Principle of Substitution, ....

[RCA]

Thought I would post this from my recent post in the Facebook AI Group.

I would just comment,

(1) Sure, theory, and protocols are one thing, practice is another.

(2) You would expect the theory to be correct, but it is trash. SCA depends on the Principle of Substitution - which the AI has apparently spent years and hundreds of thousands of dollars lobbying for. The Principle of Substitution, the foundation of the SCA, is trash and so is the technique of matched pairs analysis which builds on it. Why is the Principle of Substitution trash? - Because it only works if all of your significant variables have good measures. The issue of unmeasured variables such as condition, quality, aesthetics/design, functional utility, view, etc.,, is treated as if it were not important - yet in many areas such as the SF Bay Area, it is indeed very important. Secondly, the matched pairs technique is trash because it pulls adjustments out of mid-air. An adjustment is a difference. So, if you have adjustment A, there should be some S (subject attribute value) and C (comparable attribute value) so that A=S-C. Why do you need S and C? - These are the value contributions that go into the final sale price in the case of the comparable or the estimated sale price in the case of the subject. You need to go through the determination of value contributions to get adjustments - that is "having your feet on the ground" - your support. (One can also mention that matched pairs invariably disregards the set of unmeasured attributes as if they were not important - or assumes they are the same between comparables when they are not - and finding a good match just based on measured variables is difficult enough ...) - Now a good regression technique like MARS, capable of providing high R2 values with other useful features, can get you the value contributions of significant measured variables (it determines which ones are significant) - but it can only give you the total value of all unmeasured variables (from the residual difference between the estimate based on measured variables and the actual price of each comparable) - which, as it fortunately, turns out - is good enough. So, the theory and protocols could be corrected. The caveat being, a much higher level skill set is needed to put that theory into practice. You need to understand MARS, probably how to program in R or Python - and be able to create a workflow to handle the tedious work - without making significant errors. Not easy.

The tedious work after you have done all the work to create a good price model is:
1. Calculation of value contributions for all measured variables for all MLS comparables used in the regression - you may have hundreds.
2. Calculation residuals for all comparables
3. Ranking of all comparables by residual.
4. Calculation of Residual (CQA) Scores for all comparables
5. Manually find the rank of the subject property in the sorted comparables and assign it a Residual (CQA) Score.
6. Rerun the dataset to get the adjustments for the URAR. These may be for variables for which there are no slots in the URAR - so you have to set up methods to aggregate the adjustments into URAR fields in an Excel spreadsheet.
7. Now calculate all adjusted sales prices for all MLS comparables (hundreds).
8. Select the best comparables for the Sales Grid - move them to the top of the list. Let's say the top 12.
9. Run the next stage program - which will then extract the data for chosen comps to a spreadsheet for upload into Alamode.
10. Upload into Alamode and fine-tune adjustments for the unmeasured variables, breaking the residual adjustment into separate unmeasured variable adjustments for attributes such as condition, quality, functional utility, view, design, etc..

It is a lot of work that must be done precisely and it will give you the same adjusted sale price for each comparable - to the dollar (only because of rounding - otherwise to the cent). So, this has to be done programmatically.

 

[George Hatch]

The SC has an IRL track record. It's utility is almost universally considered by the users to be sufficient for their usage.

That's an objective fact, not an opinion.

Now you might be right that your way is better, and by all means, if you can sell it for your price then I think that's great for you. But it will be the users who decide what methodology is meaningful to their usage; not the appraisers.

 

[Salty]

Sometimes I feel like appraising is similar going to medical school then working as a doctor in a third world country for peanuts to help the greater good.

 

[Zephyr999]

The SC has an IRL track record. It's utility is almost universally considered by the users to be sufficient for their usage.

That's an objective fact, not an opinion.

Now you might be right that your way is better, and by all means, if you can sell it for your price then I think that's great for you. But it will be the users who decide what methodology is meaningful to their usage; not the appraisers.

Exactly!! Write to the audience and don't baffle with BS. Regression alone is not intuitive enough...nor can a general audience understand it so the opined value becomes a form of confusion.

 

[Functional Obs]

When I see the word "trash" used in a formal argument, I think the author must be about 15.

