Technology in appraisals AO 41

[Fernando]

Grok: In the context of the Uniform Standards of Professional Appraisal Practice (USPAP), the definition of "appraiser’s peers" focuses on appraisers who possess expertise and competency in a similar type of assignment. This criterion does not impose restrictions based on the relative level of experience or expertise among peers. Consequently, an individual with substantially greater experience and expertise could qualify as a peer, provided they demonstrate the requisite competency in assignments of a comparable nature. The emphasis in USPAP is on the alignment of professional capabilities with the specific assignment type, rather than equivalence in career tenure or proficiency levels.

I have done all kinds of residential and commercial appraisals throughout Northern Calfornia. And I only do FULL APPRAISALS. No driveby's or desktops.

So, yes, I am one of your "peers", per USPAP.

You keep screwing up Fern. You're reckless as hell.

You use this context "... when I am your "peer" using MARS regression, SKATER or rgeoda for defining neighborhoods, ...". I never heard of SKATER or rgeoda.
So no, our peers do not use what you are using. You're appraising in your own bubble world.

 

[RCA]

You use this context "... when I am your "peer" using MARS regression, SKATER or rgeoda for defining neighborhoods, ...". I never heard of SKATER or rgeoda.
So no, our peers do not use what you are using. You're appraising in your own bubble world.

As is so often the case, you are not making sense. You dropped off the first part of the sentence. "Then, WTF are you going to do when I am your "peer" .... " OF COURSE we all know you have never heard of these popular R cluster analysis techniques, more advanced than K-Means Cluster analysis. That is the point - you like to brag that what you do is what your peers do, and that is not the case. Most of us don't have a clue what other appraisers actually do. We don't have every appraiser in the US on this forum. Only a VERY small percentage. But, based on our random bump-ins with other appraisers and the number of dead-boring Appraisal Conferences we have attended, we could certainly conjecture [ notwithstanding George Dell's ify classes on R and statistics ] that there aren't, percentage-wise, many appraisers with advanced knowledge of R and statistics. But we can't know for sure. The use of the term "peer" is an EXCUSE by the people who write professional standards to be sloppy. Everyone knows it is an imprecise term, a catchall for what the incompetent standards writers can't express, and if they could express what they meant, most likely would be unable to agree upon.

 

[Fernando]

As is so often the case, you are not making sense. You dropped off the first part of the sentence. "Then, WTF are you going to do when I am your "peer" .... " OF COURSE we all know you have never heard of these popular R cluster analysis techniques, more advanced than K-Means Cluster analysis. That is the point - you like to brag that what you do is what your peers do, and that is not the case. Most of us don't have a clue what other appraisers actually do. We don't have every appraiser in the US on this forum. Only a VERY small percentage. But, based on our random bump-ins with other appraisers and the number of dead-boring Appraisal Conferences we have attended, we could certainly conjecture [ notwithstanding George Dell's ify classes on R and statistics ] that there aren't, percentage-wise, many appraisers with advanced knowledge of R and statistics. But we can't know for sure. The use of the term "peer" is an EXCUSE by the people who write professional standards to be sloppy. Everyone knows it is an imprecise term, a catchall for what the incompetent standards writers can't express, and if they could express what they meant, most likely would be unable to agree upon.

Majority of appraisals do not use much technical mathematical concepts in their daily practice. Those are our peers.
I'd taken many advanced college math classes & did very well, and I stopped at Linear Algebra because I had no need for further math in my degree field.
USPAP does not expect us to know advanced math concepts and no need when using 1004 form.
I'd given up with advanced math modeling in doing appraisals because my peers don't use it (I had more advanced math skills in the past than my peers when younger).

I'd attended many AI appraisal conferences when I was younger, and the attendees are more of commercial and assessor appraisers.
The small time independent fee appraisers do not attend those "snotty" conferences. Those male attendees are dressed in suit and ties.
I'm one of the few who dresses casually there which I would normally wear when going to people's homes.
Again, RCA, you are not a typical "peer" in doing GSE appraisals.

 

[RCA]

Majority of appraisals do not use much technical mathematical concepts in their daily practice. Those are our peers.
I'd taken many advanced college math classes & did very well, and I stopped at Linear Algebra because I had no need for further math in my degree field.
USPAP does not expect us to know advanced math concepts and no need when using 1004 form.
I'd given up with advanced math modeling in doing appraisals because my peers don't use it (I had more advanced math skills in the past than my peers when younger).

I'd attended many AI appraisal conferences when I was younger, and the attendees are more of commercial and assessor appraisers.
The small time independent fee appraisers do not attend those "snotty" conferences. Those male attendees are dressed in suit and ties.
I'm one of the few who dresses casually there which I would normally wear when going to people's homes.
Again, RCA, you are not a typical "peer" in doing GSE appraisals.

There is nothing at all in USPAP about "typical peer". Whether I am a "typical peer" is actually immaterial.

