Square footage adjustments

[EDWARD BERRY]

I THINK this is one of the most interesting subjects in appraising.

Several years ago, before powerful computers, a man named Carbone wrote a thesis for mass appraisal on this subject.

I obtained a copy from a place in Ann Arbor where they keep microfolm copies of thesis's(sic)

Basically, he inputed many sales and his "suspected" (this is for Austin) features. Then his program developed a formula for valuation, mostly for assessors.

At that time I had no education on MR, and after tahing a course on it, did not feel confident to use it.

It did show and was somewhat understandable to me, the derived value of some of the components. It gave a base value for any house and then values for the "additives"

Of course that is what most of the models do, but he took it a step more in that as additional sales were made, he inputed them and then the program recalculated the formulas.

This is something like McKissock(sic) had yeras ago but unfortunately due to high interest rates we did not have enough sales (or business) to use it.

With many good sales today it would be interesting to try that program and see if it gives decent results. (DID ANYONE ever use it)?
I believe in using all the tools availabe and MR is one that if understood could be very helpful.

Respectfully-ed in arkansas

 

[Austin]

FEDERICK: I have my program in a work sheet in a workbook. After I locked the cells I wanted to make a copy for some one, so I copied the work sheet into another workbook and put the file onto a floppy disk. When I took it home and opened the file on another computer there were no formulas and no graphics. I haven’t had time to figure that one out yet. This system you are referring to is something I came up with based on the same theory but looks like a conventional marketing grid. I make this model for simple-minded bankers and examiners. It looks like a marketing grid and it is to a point but based on a totally different theory. I want to test it out on a few people but if you don’t understand the underlying theory as I have outlined in this post you won’t understand the purpose of it, you still have to use the correct sequence of adjustments. If you understand the theory you could program your own model so it is another circular reference. Let me figure out what is going on with this copy problem and I might get back to you.

Edward: You would have the confidence to use a model if the model was self-verifying and after you finished you knew not only that the model estimated the price of the subject but also the regression equation also estimated the price of the comparables and compared them with the real world results thus verifying the accuracy of the equation. Then, you can use the graph of the results to reconcile the remaining factors plus the regression tells you actually how much of the correlation you have actually explained in the equation. This subject is so comprehensive that it is difficult to discuss. For example, I have found they you can’t use my model or any model AVM or MRA on a universe of data for complicated reasons. I first filter the data and pick about 25 comparable sales from the same property class. I then have a test to filter out affected or outliner properties, and then I do the regression on about 20 really comparable sales. Then led by the graph of GLA vs price, I do a stepwise linear regression analysis to find which variables influence price and which ones don’t. When I get the variance down to a low level or close to the trend line, I stop adjusting and reconcile the remaining factors.
A lot of people like Frederick are probably asking: “Why can’t we get a copy of this model?” The answer is the model is different for every problem. For example, in large older houses size does not correlate with price. The value factor in these houses is condition or degree of renovation. This model is not an idiot box that you plug numbers into and out comes the answer, it is a tool for extracting value influencing factors from the market. If I gave you a hammer and saw, could you build a house?
I think Edward asked me if I ever used McKissic’s model. I took his class once and got a copy of his program. We were sitting in class and Matt was doing something, regressions I think with his model and the members of the class were using the computed regression value factors for making adjustments on a conventional marketing grid. Not good in my opinion! I loaded his program onto my computer and never used it….. I don’t use programs or models that I haven’t seen, don’t understand, can’t verify etc. I have an MAI buddy that for years has been trying to get me to use Argus I think is the name of it, but no thanks, if I can’t program my own and understand it and know the algorithm and theory mesh, I won’t use it. For reasons just outlined, I would never use any store bought regression or dcf program.

 

[Farm Gal]

:lol:

Thanks Austin!

Once again you have (quite gently actually) explained why statistical tools in the hands of the uncomprehending or incompetent become dangerous!

Lies Lies and statitistics :twisted: :!:

It is precisely the combination of knowlege of the local market, AND access to a program, AND most importantly the wisdom to do what you do that I'd like to pick your brain on!

Some of the manipulations as the fairly self evident one you cite: older big homes should NOT be adjusted 1st for size then other factors... OK MOST of us figured this out about the time we did our first few big barn houses.

