Appraisal Statistics

[Deleted member 130081]

Sigh. I can look at a property and after 5-15 minutes of research of sales, on 95% or better of instances, know what the property is likely worth to the same degree of accuracy, and usually better, than any model out there. How can I do that? It's two things. #1 Experience. #2 My brain, and all brains, are super-computers. If only my brain had a print button. Alas it does not, and those that seek "proof", turn to any method possible to produce said "proof", including computer data models. While I get it, I really do, an appraisal really does need to be supported, to weed out the people who are dishonest or who might be short on internal wiring however, this quest the world is on is not the most feasible way to skin the chicken. Then again, I don't have all the answers either.

We are not Rocket Scientists, nor is anyone interested in paying us a Rocket Scientist salary to get an appraisal done. At some point, it would be nice for all involved to come back to Earth and eat some reality pie.

 

[Meandering]

And here it is,

Read it and giggle, or weep.

…Battles over alleged misappropriation of trade secrets relating to residential valuation technology and services…

And this is why you can't discuss your methodology over the internet or in detail in a report.



.

 

[RCA]

Sigh. I can look at a property and after 5-15 minutes of research of sales, on 95% or better of instances, know what the property is likely worth to the same degree of accuracy, and usually better, than any model out there. How can I do that? It's two things. #1 Experience. #2 My brain, and all brains, are super-computers. If only my brain had a print button. Alas it does not, and those that seek "proof", turn to any method possible to produce said "proof", including computer data models. While I get it, I really do, an appraisal really does need to be supported, to weed out the people who are dishonest or who might be short on internal wiring however, this quest the world is on is not the most feasible way to skin the chicken. Then again, I don't have all the answers either.

We are not Rocket Scientists, nor is anyone interested in paying us a Rocket Scientist salary to get an appraisal done. At some point, it would be nice for all involved to come back to Earth and eat some reality pie.

I hear this from appraisers from time to time, going back many years. The central question is whether you can explain how you came to your opinion of value. If you can do a good job of that, then I am pretty sure I can build a model based on that. And if I can create a model, there is a pretty good chance that a statistical program out there can derive a close look alike - good enough that if I take such a model and give it to a novice and have them compare it to sales with photos on the subject MLS, they can probably develop the same kind of proficiency that you have in a much shorter period of time. And that is part of the rationale of using statistics, it allows you to jump into an unfamiliar neighborhood and more quickly pick up expertise in that area.

I had, not to mention any names, a senior appraiser, sternly tell me not to ever go into Monterey County. I took it at the time to be a pure threat, and thought he probably had some loose agreement down in that area (he was a commercial appraisal reviewer and had all kinds of agreements not to appraise in certain areas for for particular property types because of conflict of interest issues). I did them anyway. Salinas, Monterey and Pacific Grove were the principal areas. Carmel and Carmel Valley I never got until one day I got a call. Now, these two areas are very challenging. But the Carmel one was particularly challenging. I did know an appraiser down there; and gave him a call. Surprisingly he didn't have much advice. I had been given a previous appraisal on the same property that was rather unimpressive. I was confident I could do a better job. In these types of things, I have absolutely nothing against approaching agents at other open houses and asking questions. In really difficult areas, I will drive around and see if there are any people standing out front watering their lawn or something similar. I park the car and walk down the road and chat with them to get info about the neighborhood. Older people can be real interesting to talk to. You can get all kinds of information (I could tell you some stories).

But I also use MARS to run regressions on different groups of properties - and this really forms the backbone of my opinions.

Now as it turns out, there was a civil suit involving the lender, so all the 7 appraisers (yes they wound up having 7 appraisals) were investigated by the board, as per their policy when there is a civil suit. So, well, between what my client told me (and the documents that had come into his hands as well), what I picked up from other sources, it appears of course they compared all appraisals to each other, and the fact that I accurately measure GLA per the ANSI standard and most of the other appraisers somehow independently arrived at the assessors quite incorrect value ... must have caused some problems. Also, I have a much different way of dealing with the Cost Approach that I have found more accurate than using the usual Marshall & Swift and land/improvement ratios.

