and appraiser emulation are used. Appraiser emulation is more like what an appraiser does. The most similar sales are selected and adjustments are made.

How do AVMs work?
No commercially available AVMs let the user know the equations used in calculating values, as the equations are proprietary.

Sometimes you can find out which variables are most important. A few let you see adjustments that were made on their appraiser emulation models, where adjustments are made on a grid, like on an appraisal report.

What about appraiser/user input?
Some AVMs allow user input, such as changing square footage or inputting a time adjustment, and others don’t allow any input.

The main reason for not allowing user input is to eliminate the human element from valuation. These AVMs are seen as “pure” without human errors and bias.

Good data is the key
Public records is the primary data source for all commercially available AVMs. This data is typically purchased by the AVM vendor. Some vendors collect part of their data, particularly if they work only a few states. Public records data is used for both the subject and the comps.

Freddie and Fannie use their extensive appraisal database for their AVMs.

Some vendors use additional enhanced property data, with information from MLS, appraisals or other sources.

If an address is input and the AVM has no property characteristics or previous sale (for price trending) it cannot be run. Only one vendor I spoke with, Solimar, will run an AVM without property characteristics on the subject.

If the property is in a nondisclosure state, public records are of no use unless a sales price can be obtained from another source, such as MLS.

No AVM provider has 100% coverage in the U.S. In many areas, particularly rural areas, no one compiles and sells the data.

National AVM vendors compete on how many counties they can cover.

What about MLS data?
MLS data is widely available online now. Previously real estate agents did not want their listings and sales to be public.

Assessments – a significant threat to appraisers
In many states, tax assessment information can be used to determine values. This is particularly useful for a local appraisal firm or lender, who is familiar with local assessment practices. Some assessments are at 100% of market value and some are at 6%. If assessments are equalized in a neighborhood or city, all you have to do is apply a factor to the assessment.

Who inspects the properties?
One of the major weaknesses in AVMs is a lack of a physical inspection of the properties.

Most lenders I have spoken with using AVMs for home equity and seconds don’t have anyone drive by the properties. They see them as credit loans, with the collateral very secondary. Banking regulators, and many investors, agree.

Methods such as verifying an address with the U.S. Postal Service and using aerial photography are possible strategies for cutting down on fraud. But there is no substitute for a physical inspection.

Freddie and Fannie are taking a more conservative approach (for now) and having appraisers inspect the properties to see if there are any problems.

Who has developed AVMs for lenders?
The oldest AVM for lenders I know about is PSAR (Property Survey Analysis Report), which was developed in the 1980s by an appraiser, Robert Maxfield, Sr. for use in his probate and divorce appraisals. When I started my appraisal business in 1986, one of my primary lender clients was using PSAR for seconds and home equity loans.

There are many, many AVMs available today. Zillow uses their AVM for home “prices”.

Use of multiple methods
Many AVMs use more than one method of valuation. Typically appraisal emulation and purely statistical valuations are used. The more reliable estimate is given more weight in the valuation.

How do you know the accuracy of an AVM output?
Many outputs give you an “accuracy range” similar to a standard deviation.

Lender users can set a cut-off for acceptance or rejection of a value. For example, accepting any values with a 10% or less variation.

What’s the best AVM?
No AVM is perfect. They all have data problems. The best AVM for a user gives the most accurate results with the most “hits,” for their “comfort level” of acceptable risk. Some models perform better than others, but all depend on data to get reliable results.

What about USPAP? 

If you are only running an AVM (typically typing in an address) and are not “signing off,” you are not subject to USPAP, and are simply performing a clerical function. If you are asked to sign off on AVM values, be sure to read USPAP.

To read more, go to, click on Appraisal Standards Board, then click on USPAP.

Opportunities for appraisers
If you’re in an area where public records information is poor and tax assessments are very inaccurate, you’re pretty safe from AVMs.

If you’re in an area where AVMs are a threat, while your competitors have their heads in the sand, you can expand into appraisals that aren’t automated, such as relocation and REOs.

Where to get more information

Google Automated Valuation Models to see what is available.

Fannie warning letters – GLA adjustments and lots more coming(Opens in a new browser tab)

Data is King for Appraisers