But I would be willing to bet that you have some 500+ scenarios packed away inside your skull based on what you have learned about not only the program, but your market that would be pearls of great worth. I wouldn't ask you to cast them before swine, but there are more than a few of us out here who'd like to get a little boost into your expertise level. I don't want to learn by rote, but I'd sure like to see what works in yoru area, then comehome and play with a bit more knowlege in mine!

 

[Fred]

Been waiting for this to slow down to chime in.

Austin,
When you say "work the cost approach backwards," I am not sure what you mean. To adjust for size, I usually back-out the land value from the market prices, so that the amount considered is the contributory value of the improvements only. When I do this, I am looking to see what the rate of change is to the marginal contributory value of square feet of improved space, throughout the range of size possibilites. If the land values of all the sales are equal, or contributes equally, you can leave it in, but if it is not, you have to adjust for the difference or preferably take the land out.

So, does work the cost approach backwards, mean take out the land value?

 

[Austin]

Steven:
Now you are getting to the heart of the problem. Try this when you have time. I also use an Excel self programmed cost approach form. Do the cost approach in the same sequence as the cost manual calls for starting with the selection of the basic house cost factor and proceed in the book sequence. Then after you have finished, go back up and estimate what percentage of the total replacement cost each feature contributes. What you will find is that the basic unit cost factor, the one from the Marshall Valuation Tables, accounts for about 70% of the total improvement cost and the next most significant factor is about 11% and it tapers rapidly from there. I did a graphical analysis once of the entire cost manual. For example, in M & S cost manual, a 1,000 square foot dwelling after adjustments for location can range from $30 to about $140 per square foot. If you don’t do a very good job of picking the correct basic cost factor you could easily erase the significance of the remaining 30% of value factors. Miss the quality of construction factor by a small amount and you just wiped out a fire place, deck, and carport. This shows you what is significant and what is not. If you never interpolate between quality classifications when using the cost manual, you are definitely screwing up. This is the basis of the appraisal problem as I see it: We start off with a highly significant number that reflects the quality of construction that is without question the most significant value determining factor. This number is really pregnant. Then we add additional value factors, and then we add land. Then we expose it to the market and strange things start happening. All of these factors create enterprise value in the total property. Some times the factors have synergy and the value is well above cost, and some times they don’t. Every one of the value factors from land to dwelling to basement to carport etc., makes some contribution to value we hope, but not like it does in the cost approach. The difference is due to synergy of factors or obsolescence of the total property enterprise. This results from the matched or miss-matched value components as they interact with the market. This process is in a constant state or change.
For this reason, in my view, you can’t back the land or any value factor out of the mix because they are inseparable. If you backed the quality of construction component cost factor out and replaced it with another one the whole chemistry of the property would change and all factors would then make a different contribution. In plain English, you can’t solve this type problem by making line item adjustments alone because you can’t measure the synergy or obsolescence cause by the total package, so you have to deal with the total aggregate of the property enterprise which consist of all value factors working in harmony, hopefully.
How can we do that? I say it has to be dealt with statistically and least sum of the squares regression is the perfect method of doing that. This chemistry of complementing value factors is what complicates the appraisal process. If it worked in the sales comparison approach like it does in the cost approach, life would be easy. When I use to play with regression analysis sometimes I felt like I was trying to communicate with some living being that was locked in a box and I had to figure out how to get it out, then one day I realized it was a living being. Its name was “The Market.” He is real, I swear, I have talked to him! He gave me a message for you appraisers: "Catch me if you can and when you do, I will tell you all of my secrets!"

 

[Fred]

Austin,
First, I would like to say how much I appreciate the time you put into the forum and I agree with the principles you advocate, pretty much right down the line. However, I must confess that I have trouble understanding you. The one about the shirt buttons is a riot. I may use that one.

You say, <span style='color:blue'>" ...you can’t back the land or any value factor out of the mix because they are inseparable." </span>

Maybe my use of the terms 'back-out' the land value or 'adjust' for the differences is misleading, but here is the problem. I do not see how anyone can determine the size adjustment among a set of house sales with differing land values, without accounting for the land value separately in the model.


