Square footage adjustments

[Austin]

Steve: In your first paragraph you clearly get the point when you say one of the comps is out of line after equalizing which indicates that something other then physical differences are affecting price. That is the whole point of adjusting for physical factors first. If these physical factors correlate with price and explain the factor/price relationship then obviously nothing else matters.
Then in your next paragraph you digress claiming that you just know a time adjustment is justified and mention cash equivalency. If a sale is two years old and not cash equivalent, but is comparable otherwise, and you equalize it to near the trend line by accounting for physical factors alone with current comparable sales, then how could time or cash equivalency be a factor? If they were a factor the sale would not equalize near the trend line. Again, that is the whole point of the theory. If the equation, trend line, explains the relationship of value influencing independent variables purely on physical factors alone, then how could you justify saying that a time adjustment or cash equivalency adjustment is justified? How many times have you seen a cost approach on a new dwelling with additions for time, cash equivalency, location, view, room count, etc.? If the market reflects these factors at all they must be residual factors that can only be detected after all else is accounted for. If you count your chickens and none are missing, then why go looking for chickens that aren’t missing? If you count your chickens and one is missing, then you go looking for one chicken. If you know your chickens, you will know the chicken when you see it.
As you said: "What is wrong with using your head?"

 

[Steve Owen]

I think your flaw is relying on math to the exclusion of common sense. You say that physical factors must be accounted for first. By what logic? The view, location (which actually is a physical factor) quality of construction (also a physical factor), and conditions of sale are all factors that might as easily affect value as the size, number of garage spaces, etc. Because they don't fit your neat little graph, you discount them. So, if you have two houses that are identical physically, but have a different sale price, the cause must be something else. Of course, the market is not perfect, so it could just be a different price.

But, if I look at the two properties and notice that one has straight line construction and the other one has a hip roof with lots of ins and outs, then I'll bet the difference in price is caused by the design and appeal. There is simply nothing wrong with an appraiser being smart enough to notice that factor first.

Frankly, I don't think there is anything wrong with you adjusting for the physical first before such subjective items as "view." However, you are making a grave error when you don't bring the comparable to a cash equivalent basis for market conditions and conditions of sale before adjusting for any other factor. For example, if normal practice is for buyer and seller to split closing costs but the comparable sale in question was sold with the seller paying an additional $1,500 in closing costs, then that house sold for $1,500 less than the reported sale price. This has nothing to do with any physical characteristic - the house actually sold for $1,500 less than the price on the sale price section of the grid. If you don't adjust it you are guilty of a substantial error and most likely will come to a misleading conclusion. The same principle holds true for market conditions (time) and other conditions of sale, such as below market financing.

Most of these adjustments are rare for the typical residential appraiser. Usually we can find good comps that don't require such adjustments. That doesn't mean we should ignore the necessity of taking them when they are needed just because it doesn't fit the preconcieved notion of how some mathematical equation or set of equations should work.

 

[Farm Gal]

Austin:

I really DO wish you'd show up in FLorida, as with my background in stats (inclusive of plenny practice with early SPSS versions) I have enough stats knowlege to hold my own on THAT end.

What I don't have is the ability to use that knowlege effectively in analysis of RE! I know that within about three hours of conversation I could learn enough from you to aid that process considerably. I am going that direction anyway, but would sure like a few pointers from someone who has trod the path.... Austin has messed with this stuff for countless hours and larned a few things along the way.

I'd like ot shorten the learning curve!

Folks if you have not yet taken any college level course in advanced stats, specifically one with up to a year long course in MLR, access to computer labs with good teachers, best take this chance to do so. You are likely going to need it to stay ahead of the college trained folk coming up behind you...

Steve:

You are WRONG, buddy. All the items you mention can and should be codified, quantified and included in a analysis IF they are pertinent! When you run your analysis and you get a heteroscedatic graph, there is probably a difference that needs to be eliminated or worked with! Again the beauty is that you can fairly easily identify what is and is not pertinent!

But here is where WE the trained value experts in our local market have the BIG HUGE advantage over some desk jockey 6 states away: we have more knowlege about what buyers and sellers think and do in our area than that person unfamiliar or that machine! It's best to TEST that knowlege: you may find that what you thought was 'real important' may not have the import you were thinking. This is where some folks get real testy with Austin, but I suspect to a huge extent that some of the adjsutments we make using 'paired sales' and 'knowlege of the market' are so much hooey! Or at least not as $$ significant as we thought 8O .