What is material is whether you are competent to provide a reasonably accurate estimate of Market Value. The typical expectations are within +/- 5% of the true Market Value. Market Value is typically defined fairly accurately itself. If you are thrown into the midst of residential properties in most areas of the SF Bay Area, and you have to compare the subject to other properties, none of which share more than 80% of the important features of the subject property, objective adjustments will have to be made to 10 + features, many of which vary in a non-linear fashion with the important features. In fact, they may vary with 2-3 variables at a time, aka "interactions". If you use Multi-Linear regression, you very likely are not better than an R2 of 0.40, that is to say you will not be able to account for more than 40% of the variation in the price of properties in the subject market area through your "model." You also well be left with a residual that is worthless - that you can do anything with. Another "competent" appraiser who uses MARS regression, can get the the R2 up to 80%, which also gives a residual usuable for further adjusting the other 20% of value,i.e. the unmeasured variables such as condition, quality of construction, functional utility and so on - and do it with constraints that make the residual side of adjustments perfectly accurate in terms of the impact on the value conclusion.

So, understandably, one appraiser using MARS and RCA can, in such cases, provide an accuracy within +/-2%. - And you are really talking about an objective conclusion that has perhaps only 40% accuracy. OF COURSE, what you will do in such a situation is resort to subjective comparisons and simplistic methods that provide only very rough and risky estimates of adjustments. The difference is a matter of luck, of how good you are at guessing, plus a certain amount of luck. Bad things can happen.

The MARS appraiser comes along and can say that such-and-such totally objective methods yielded such-and-such values until he gets within +/- 2% of market value. He can actually construct a very good logical argument for every adjustment made.

You made a mistake not taking Linear Algebra. Until you learn Linear Algebra, you are kind of wasting your time studying math. AI LLMs are based on Linear Algebra and partial differential equations - mostly. A lot of the statistical methods we use are also based on the same.
=====

You can present all the excuses in the world for not using MARS. At the end of the day, people will think what they will. Either you get good results or not. And either you convince people that you're biased or not. That's between you and your clients and respective homeowners.

 

[Fernando]

There is nothing at all in USPAP about "typical peer". Whether I am a "typical peer" is actually immaterial.

What is material is whether you are competent to provide a reasonably accurate estimate of Market Value. The typical expectations are within +/- 5% of the true Market Value. Market Value is typically defined fairly accurately itself. If you are thrown into the midst of residential properties in most areas of the SF Bay Area, and you have to compare the subject to other properties, none of which share more than 80% of the important features of the subject property, objective adjustments will have to be made to 10 + features, many of which vary in a non-linear fashion with the important features. In fact, they may vary with 2-3 variables at a time, aka "interactions". If you use Multi-Linear regression, you very likely are not better than an R2 of 0.40, that is to say you will not be able to account for more than 40% of the variation in the price of properties in the subject market area through your "model." You also well be left with a residual that is worthless - that you can do anything with. Another "competent" appraiser who uses MARS regression, can get the the R2 up to 80%, which also gives a residual usuable for further adjusting the other 20% of value,i.e. the unmeasured variables such as condition, quality of construction, functional utility and so on - and do it with constraints that make the residual side of adjustments perfectly accurate in terms of the impact on the value conclusion.

So, understandably, one appraiser using MARS and RCA can, in such cases, provide an accuracy within +/-2%. - And you are really talking about an objective conclusion that has perhaps only 40% accuracy. OF COURSE, what you will do in such a situation is resort to subjective comparisons and simplistic methods that provide only very rough and risky estimates of adjustments. The difference is a matter of luck, of how good you are at guessing, plus a certain amount of luck. Bad things can happen.

The MARS appraiser comes along and can say that such-and-such totally objective methods yielded such-and-such values until he gets within +/- 2% of market value. He can actually construct a very good logical argument for every adjustment made.

You made a mistake not taking Linear Algebra. Until you learn Linear Algebra, you are kind of wasting your time studying math. AI LLMs are based on Linear Algebra and partial differential equations - mostly. A lot of the statistical methods we use are also based on the same.
=====

I did take Linear Algebra. I remember it wasn't that hard. I should have taking more advanced statistics classes which would have been more handy in appraising. Still, fundamentals of statistics have helped me in knowing what is important or not in analyzing data.

You can present all the excuses in the world for not using MARS. At the end of the day, people will think what they will. Either you get good results or not. And either you convince people that you're biased or not. That's between you and your clients and respective homeowners.

You and I believe we are better than our peers. Lenders are lucky to have us as their appraisers.
You use mathematical models to narrow down the most likely market value.
I use my experience and intuition which is faster and just as accurate as your impractical computer models at this time.

 

[Fernando]

Speaking of computer models, I check on Zillow which uses their proprietary algorithmic valuation or Zestimate.
Many time it's not as accurate but within range because of bad input information from subject's characteristics and bad comps.
Zillow is $16 billion company and has hundreds of RCAs working together and having a much better computer model.
Every month about 243 million users go to its portal, thus I can't ignore Zillow in how they "create" valuation perception for the public.
Thus I use technology and my experience and intuition which makes me better appraiser than any existing computer model out there now.

