How Many Comparables Are Too Many

[TRESinc]

Bert, I find it questionable that you have this ability yet none of the big data companies (that have a lot more data than you) can do this. How do you do what you do, by yourself, when those companies have very specialized people who are highly educated who cannot get to your precision? You are better than the company with 20 guys with a degree in statistics?

Pardon us all who do not believe that a comp from 2001 has any significance to a sale in 2019.

i had the same thought. i even asked bert why he was wasting his time completing appraisals but got crickets in response.

See, that's the problem. If House Canary were smart they would hire me; only, I **** on Python and the CTO loves Python and won't talk to me, - because he could care less about appraisal. Only I'm about the only guy in the world that could help them.

well i guess that about says it all. random_guy_01 on a message forum is smarter than everyone else. got it.

 

[Terrel L. Shields]

Pardon us all who do not believe that a comp from 2001 has any significance to a sale in 2019.

No but I do retrospectives. My oldest was from a DOD in 1969 and last year I did one from 1994 and one from 1986. So all that info is still useful and I still keepm all those old MLS records from the quarterlies they used to publish.

I can see trend analysis as being predictively of long term changes. Tracking land prices here will show break points in value as property transitions from agri to development.

 

[TRESinc]

No but I do retrospectives. My oldest was from a DOD in 1969 and last year I did one from 1994 and one from 1986. So all that info is still useful and I still keepm all those old MLS records from the quarterlies they used to publish.

and how much of the data from those retro appraisals from 1969, 1994 and 1986 did you apply to ones that had an effective dates from 2018?

 

[RCA]

"Pardon us all who do not believe that a comp from 2001 has any significance to a sale in 2019."

I agree....
And that includes land sales from years ago too....

And you fall for an argument based on a false syllogism. No one said a sale from 2001 has significance for a sale in 2019. However, if a model can predict the sales prices for all 350 sales in an area going back that far in time with an R2 of over 0.99, it is an argument supporting the strength of the model. That is something you should think about. ... Now you are wondering how that is even possible. Well, to tell the truth, I'm cheating a bit with residuals. But even without residuals, I think in this particular case the R2 was around 0.91.

Does that mean that appraisers can be displaced? No. But take a tool like this and give it to a real estate agent trained a bit and with a good knowledge of the neighborhood and they will be able to do a fantastic job with their BPOs. However, honestly, it may become a problem when an appraiser's opinion conflicts with one of these models. And thus, appraisers will be wise to take them into account, - in many areas. But it is years off before this becomes a problem of any magnitude. Progress in the real estate industry occurs at a snail's pace, - that you can be sure of.

 

[Amy Perkins]

Great advice!! In the commercial world I am ecstatic if I have 3 good comps. But if I am appraising a new Taco Bell and find 8 similar sales, i'll use them. Often times I am able to prove tweaks to the cap rate by analyzing the comparable sales.

I would say it depends. If you have no good comps and a lot of features to bracket use more sales. Sometimes you just don't have enough sales.

 

[ucbruin]

And you fall for an argument based on a false syllogism. No one said a sale from 2001 has significance for a sale in 2019. However, if a model can predict the sales prices for all 350 sales in an area going back that far in time with an R2 of over 0.99, it is an argument supporting the strength of the model. That is something you should think about. ... Now you are wondering how that is even possible. Well, to tell the truth, I'm cheating a bit with residuals. But even without residuals, I think in this particular case the R2 was around 0.91.

Does that mean that appraisers can be displaced? No. But take a tool like this and give it to a real estate agent trained a bit and with a good knowledge of the neighborhood and they will be able to do a fantastic job with their BPOs. However, honestly, it may become a problem when an appraiser's opinion conflicts with one of these models. And thus, appraisers will be wise to take them into account, - in many areas. But it is years off before this becomes a problem of any magnitude. Progress in the real estate industry occurs at a snail's pace, - that you can be sure of.

I'm not criticising your preference for data collection and saving them for nearly 2 decades....

But I'm neither an agent nor in the business of predicting....

 

[Ariba]

I have done appraisals in Pacifica and elsewhere, all difficult areas, and often use 9+ comp. Here is an appraisal from a ways back:

Comp Adj. Sale Price % Diff
Comp 1, $1,213,195, 0.03%
Comp 2, $1,211,799, 0.08%
Comp 3, $1,214,924, 0.18%
Comp 4, $1,215,866, 0.25%
Comp 5, $1,207,219 , 0.46%
Comp 6, $1,206,509, 0.52%
Comp 7, $1,230,169, 1.43%
Comp 8, $1,192,095, 1.71%
Comp 9, $1,235,656, 1.88%
Comp 10, $1,238,876, 2.15%
Comp 11, $1,174,596, 3.15%
Comp 12, $1,173,149, 3.27%
Comp 13, $1,252,709, 3.29%
Average $1,212,797

Consider that the GLAs ranged from 1,100 to 2,760 sf and the sales prices from $825,000 to $1,300,000, the condition and quality from near the bottom to the best. The adjustments were applied consistently across all comps without exception. This is not easy to do.

As to why so many comps? I am in an area where the subject can very easily be quite dissimilar from EVERY comp, so that you are completely dependent on adjustments for value. You prove the quality of your adjustments in two ways: (1) The consistent application of a model developed from a likely complex and sophisticated regression that few can really understand and (2) Applying your model to a large number of comps and getting a narrow band ( < 5% ) of adjusted sales prices.

Why not just use all the sold comparables in the MLS and divide by the number of sales to get an average? Common sense goes a long way!