Using MLS data to generate an AVM

by | May 29, 2014 | MLS and NAR | 0 comments

The question of whether there is any real value to using the MLS data to generate an AVM was recently discussed in the RealTalk forum. Here are some excerpts:

Steve Erwin asked –

“Is there any REAL value to using the MLS data to generate an AVM?”

First off for those who may not know, let’s understand what is an AVM (Automated Valuation Model). It is an estimate of a property’s value using a proprietary formula based only on PUBLIC RECORDS. AVM’s have been around for over 15 years and there are different types.

1. Hedonic – treats the property as a bundle of characteristics (# of bedrooms, # of baths, etc.) the sum of which determines the estimated market value.

2. Indexed – predicts valuation based on sales trends in a geographic area. Specific property aspects are not considered except to select a number of similar properties.

3. Blended – combines both sales trends and property details: uses statistics that show the accuracy of predictions for both hedonic and indexed models over time and gives each value a weight according to the provider’s proprietary formula.

The problem is that these usually assume average condition of the property without allowing for upgrades or the value of those upgrades vs. comparable properties. Each AVM can come up with a completely different value. For example – $134,000 for Trulia, $141,000 for Chase Home Value, $138,470 for AOL Real Estate, and $136,519 for Zillow. Will the real value please stand up?

Realtor Property Resource ( that can be used by any REALTOR as part of your NAR dues, that produces an RVM (Realtor Valuation Model) which is a nationwide parcel centric database with over 146 million property records. It is using the Blended approach of not only using the public records but is also adding in MLS data of current and off-market information into a proprietary algorithim that produces the most reliable valuation product available. This system is owned by NAR… which is us.

So perhaps you are misunderstanding. No one is talking about a valuation model that uses only MLS data… and is not also blending in the public records too. This blend that other AVM systems have not been able to have access to – our active, pending and sold MLS data – gives an even more complete picture as to the value of the property in question. This MLS data is coming into the RPR system virtually in “real time”.

Please feel free to contact me if you have more questions as I have been teaching RPR classes for the past 2 years for NAR. I have probably given over 125 “live” classes in that time.



********************** Win Singleton, CRB, SRS, SFR, e-PRO Associate Broker Long & Foster Real Estate, Inc. (703) 536-7631 Licensed in Virginia **********************


From Mark Jay:

“Steve Ervin asks:

Is there any REAL value to using the MLS data to generate an AVM?

More and more listings are being sold OUTSIDE of the MLS systems. .we have been informed that slightly over 43% of transactions closed in 2013 didn’t pass through a Multiple Listing Service.”

I would think that a data source (the MLS) that is missing that much data would be a pretty unreliable foundation for any Automated Valuation Model system.

Mark Jay comments:

Okay, if 43% of transactions are outside of MLS that means that 57% are marketed and reported INSIDE MLS.

Let’s say that there are 10,000 sales within a geographic area serviced by an MLS. That would be 5,700 sales reported in MLS. That’s a pretty large sample set upon which to perform a statistical analysis upon, don’t you think? Sure, it would be nice to have ALL the population data (all the ‘arms-length’ transactions) within an area but if you don’t, taking-statistically speaking-a huge sample size is essentially, just as good.

Now, it might be that the 4,300 non-MLS sales differ in some way from the 5,700 MLS sales, but it’s likely that difference, if significant, would be systematic and therefore able to be accounted for as another variable, wouldn’t it?

What makes the MLS data amenable to use in statistical modeling is that property data is the ease of tabulation through the RETS standards for MLS data required some time ago now by the NAR promulgated MLS Model Standards. The most important RETS data can be used to predict sales prices and that’s the objective of AVM models. An entity can simply do the work of analyzing the non-MLS sales (a statistically significant sample size) and develop an adjustment factor. Having a sample size as large as 57% of the population universe should be no impediment to having an accurate sales price prediction engine (AVM)

I’ll go on a little farther..

