There will soon be no need for residential appraisers.
I remember years ago, doing education for my merging AIREA/SREA chapter(s) in San Diego. AVMs were just coming to our attention. I invited a MAI who happened to also be a PhD in economics, Phil Mitchell. The room was full.
Someone asked about those “automated algorithmic computer calculated appraisals.” Could a canned program do as well as an appraiser? Without hesitation, he answered: “Not only will these models be able to appraise – they will do it better!” That was 24 years ago. He was right, partially.
I believed then that certain things could be better done by a computer. And still do. Yes, certain things are done very well by a computer, with the right data. Other things are done better by a human. Particularly by a human who knows the subject-matter.
It is true, AVMs have taken a large market share. However, they have not replaced the appraiser. Why are we still in this middle mushy area? Why no winner after all these years?
Let’s clarify a couple of issues. An appraisal is defined as a valuation done by an appraiser. “AVM” is actually an industry, not a type of model. Each competing AVM company has its own model. Different models do better in different places. And different AVMs have better or worse ‘coverage.’ (How often they can provide a number.) And each company keeps its magic formula secret. Secret. Else another company will steal its algorithm.
AVMs are reaching a point where improvements in algorithms bring little or no improvement in accuracy or precision (trueness or sureness). In fact, the issue now is that there is no consistent way to measure trueness and sureness! (I am currently a member of the AVM Standards Task Force, sponsored under the “Congressionally Authorized” Appraisal Foundation.)
Appraisers also have reached a point of little or no improvement in accuracy or precision. The issue here: there is really no way to measure trueness or sureness. The standard under USPAP (Uniform Standards of Professional Appraisal Practice) is “belief.” The test in USPAP is “credibility” or “worthy of belief.” How do you measure believability or worthiness? An appraisal is an opinion. This is the problem. Clients, including Fannie Mae, Freddie Mac need quantitative results, not opinion.
Neither AVMs nor appraisals (nor any other ‘method’) appear sufficiently reliable, given cost and speed. One reason, we are approaching the limits of unenhanced data. “Adjustment support” and algorithms can only work with the data available.
AVMs are stuck with whatever data can be obtained online immediately. (This includes so-called “non-traditional” variables.) What is missing is twofold: 1) the quantification (or lack) of ‘order’ (nonparametric) variables which have no measurable quantity or category. My expert witness specialty is view value, which can be the greatest value element in some areas, and; 2) certain modeling decisions which depend on human-unique abilities to generalize and identify outliers and fit curves.
Appraisers, residential and non-residential, are stuck with legacy practices, forms, and cells in spreadsheets — which depend on judgment instead of modern models and analytic algorithms in data selection. As we are taught — “picking comps” is circular. A good comp is competitive. A competitive sale is similar. And then – a similar sale is “one that can be compared to the subject! The problem here is that when you start with subjective judgment . . . you cannot achieve an objective end. Clients are concerned with one of three things: collateral risk, investment potential, or equity enforcement. Everyone has an opinion. What clients need is measurable analytics, risk scores, and other products and services, which can be prepared by appraisers or qualified asset analysts.
Clients are concerned with one of three things: collateral risk, investment potential, or equity enforcement.
The problem is in the data. The solution is in the data. But it is not a quick fix. Getting the data right means cleaning, scrubbing, and organizing the data so it becomes useful information for good decisions.
We will come back with a Part II to this question: “Is Residential Appraisal Finished?” In TAAR (The Asset Analyst Report) we consider more specific solutions for our subscribers. We are currently developing high-quality on-line education on this topic. Valuemetrics.info
May 27, 2020 @ 12:43 pm
You’ve drilled down to the essential issues, as you usually do (although we might disagree on whether an appraisal opinion is inherently bad). I think one problem is that we’ve let the form and format of the appraisal report take over the valuation process; Scope of Work becomes a mere afterthought. A shameless plug for my most recent blog post that discusses this issue:
https://coastlinerealtyadvisors.com/has-the-appraisal-been-hijacked-by-the-report/
By the way, I remember Phil Mitchell, who passed away many years ago . . an excellent and thoughtful practitioner.
May 27, 2020 @ 1:33 pm
Mr. Dell, for the most part I agree but I think what will really do appraisers in will be AI and the growth of Big Data. AVMs have always been as suspect as the data fed to them. That is changing with the UCDP and we appraisers are helping. Most every time our appraisals are uploaded for conventional financing, residential property data for the subject and comps is cataloged. The property characteristics are honed over time with each entry verifying once and for all each property’s physical characteristics. Unlike MLS data, all the typos and incorrect room counts and the like will be removed from the data set. Add to this all the 3D modeling from Matterport cameras and soon more precise categorization of quality will be included as well.
