Principle of Substitution, ....

[shrubberyvaluation]

Oh she is now famous: https://www.millerandperotti.com/StaffProfiles

License intact, not a scratch.

Look, I have seen, I know of - many appraisers, highly successful ones, who have had major problems and just kept going. All kinds of legal and board problems. Yep. Bull-headed enough to get in trouble and keep going.

License intact does not mean business is intact. I'm not claiming she did anything wrong, let alone something that would warrant her license being revoked. I am talking about clients not wanting someone who has a poisoned public image even if through no fault of their own.

 

[Joe Flacco]

Countless. How many parcels of real estate are identical in every one of those variables? How many buyers have identical life experiences while simultaneously affected by exactly the same circumstances, and make identical decisions as each other in every circumstance in life? The fact is, except as the buyer in purchase of your house, you don't have a clue what other buyers and sellers think or act upon in a real estate transaction. You know what you think "you" would do in their place, but you don't have a clue. It is simply a fantasy to believe you are recreating the motivations of every buyer and seller in the market. You don't have a clue about 99.9% of their motivations. What are facts, though, are sale prices and property characteristics. The results of a those differing motivations is what we analyze.

Yes, countless.

 

[RCA]

That is all good and fine but larger set of data is always inferior to smaller set of data. No question about it.

You don't know much at all about regression. The larger the data set - the more intelligence it contains for extraction.

Now, where you are going wrong is that probably all you know is linear regression. In that case, what you are saying is true. If you are doing linear regression, then a smaller data set will give you a tighter fit ---> at the risk of making it less reliable. However, you need to be using MARS regression - where the lines can bend at "knots," and it goes through thousands of calculations to find significant variables and then models for each of the variables to account for differences in value contribution across different ranges. Even with MARS though, too large of a data set can cause problems when your variable behavior drastically changes over time. Although, you can probably get around the problem by increasing the number of interactions from 1 to 2 (or possibly 3).

 

[RCA]

License intact does not mean business is intact. I'm not claiming she did anything wrong, let alone something that would warrant her license being revoked. I am talking about clients not wanting someone who has a poisoned public image even if through no fault of their own.

Do you actually know anything here? Or are you just guessing? I would bet (and I admit I don't know) - that she did a ton of work in 2021 and 2022. That house was in a small group of houses setting on poles - that were really very difficult to appraise.

Hypothetically, assume you come to the conclusion that the value of a house is between $800,000 and $1,200,000, but you can't be more accurate than that because of a lack of evidence. Your client insists on a point value. What would you conclude? If you use that middle point of $1,000,000, you could certainly claim that is unbiased. But then if the real value is $800,000, you would be $200,000 too high. What would you do?

 

[Joe Flacco]

You don't know much at all about regression. The larger the data set - the more intelligence it contains for extraction.

Now, where you are going wrong is that probably all you know is linear regression. In that case, what you are saying is true. If you are doing linear regression, then a smaller data set will give you a tighter fit ---> at the risk of making it less reliable. However, you need to be using MARS regression - where the lines can bend at "knots," and it goes through thousands of calculations to find significant variables and then models for each of the variables to account for differences in value contribution across different ranges. Even with MARS though, too large of a data set can cause problems when your variable behavior drastically changes over time. Although, you can probably get around the problem by increasing the number of interactions from 1 to 2 (or possibly 3).

I'm no expert on it but know enough about it. But I understand real estate and the data that is databased, organized, and available for regression or MARS. That type of analysis ignores anything that is not organized in a database to be analyzed. You pretend the countless number of variables do not exist. Plus the variables are dynamic depending on combination of other variables. Real estate market is very complex.

 

[TerryRohrer]

Only for the simplest of properties can an algorithm or large dataset methods be effective. In which case it doesn't even matter if it is three comps or 30. Anything even a little bit complex is better off with manual analysis.

On what basis are you making that claim? One option you have some understanding, the other you don't have a clue, so the infallible one is the one you can spell?

 

[TerryRohrer]

Yes, countless.

Well, at least you understood the one word sentence. Unable to respond to 2+ word questions?

 

[shrubberyvaluation]

Do you actually know anything here? Or are you just guessing? I would bet (and I admit I don't know) - that she did a ton of work in 2021 and 2022. That house was in a small group of houses setting on poles - that were really very difficult to appraise.

Hypothetically, assume you come to the conclusion that the value of a house is between $800,000 and $1,200,000, but you can't be more accurate than that because of a lack of evidence. Your client insists on a point value. What would you conclude? If you use that middle point of $1,000,000, you could certainly claim that is unbiased. But then if the real value is $800,000, you would be $200,000 too high. What would you do?

I'm guessing that lender clients would rather not use one appraiser than the elevated risk. It is not hard to see how this could be a hassle and even costly for a lender to use the appraiser when there are complaints about value. I'm not saying the appraiser did anything wrong, I'm saying that its more ammo for borrowers complaining if they google the name and see it associated with a bias lawsuit regardless of the facts.

What is "real" value. I normally deal with market value. I appraise for market value when that is my assignment.

 

[TerryRohrer]

I'm no expert on it but know enough about it. But I understand real estate and the data that is databased, organized, and available for regression or MARS. That type of analysis ignores anything that is not organized in a database to be analyzed. You pretend the countless number of variables do not exist. Plus the variables are dynamic depending on combination of other variables. Real estate market is very complex.

How do you ever complete an appraisal? You seem to be suggesting that you manually account for each of those countless variables. The reality is that you ignore or assume some unsubstantiated notion about more of them than you analyze, and claim they are identical in those regards. But I suspect you reports don't include a fraction of the assumptions you are actually relying on.

 

[Joe Flacco]

How do you ever complete an appraisal? You seem to be suggesting that you manually account for each of those countless variables. The reality is that you ignore or assume some unsubstantiated notion about more of them than you analyze, and claim they are identical in those regards. But I suspect you reports don't include a fraction of the assumptions you are actually relying on.

That is the point of narrowing it down to 3-6 properties. Why would you want to add more variables and then just pretend they do not exist?