- The point is, I can't do this sort of thing without regression. And I doubt anybody can. That is not to say a good regression tool can allow you to jump into any market and property type and compete with a local appraiser - who may simply have data that no one else has access to. But, I tend to think in many cases, not all, that the expertise of "local" appraisers is way overvalued. Yes there can be highly qualified local appraisers - but they are often surrounded by hordes of far less qualified appraisers (many of whom have college degrees - even in real estate). And if you don't know as a fact that your "peer" appraisers in an area are distinctly more qualified than you in appraising property on their turf, don't assume that is the case! There is invariably a wide range of competence levels - and paper qualifications are no guarantee of competence.

 

[Terrel L. Shields]

he central question is whether you can explain how you came to your opinion of value. If you can do a good job of that, then I am pretty sure I can build a model based on that.

Models vary by modeler. That's why there are 3000 global warming models but zero that accurately estimate the existing historical climate record when run backward.

A $100,000 pool becomes invisible to regression if you have no comp with similar pools. $15,000 pools don't count. A cost related adjustment applying functional obsolescence as % might be better. I like to eliminate adjustments using regression to determine when something isn't driving value.

 

[RCA]

Models vary by modeler. That's why there are 3000 global warming models but zero that accurately estimate the existing historical climate record when run backward.

A $100,000 pool becomes invisible to regression if you have no comp with similar pools. $15,000 pools don't count. A cost related adjustment applying functional obsolescence as % might be better. I like to eliminate adjustments using regression to determine when something isn't driving value.

A lot of appraisers think: Regression is supposed to provide a perfect answer or it isn't any good at all. Or how about this variation: You can't alter the model that a regression method gives you? That is why you are not supposed to use these methods unless you really understand them -- and the first step is understanding that they are good at modeling ONLY the data they are given - but do not provide direct help in valuing data they are not given. So if you give your software 30 comps to look at and only one has a swimming pool, there is nothing it can do with it and not overfit.

What you do is use the regression model to calculate the "contribution" of the variables it can deal with (those input and with sufficient non-empty values) and you, the appraiser, then add on the contribution of the others it knows nothing about. So, typically with something like MARS it is going to give you extremely good estimates of the contribution to value from its domain variables, typically GLA, Lot Size, Bath Count, ..., but it will not be able to tell you how much Condition, Quality, View and other features that are typically not in the MLS quantitative data for comps. If you know your neighborhood, then you should be able to rate your subject in comparison to the comps in respect to Condition, Quality, View, and other unusual features - and know the market reaction tot he difference. In fact regression can help you in that respect as well --- if you know he ins and outs of using it.

Well - of critical importance on that last point, is that regression will give you a value for the comp that assumes that the unknown variables all have average values. So, if the comp is in fact above or below average on those unknown variables, then you make an adjustment for the difference. For example, if the average condition of the comps is C-4, it will assume your each comp is C-4. Same for Quality and any other unknown values. If you take the model produced for the comps and run it against all comps, and then take the difference between that estimate and their actual price and then rank the differences and look at the range, you can get an idea of what kind of adjustment needs to be made for all unknown variables combined. In fact, an experienced appraiser will usually be able rate any property as being at the Nth percentile in the combined features - and you can then in fact model that with an equation and get a pretty exact value. -- Of course assuming you really know your neighborhood so well, that you can pigeonhole your comps as being better than a certain percentage of the other comps with respect to quality, condition, .....

 

[Deleted member 130081]

A lot of appraisers think: Regression is supposed to provide a perfect answer or it isn't any good at all. Or how about this variation: You can't alter the model that a regression method gives you? That is why you are not supposed to use these methods unless you really understand them -- and the first step is understanding that they are good at modeling ONLY the data they are given - but do not provide direct help in valuing data they are not given. So if you give your software 30 comps to look at and only one has a swimming pool, there is nothing it can do with it and not overfit.