AVMs are only as good as their programmers and the data they use. AI on the other hand will continuously improve its own algorithms, testing and re-testing the models constantly. AVMs only recognize the outliers they are programmed to identify. AI learns to identify them on its own.
AI, AVMs and Big Data will one day take a large share of the valuation market, particularly in urban and suburban markets where data is plentiful and uniformity of properties exists. However, they will never replace appraisers entirely. Appraisers will still have a place in rural markets where data is scarce and dissimilar and where many property characteristics cannot be mathematically quantified. We will be needed in-town for unusual or oddball properties, even for relatively mundane aspects like a garage conversion. Unfortunately, all the cookie-cutter work that is most appraiser’s bread and butter will be taken by automation leaving us with increasingly complex assignments.
May 28, 2020 @ 6:28 am
Excellent post, George. You have distilled the issue in a nutshell.
“The issue here: there is really no way to measure trueness or sureness.”
However, there is a way to measure, or at least opine, on “reliability.” Valid data correctly analyzed leads to more reliable results. I am not saying that reliability equals accuracy. I suspect that these terms are not defined in USPAP for a reason… the standard is for credibility. Two appraisers could analyze the same data, both with technically correct analysis, and still come to two different opinions of value. Both would be credible if the data is the same and no errors are made in analysis. Neither would necessarily be accurate or inaccurate… but, one might be more reliable if that analysis had more depth.
If best available data is somewhat sketchy, the (still credible) opinion of value might be somewhat less reliable. Appraiser’s should own up to this and describe the data and why it is not perfect. That is something AVMs cannot do, at least not yet, and IMHO, gives us a competitive advantage.
One additional comment. The discussion in your article appears to be about value or market value, or conversely, AVM results. There is no way to measure trueness or sureness for opinion of market value, you can only analyze data and method. However, there is a kind of measure of trueness when the appraisal is for Anticipated Sale Price. RELO companies grade appraisers on whether the property sells within time frame and how far from ASP the eventual selling price is. That grading is not necessarily valid for an individual assignment, but when all assignments are analyzed it can help determine appraisers that are more “true.” I caution appraisers and others not to attempt to apply the same kind of grading to appraisals for market value. There are too many factors that could impact eventual price of the next sale to make such a grading valid.
May 28, 2020 @ 12:14 pm
Hmmm. So I write my own software and now have my own “model”. The model is not reproducible and I don’t have to show anybody what data was used. It’s a secret. In my model, I just used Zillow and cross reference with county property records. If I want to. Wow! and then you have the BALLS to say it’s an appraisal. NOT. It’s a property “score” based on Magic. Anybody can come up with a property “score”, the property owner, broker, AVM, etc. It’s cheap and fast. The AVM can not show you current photos of the interior and exterior and can not give you a current value. For loan to value ratios that are low, an AVM would be fine. But as an investor that purchases mortgages, I don’t want a “maybe” score. I want to know the condition of the interior and exterior as of the date of the appraisal, not when it sold 3 years ago. As an investor, I want to know my money is safe. Earthquake or fire, the house burns down. There were no safety straps on the water heater, no smoke alarms, and everyone died. It turned out that the property was a trashed out rental, or the owners were using it as a “grow house”, or even a meth house. Either way, as an investor, I just lost my money!
May 28, 2020 @ 5:46 pm
George is on track with his program. If the Residential Appraiser goes away it will be their own fault for not embracing what George is trying to teach and adopting. AVMs are statistics. Most statistics were developed for Gaming/Odds applications in early 1800s evolving to science in the 1,870 to 1,900 era. AVMs do not produce a believable report just a probability and error rate they call an FSD. Which is just the standard deviation expressed as a probability. Statistics can give the probability of a value being believable. What they can’t do is tell you when it is believable. Just like statistics can provide the probability of getting heads or tails when flipping a coin. But it can’t tell when heads or tails will be Flipped. That is still a random event. AVMs provide a random data point, believability is a human subjective interpretation. Statistics do not provide believability, they are just a random data point at a point in time with a probability of being True. No better than flipping a coin.
February 18, 2021 @ 11:12 am
An important factor that is not mentioned is that users are able to insure for the variance with AVM models, as they can do for the appraiser. That puts the two products on an even playing field with regard to risk, so it is the fastest and lowest price provider that the free market will reward given the equal risk of both products. The only future growth in residential practice will be in litigation and other non-lending appraisal work.
No Time for Time Adjustment? - George Dell, SRA, MAI, ASA, CRE
April 12, 2023 @ 9:56 am
[…] am told that about half of form residential appraisal reports had “zero” change in prices for market conditions, while at the same time […]
December 25, 2023 @ 10:02 am
Real estate is an imperfect market which can only be analyzed by humans. Every real estate transaction has an emotional characteristic that must be addressed as a sale condition adjustment to estimate the typical buyer and seller as defined by market value. AVMs will not be able to estimate market value as defined.