The beauty of a multi-linear regression which Austin has tried time and time again to explain to us knotheads, is that you can rather quickly eliminate factors with little relevence OR remove those sales which are such outliers as to tweak your curve. When you get an outlier and make a few phone calls it is stone amazing how quickly you get "the rest of the story"!~ OK so sale #123 is inappropriate to use because 5 people got shot there four weeks ago and it has stigma. If you are looking to solve for stigma you may have a need for this sale, otherwise toss that puppy OUT and move on with your analysis!

Now some of us are either familiar enough with our market, already know the scoop on that particular sale or are savy enough to have already tossed that sale as unsuitable, but the MLR process can let you use a significantly greater number of sales and develop information that no human on this earth could process in the same manner. And someone form outside your market may already be running a similar study on your comps and if they find that all three of your comps are outliers thay may be calling to ask a few questions or RIGHTFULLY sending the thing out or review :twisted:

 

[Austin]

Steve: “By what logic do I account for physical factors first?” Answer is the cost approach theory and the fact that the sequence of adjustments comes from the market. Read the chicken example in my last post. You can’t have one theory of value in the cost approach and a totally and contradictory theory in the sales comparison approach. Don’t forget, the sequence of adjustments comes from the market and is not one of your personal preferences. Appraisal theory is like buttoning up your shirt. If you get the first button in the wrong hole, there is no way you can work it out at the top or bottom. You are insisting on importing adjustments into a marketing grid without any justification, which is putting the first button in the wrong hole. How can it be clearer-if you pick a group of comparable sales, then all adjust to the trend line, then there ain’t no chickens missing. If one sale is out of line, go looking for a chicken with location stamped on its head.
You did bring out one good point about cash equivalency adjustments. If all of the sales were not cash equivalent, then you would have to adjustment them all right out of the shoot. But if only a few are not cash equivalent then the model will indicate that fact and even show you how much adjustment is required for cash equivalency. If the non-cash equivalent sale is $5,000 above the indicated price range after equalizing, then that is support for a $5,000 adjustment for that factor. Same for location. If one of the sales has a superior location, then it will show up on the graph and tell you how much the location adjustment is. The case is most often that you end up with a sale out of the range due to a lot of factors. The beauty of this system is it smokes out these oblivious factors causing the price variance, but in reference to the trend line, you know how to gauge the variance or adjustment. You can’t import data into a data set and say: “Boy! Sales in this location are always $10,000 higher than in that location so I will make an adjustment for it.” How did you extract that sequence of adjustments from the market? If you have a variety of sales superior location will show up. That is the reason I don’t like to use clone sales. They don’t tell you anything about the market in general. Once you pick the comparable sales you have to stick with the data set. If you bring in excess baggage you will upset the apple cart. All of those market condition, time, etc., adjustments you mention can be accounted for only by using sales in the data set to smoke them out. If a recent comparable sale adjust equally to a one year old comparable with pyisical factors only, then how do you justify a time or condition of sale adjustment. I think your match pair theory is catching up with you. With match pairs you can prove the world is flat. You can also prove it is round.

 

[Steve Owen]

If all of the sales were not cash equivalent, then you would have to adjustment them all right out of the shoot. But if only a few are not cash equivalent then the model will indicate that fact and even show you how much adjustment is required for cash equivalency.

Now you're talking. I have said before that I don't believe I have a problem with what you are doing, as well as I can understand it from what is posted, but do have a problem with some of the statements you make, or at least with the way I interpret them. That is the case here, I believe.

As long as your model shows an adjustment is needed and accounts for it then I have no problem with your practice. If you are adjusting in dollar increments it makes no difference what you adjust first on an ordinary grid (it might make a difference with your particular multiple regression model, of course). However, this gets a little dicey if you are adjusting by percentages. The reason for this is that the actual price at the top of the grid on the form is not the price from which you must adjust when the price is not cash equivalent. That is why most instructors I have had refer to such adjustments as "above the grid" adjustments.