What you do is use the regression model to calculate the "contribution" of the variables it can deal with (those input and with sufficient non-empty values) and you, the appraiser, then add on the contribution of the others it knows nothing about. So, typically with something like MARS it is going to give you extremely good estimates of the contribution to value from its domain variables, typically GLA, Lot Size, Bath Count, ..., but it will not be able to tell you how much Condition, Quality, View and other features that are typically not in the MLS quantitative data for comps. If you know your neighborhood, then you should be able to rate your subject in comparison to the comps in respect to Condition, Quality, View, and other unusual features - and know the market reaction tot he difference. In fact regression can help you in that respect as well --- if you know he ins and outs of using it.

Well - of critical importance on that last point, is that regression will give you a value for the comp that assumes that the unknown variables all have average values. So, if the comp is in fact above or below average on those unknown variables, then you make an adjustment for the difference. For example, if the average condition of the comps is C-4, it will assume your each comp is C-4. Same for Quality and any other unknown values. If you take the model produced for the comps and run it against all comps, and then take the difference between that estimate and their actual price and then rank the differences and look at the range, you can get an idea of what kind of adjustment needs to be made for all unknown variables combined. In fact, an experienced appraiser will usually be able rate any property as being at the Nth percentile in the combined features - and you can then in fact model that with an equation and get a pretty exact value. -- Of course assuming you really know your neighborhood so well, that you can pigeonhole your comps as being better than a certain percentage of the other comps with respect to quality, condition, .....

LOL. You are a nice guy and clearly not dim, however I would opine you prefer to complicate things, and over-look significant flaws in favor of said complicated methods. A typical lending appraisal needs to get done for a small amount of money. Lending pays for the party. Few other clients are willing to pay large amounts for appraisals as well. It is illogical to incorporate, and worse to promote, complicated methods as solutions, when the solutions are questionable at best, and the contributory value of employing the method is also questionable at best, and the cost of employing the method is high. Your whole thing lays on the foundation that the differences in prices paid from one property to the next, which end up being the elements of value we adjust for, occur in the market consistently. They don't. People just don't buy property like that. Do people pay more for more and less for less? Yes they do. Do they walk around with a clipboard, compile the various elements, assign a dollar value and arrive at a purchase offer price - LOL - uhm...no. Do they then do these things sub-consciously? Well, that is the ASSUMPTION isn't it? But is that assumption correct? I don't believe it is. I think people purchase emotionally and irrationally, and at the very same time, pay more for more and less for less. If my conclusion is true, how would someone go about programming that into a computer model, using only the tangible data available, which might I add is not collected consistently whatsoever and has more holes than swiss cheese? Its one of those things anyone with a decent math ability would look at and say sure, you could do it, if you had a billion dollars and even then, the results would always be questionable. Ask me about my rocket scientist story sometime.

I dunno...maybe I will be proven wrong someday. I am open to that. I highly doubt it, but I never say never.

 

[RCA]

I view that line of thinking

LOL. You are a nice guy and clearly not dim, however I would opine you prefer to complicate things, and over-look significant flaws in favor of said complicated methods. A typical lending appraisal needs to get done for a small amount of money. Lending pays for the party. Few other clients are willing to pay large amounts for appraisals as well. It is illogical to incorporate, and worse to promote, complicated methods as solutions, when the solutions are questionable at best, and the contributory value of employing the method is also questionable at best, and the cost of employing the method is high. Your whole thing lays on the foundation that the differences in prices paid from one property to the next, which end up being the elements of value we adjust for, occur in the market consistently. They don't. People just don't buy property like that. Do people pay more for more and less for less? Yes they do. Do they walk around with a clipboard, compile the various elements, assign a dollar value and arrive at a purchase offer price - LOL - uhm...no. Do they then do these things sub-consciously? Well, that is the ASSUMPTION isn't it? But is that assumption correct? I don't believe it is. I think people purchase emotionally and irrationally, and at the very same time, pay more for more and less for less. If my conclusion is true, how would someone go about programming that into a computer model, using only the tangible data available, which might I add is not collected consistently whatsoever and has more holes than swiss cheese? Its one of those things anyone with a decent math ability would look at and say sure, you could do it, if you had a billion dollars and even then, the results would always be questionable. Ask me about my rocket scientist story sometime.