The idea behind AVM is for “larger” MLS participants to be able to develop an income stream ancillary to providing traditional Real Estate Brokerage Services by satisfying the demand lenders (and other users of this sort of information) have for obtaining more objective and reliable information on property values as mortgage loan security (collateral) than appraisers can comparatively provide. MLS entities themselves could provide AVM services but that would be clearly outside of the Core Mission and I doubt the membership would allow it, because of that and other reasons. Large, market dominating, Real Estate Brokerage Services providers (Big multi-office Brokers with 20 to 30% market share) want to add a revenue stream and can do that with just the “slightest” permission from NAR and NAR model MLS entities. In fact. a large broker with a 20 to 25% share wouldn’t even need to use the entire MLS data base to generate an AVM that a large lender could use in lieu of an appraisal, it could be argued, from a statistical point of view. The largest and the second largest broker in my market could generate a statistically robust AVM just from their internal sales. (and remember, this large broker COULD-if they had them-add back in “pocket listings” All that’s need is a large enough population sample size. Remember, too, all the lender really wants is something MORE reliable and less costly than an appraisal. The AVM data source doesn’t have to be “perfect”, as in containing ALL the sales data, it just has to be better IN RESULTS than than the predictive ability of the appraisal process in terms of risk management and cost.

The sad reality for appraisers vis a vis one through four family valuation is that the way they do what they do has and for many years has been at the point of “technical exhaustion”. Appraisers can be easily “bought” in a way that a mathematical model can never be and the cost, because of the labor component, is just TOO high. What has happened to Mortgage Loan Underwriters will soon happen to appraisers. In mortgage loan underwriting, the originator enters a number of variables into a computer program and the program approves the loan. The information entered is verified for accuracy by what is called a “validator” and the loan is closed. As long as the income, assets, and the property value entered into the underwriting engine (a computer program) is verified or validated by a person with a skill set up to the task of comparing numbers on a form, then a highly trained and deeply experienced underwriter is not necessary except in the increasing rare case of a loan that has to be “manually” underwritten. In the future a “validator” will simply order an AVM product, look at that number to make sure it supports the number in the Automated Underwriting Engine and the loan will close. The AVM number will be ordered online at a cost of a few dollars-say $20 or $50 or so. competition will bring that number down-and received in a matter of a minute or so and the loan will close. There will be no need for an appraiser to “go out”, look at the property, take pictures-front, back, street scene, etc.-and then submit a report 10 days to 2 weeks later.

The short version is that AVM WILL happen and a full record of every sale in a market area isn’t needed, even the internal data base of a large broker should work. AVM WILL work because the lenders want it. The only little thing that is needed is permission to use the MLS data base-even if it’s to only “fuel” a proprietary AVM with the Broker’s own sales data…


Mark Jay comments:

The hardest part of selling is generating REAL qualified prospects for your good or service.

Steve Ervin replies:

Mark…I completely agree with your comment and the rest of what you wrote here. the basic problem can be understood by the data in NAR’s annual survey of home buyers and sellers. The FIRST thing to look at is that NAR says that on average people start looking on the internet 18 to 24 months BEFORE doing a transaction. The study goes on to say that home buyers START working with a Realtor on average only 12 weeks before doing a transaction. But here is the important fact appended to that…the home buyers say that they started SEARCHING for a home to buy ONLY 3 weeks earlier. In other words almost ALL of the time prospects spend on the internet they simply are NOT prospects.

I totally agree with you that most lead companies selling leads to Realtors have absolutely no interest in the QUALITY of leads they generate. for years I have been working on a solution to this. And I think I have come up with on that will work. Instead of finding ways of tricking consumers into filling in forms so that information can be sold as a crap lead…I am building a system that is designed to HELP Realtors generate leads from their listings. The system is described here …but just the technology to generate more leads does not solve the problem. The leads still needed to be screened to determine if they were actually of any value. So I have been putting together a service to follow up on all the leads generated…..and instead of delivering RAW contact data, I plan to deliver profiles of each prospect identifying why they made an inquiry and where they are in the buying process.

I am planning to offer these services to Realtors free of charge…including providing the rider signs, etc. There will not be any “freemium” offers where the base level is free…but enhancements ( enhanced listings for example) are available at an additional charge. The entire marketing system and follow up service will be provided at NO charge. What I am asking for in return is that I can sell these leads to a lender or some other NON-REALTOR service providers that provide products and services to home buyers. Unlike Zillow,, etc. the leads generated by a Realtor’s listings will never be shared or sold to another Realtor. And of course we will not put any restrictions on the Realtor providing those leads to their preferred mortgage supplier or other service providers they feel can help their client.

I would really line to talk with any agents and brokers out there who can give me feed back and advice on what works for them and what does not…before I do the official launch of the system in the next couple months.

Steve Ervin 727-320-5436 Direct




Submit a Comment

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.