Here is as example. Below is my sample from a perfect market. The square-foot contributory values is exactly the same for each property. Or, if you prefer to say, price varies with size in a perfectly linear relationship.

Sale	SF	Price

1	2,500	281,250

2	3,000	352,500

3	3,200	396,000

Now, show me how you get the size adjustment out of this sample, without asking me the land values.

 

[Austin]

Steven:
The size adjustment based on the data you presented would be the slope of the trend line, which is roughly $160 per square foot. One of your sales most likely sale 1 has a land value $11,000 higher than the other two sales proportionally, but it is hard to tell which one because there are only three data points. You omitted a subject property so the only answer I can give is the slope of the trend line.

The next question is “I do not see how anyone can determine the size adjustment among a set of house sales with differing land values, without accounting for the land value separately in the model.” You don’t determine a size adjustment (size or GLA is a vicarious variable meaning it represents more than size. It also includes different site values, noise, etc.), you equalize the trend line by using iterations with guesses of the size adjustment until the trend line is level or equalized, then you can reconcile the residuals or different land values from the data points to the trend line to gauge the different land values. We are using least sum of the squares to average out the noise after we adjust for physical differences. Excess land is a physical factor and would have been adjusted for in the process so I assume you mean different site values. Your question as posed raises questions of highest and best use or comparability of comp sales in my mind.

I am not sure if you are referring to what I discussed in my last post above about why you can’t back out the land value. I think the answer to that question is when you extract the land out of an improved property then you don’t have a problem if the highest and best use is as improved or there is little obsolescence in the property as improved. On the other hand, if the highest and best use of the land and improvements are different or there is significant obsolescence in the property and you extract out the land value as though vacant based on its highest and best use value you may have a problem because the land may not be contributing the value as though vacant. I know in appraisal theory we allocate all of the obsolescence to the improvements, but I do not think the market does it that way because the land is part of the total property enterprise value and thus a covariant variable that cannot be separated from the whole. How would you know what land value to take out?

I think I answered your questions.

 

[Caterina Platt]

Austin,

Have read many of your past posts, and reread most of these a couple times. This is way too comprehensive, complicated, etc. for my little brain to understand via posts. Possibly books may help, but at this point, I'm afraid it would have to break it down to the 'place thumb and forefinger around the pencil...' super basic steps. Hope I'm not the only one fighting the glaze over effect as I read, reread, and try to decipher the meanings here. For me anyway, this would have to be at least a couple week course with 'hands on' exercises. Some of it hits home with the neurons, but as you say, I'm sure it's just enough to make me dangerous.

If you ever get the urge to teach this, please, please, do let us know. I'm sure there would be many takers here, me for one. Developing such a course is a formidable task, I would expect to pay a good chunk for such a valuable one as this. [/i]

 

[slacker]

Steven:
The size adjustment based on the data you presented would be the slope of the trend line, which is roughly $160 per square foot.

Austin,

Did that $160 psf value start at about $61 psft? I'm trying to play along at home.

From Steve's data I'm using GLA as my dependent variable and price per square foot as my independent and came up with a slope of approx 61.

Using an assumed subject GLA of 3000 sf I graphed my adjusted price per square foot using the 61 as a slope. From there I kept increasing that number until my graph was relativiely flat. That final number was at or about your figure of 160.

Am I on the right track with this?
I noticed that the graph was relatively flat with a figure of anywhere between 135 and 180. How do you decide on the final number? Or do you just shoot for the middle?

 

[slacker]

I noticed that the graph was relatively flat with a figure of anywhere between 135 and 180.

<span style='color:blue'>Oooops!

Well, I just realized that the trend line "appears flat" at $160 psf but it has a lot to do with the scale of the graph. My trend line is actually the flatest at about $43 psf which makes more sense to me than the $160.

Based on an assumed subject GLA of 3000 sft. At $160 psf the range of my adjusted price per square foot was 113.75 to 144.5 with a trend line that "appeared" to be flat.

At $43 psf that range was 120.5 to 121.25 with a line that was "actually flat".

Oh well, it was fun while it lasted. In the end that $43 bucks is just about 1/3 of the average price per square foot of the comps.

Seems like by using the trend line I just took the long way around to get to the same figure I would have used in the first place.</span>