I dunno...maybe I will be proven wrong someday. I am open to that. I highly doubt it, but I never say never.

I view that line of thinking, which is common among appraisers, as putting the cart before the horse. Yes, if you can do a adequate appraisal and make enough to pay the bills, then what can any one say?

But a few of us regard appraisal as an art, I suppose. The whole concept is common in professions. In software development, we have many programmers who view their job as an art (https://www.amazon.com/dp/0201896834/?tag=realestatappraat&tag=realestatappraat). The closest I can find in appraisal is https://www.appraisalinstitute.org/...l-the-art-of-appraisal-review-second-edition/, "The Art of Appraisal Review".

The term "Art" seems to be somewhat correlated to some kind of professional stature.

Anyway, back to the subject. Yes, there must be more than one appraiser out there, like me, that gets absorbed in appraising for accuracy. Lenders are however, more pragmatic, everything is a question of risk for most. And the concept of making an appraisal as accurate as possible is a bit alien; rather it should be as cheap as possible.

So, your are right. But then again, I never said using regression at this level was just for lending. Some poor folks really do care what something is worth and/or what it will be worth in the future - to the nearest N dollars - because they can't really can't afford to loose much money on their investment. And honestly, if you are going to invest $800,000 in a home, $1000 or more for an appraisal can make a lot of sense.

Oh - and watch out for reviewers, especially the ones on the state board. They can get very nit-picky -and it just keeps getting worse.

 

[Elliott]

If an appraiser has come to a method, using a statistical analysis that contributes to their confidence and contributes to the support of their opinion of value, I'm not one to tell them they shouldn't. By all means, go with it. But recognize the method you've developed is a function of your experience in your neighborhoods. Similarly, don't disparage the appraiser who understand their market in a different way and uses a more simple direct method, maybe guilty of being old school and maybe relying on their experienced opinion. Anyone who has taken a good statistics class from a good instructor should have learned about the limits, facilcies, and misuse of statistical methods. Remember, 99% of the polling done in the 2016 Election, that spent millions of dollars, using the most sophisticated statistical methods and equipment, weren't able to replicate a 50/50 human decision and got it wrong.

I think good appraisers learn about their markets, find out what works in the lending world, and what isn't considered acceptable.

 

[RCA]

Well, I certainly believe they should just leave it to appraisers to appraise and not subject them to review by underwriters. And, I think appraisers should have leeway to appraise as best as they can. - But under the assumption that we can monitor their accuracy. Tell me exactly how someone can verify that you are a good appraiser? It's kind of difficult. Some appraisers provide a fairly solid argument and many don't even supply adjustments (well at least as of some time back).

Statistical modelers compete with each other all the time to create accurate models. The same should be done for appraisers. I tried to figure out how we could come up with competitions. - But it is in reality a bit difficult.

It's up to our clients to decide how to judge us, although a lot of people question their judgement. I'd like to see competition with some difficult appraisal problems that require comparison with different property types, due to lack of comps, crossing typical market area boundaries - again due to lack of comps, homes at the high end, homes in need of significant repair and so on. A lot of this takes us into the Cost Approach and Income Approach.

 

[Randolph Kinney]

Reminds me of CompCruncher.

CompCruncher is bringing Computer-Aided Appraising Technology to the table to help appraisers appraise better. CompCruncher makes the computer work for you:

• It gathers data for you;

• It helps you understand it;

• It helps you analyze it;

• It helps you reconcile it;

• It automatically builds the report for you

Adding Science to the Art of Appraising