 

[George Hatch]

The principle of change continues unchallenged. The technology evolves. With that said, what we are trying to emulate are the perspectives and actions occurring in an inefficient and imperfect market. The primary usage of an appraisal is to avoid the gross overencumbrance. These two factors are of effect on the expectations of the users. We can expect those expectations to evolve, and we can expect that appraisers will respond accordingly. So "never" is obviously not an appropriate or reasonable response.

In the meantime, the reality of our business looks like this:

Q: "Can you include this analysis in your appraisal report?"​

A: "Sure. What's the fee?" ​

​

IMO

 

[Terrel L. Shields]

The principle of substitution applies to all three approaches. Since we are dealing in determining a CASH VALUE perhaps we should be relying upon the income approach not the sales nor cost approaches. Everything gets translated into dollars anyway, right?

 

[RCA]

I didn't write down in great detail the problems with the Principle of Substitution - just enough for you to think about.

But most here are clearly biased to support the principle through, rather than analyzing the details.

The Principle of substitution completely depends on the concept of "similar properties". If you read the 14th Edition and similar appraisal texts, the role of unmeasured variables such as condition and quality is sometimes mentioned - but at critical junctures ignored. If you look at the residual graphs in the analysis I did of the cities in San Mateo County, (e.g. Atherton: https://pacificvista.net/price_models/elementor-34/ Page 12 ) you will see:

1. The values of unmeasured values are almost never the same between two properties. They are always different. In the case of Atherton, the value contribution of all unmeasured variables, ranges from -5,137,877- 6,771,876 - that is a range of about $12M. You cannot just push these to the side and ignore them. Yet, the characteristics in this group do not have any measures, so ask yourself how are you to even objectively compare them? You can't. And that is the big lie you have succumbed to. You pretend they simply don't exist - you pull something out of thin air based on your so-called experience - which is completely questionable because you are incapable of any kind of sophisticated analytical capabilities to deal with such properties - which are all unique.

2. Big Lie. 14th Edition, Page 31: Beautiful Obfuscation: "Although the principle of substitution applies in most situations, sometimes the characteristics of a product are perceived by the market to be unique. The demand generated for such products may result in unique pricing. For example, a market may not have ready substitutes for special-purpose properties like a historic residence, medical office building, or high-tech manufacturing plant. In those situations, the appraiser may have to research substitute properties in a broader market or employ analytical techniques appropriate for limited-market properties." Does it really apply to "most situations"? Hardly. All properties are unique, simply because they are situated in different locations. So, you may indeed define some rules for classifying properties as "similar" - but no one has done that. If we are talking about residential properties in the SF Bay Area, to find truely similar sales transactions within a recent period is nearly impossible. Rather you need to analyze the differences by creating good model and then make adjustments to those differences. And when you do this correctly, you will find that the Principle of Substitution is worthless.

3. There's another problem. That is the circular thinking involved in the descriptions of the Principle of Substitution. It is always based on the assumption that the "typical" or "average" potential buyer knows if and when properties are "similar." Appraisers don't have a rule for defining what is "similar." And every buyer has their own perception of such things - which we can't measure because we can't see into their thinking. So we have no way of calculating what "similar" really is to the typical buyer. Without that, substitution really isn't possible. That is to say, unless you have an objective way of defining a set of similar properties - you have no way to ascribe the "lowest sale price" to one of those properties. Dead end.

So, people can look at a set of houses and they can rank them on some set of criteria based on their perception of similarity. And different people would probably come up with similar rankings. But the issue of defining the value differences between each house in their ranking, - the job of the appraisers - that is a different story. It is not easy. In my Atherton ranking, two houses may be close in vale if they are in the middle - or quite distant if value if the are near the extreme ends. I can define that fairly accuratly based on a ranking. - But it takes a specific technique, a technique that does not involve substitution. - No, not at all.

 

[George Hatch]

And yet, those properties get appraised and the users make their decisions based on those appraisals all the time. Isn't that weird?

 

[George Hatch]

You might get a more robust discussion of these and other quant-related themes with the appraisers who participate in Mr Dell's blog and groups and such.

Home - George Dell, SRA, MAI, ASA, CRE %

George Dell, SRA, MAI, ASA, CRE. Contact Us Analogue Blog by George Dell. Free Weekly Blog Posts. Stats, Graphs, and Data Science Classes

georgedell.com