 

[Terrel L. Shields]

Flood maps.

Flood maps can be wrong for a number of reasons. I personally sit on a hill waiting for water to recede in June of 1974 - and it filled the valley before me. The maps to this day only show about half the area that I KNOW was covered in water then. The magnitude of that rain has only been repeated twice in historical records. 1927 when the Mississippi River Basin was flooded leading to the creation of the TVA as well as flood control dams along it, and in 1892 when it flooded many small towns in this area. that's roughly 50 years apart (we're due another by that metric) but the 100-year flood map is not even covering what I saw with my own eyes - And at the time, I had been working with flood routing, so I understood what I was seeing.

BTW, I don't depend upon Realist for much of anything.

 

[Fernando]

Flood maps can be wrong for a number of reasons. I personally sit on a hill waiting for water to recede in June of 1974 - and it filled the valley before me. The maps to this day only show about half the area that I KNOW was covered in water then. The magnitude of that rain has only been repeated twice in historical records. 1927 when the Mississippi River Basin was flooded leading to the creation of the TVA as well as flood control dams along it, and in 1892 when it flooded many small towns in this area. that's roughly 50 years apart (we're due another by that metric) but the 100-year flood map is not even covering what I saw with my own eyes - And at the time, I had been working with flood routing, so I understood what I was seeing.

BTW, I don't depend upon Realist for much of anything.

I kept up with the flooding and landslides in my territory over past decades so I know exact locations where it happened. Flood maps can only know general expectation of flooding.
Experienced appraisers still know best. Hear that GSEs.

 

[Vermonter]

Experienced appraisers still know best. Hear that GSEs.

Smart. Summon the GSEs to read your posts and we'll all be screwed

 

[RCA]

Speaking of computer models, I check on Zillow which uses their proprietary algorithmic valuation or Zestimate.
Many time it's not as accurate but within range because of bad input information from subject's characteristics and bad comps.
Zillow is $16 billion company and has hundreds of RCAs working together and having a much better computer model.
Every month about 243 million users go to its portal, thus I can't ignore Zillow in how they "create" valuation perception for the public.
Thus I use technology and my experience and intuition which makes me better appraiser than any existing computer model out there now.

God, do you speak nonsense! Zillow does not use RCAs, and they don't even have a "model"! They use a black-box approach that effectively assumes what they don't know is just the average. That is, if they don't know the condition of a home, which they often don't, then they assume it is the average for that property type in its neighborhood. They can't be that accurate, simply because they don't have inspection-level information on most of the homes they value.

In my opinion, you are not much better than the black box, at best. You operate on seat-of-the-pants judgments, which invariably use your subjective observations and patterns you have subjectively picked up over the years. You have nothing much more than a rule that "The price must be somewhere in the neighborhood of homes sold in the past year that are relatively similar to the subject. You fidget around with "matched pairs" to get some dicey adjustments and, in the end, make sure that value looks like it is in the right ballpark. You make many assumptions that, if investigated, might not pan out. I can see where you would easily get in troiuble.

Of course, I only know what I see on this forum. Maybe you are just being sloppy on the forum. I don't know. But from appearances on this forum, I'd be more cautious if I were you.

As far as Zillow's accuracy. Well, let's compare Redfin and Zillow. Grok says they both have about the same accuracy:

Grok: "Based on the most recent official data published by each company, Zillow's Zestimate shows slightly higher accuracy in home price predictions than Redfin's Estimate. This conclusion is drawn from nationwide median error rates, which measure the percentage difference between the estimated value and the actual sale price.

Zillow reports a median error rate of 1.83% for homes actively listed on the market and 7.01% for off-market properties. In contrast, Redfin indicates a median error rate of 2.00% for on-market homes and 7.69% for off-market homes. These figures suggest that Zillow's predictions are marginally closer to actual sale prices, particularly for properties not currently listed.

It is important to note that both tools rely on algorithms that incorporate public data, such as property details, recent sales, and market trends. However, their accuracy can vary by location, property type, and data availability, and neither should substitute for a professional appraisal. Historical comparisons from earlier years sometimes favored Redfin, but current statistics indicate Zillow's edge in precision."

That makes sense, because according to my information, they both use pretty much the same statistical methods (The differences they have are usually in the data. One may have information from refinance appraisals, another not. Major upgrades usually only get noticed if and when there has been a sale. And, the local planning and/or building departments don't always report changes. .... )

Now, let me give you an example of a house I am familiar with as a comp or subject: 120 Hillcrest, Berkeley, CA.

Currently
Zillow Estimate: $6,649,400
Redfin Estimate: $4,447,552

Zillow is 50% higher than Redfin. That is as of 1/6/2026.

Fern, from my perspective - you are nothing but a Nut Case. You don't know ****. You don't really know you're own **** - because you are so messed up in your own logic and knowledge.