 

[Austin]

Steve:
The main thing I have learned from playing with regression is that you can only measure things when you have something to contrast them with. That is why I preach against these state appraisal boards that go after people because they don’t like their comp sale selection. For example, if two sales are equivalent except for cash equivalency, then the only way to measure that factor is by comparing the equalized prices of the one with and the one without. For every such factor, there has to be some sales with that factor and some sales without that factor. In other words, if you can't bracket it you can’t measure it. Basically that defines the regression process. What regression analysis actually does is take a lot of equations, in our case comparable sales, and by adding many sales with factors that some of the other sales don’t have, we can solve for the unknown factors.
For example: X+Y+Z=10. If we can find a lot of equations where we know X, some where we know Y, and some where we know Z, then we can work the regression and find out what X, Y, & Z are. These factors could be any number but only one set of numbers will satisfy the original equation. Now if we used cloned sales like the state appraisal boards insist upon, then we could never solve the equation.
In real estate appraisal, it is a little more difficult because the X, Y, & Z’s represent the center of a bell curve for a range of that factor meaning the factors have a range of values. Some factors are more significant than others with greater weights, etc., which only a regression analysis can detect. For example, if X is 9.9 in the original equation above, then the values of Y & Z may not even be significant given the scope of the problem. Regression will point out forces that you don’t even know exists like the synergy or covariant variables (factors that correlate to each other). Sometimes these factors have synergy meaning 2 + 2 can =6, and then there is dyenergy in which case 2 + 2 can =3. It depends on the chemistry of the property. When you have a lot of these value factors that show synergy and dyenergy they interact in such a way that not even conventional regression analysis can measure them because they exhibit what is know as multicollinearity meaning the regression process can’t separate the factors out because the value factors correlate with each other. That is where the algorithm comes into play. You have to have a system or algorithm to deal with multicollinearity, and it is also why you can’t solve these problems with line item adjustments. This means you can’t deal with value factors one at the time, you have to deal with them in the aggregate in the correct sequence. That is another reason AVM will never work, because AVM cannot determine the correct sequence of adjustments. An example is age adjustment. Sometimes that is the first adjustment you make, and some times you don’t make that adjustment at all. Sometimes size is the only adjustment necessary, and sometimes size doesn’t matter at all. The bottom line is, if you can’t match it with sale data, you can’t measure it, and you certainly can’t go outside the data set with matched pairs and bring foreign data into the equation.
I think it would be worth while as a teaching aid to each appraisers regression just to get them thinking correctly. If they don’t learn anything else, at least they will understand the proper sequence of adjustments and that you can’t corrupt the data selected by bringing junk data from outside.

 

[Pamela Crowley (Florida)]

Austin wrote:

... I preach against these state appraisal boards that go after people because they don’t like their comp sale selection.

At least in Florida the comp selection item is due to ignoring similar comps in the subject subdivision to use comps from other areas that will inflate the value. If an appraiser here is nailed for poor comp selection, it would be obvious to anyone what they did and why.

I realize, especially after attending the NCAB meeting, that not all boards are realistic.

 

[Mike Garrett RAA]

I really really agree that if there were "better" comparables in the same neighborhood and the appraiser has gone outside the area just come up with a higher value...they should be dealt with to the fullest extend of the law.

Review one where there were a half dozen current sales of the same model in the same subdivision that the appraiser conveniently over looked. This is out and out fraud!

 

[Austin]

Mike & Pam:
The comp selection problem you two are concerned about result from using a limited number of sales. By using only three comparable sales you can cherry pick sales from the same cookie cutter subdivision and skew the price estimates just by picking high or low sales. That is an even better agrument for a regression method. If you use 20 sales from a variety of neighborhoods not only can you not defraud but you have the support for the location differences right on the graph. The better neighborhood sales are in the upper range and the less better neighborhood in the lower range. Then too, if the neighborhoods are as cookie cutter as you two describe, who needs an appraiser to certify that model 2A's sell for $75,000. That is AVM territory. There are no neighborhoods like you describe where I live. If they start out that way they evolve into something else with additions and remodeling. It must be a boring life appraising cookie cutter subdivisions. I use to live in Colorado Springs where Mike lives and I remember the cookie cutter subdivisions my soldier buddies lived in. When I use to go to the airport I remember it was surrounded by a cattle ranch and I use to see the rancher and his kids out ridding the range. When I left in 1971, the ranch had been sold and it was flagged for a huge subdivision. I would hate to see it now. I think AVM will eventually take care of your concerns in that type environment.

 

[Frederick R. Ruffell]

Hey Austin,
Now that you know how to lock the cells in Excell whould you be willing to sahre a copy of your spread sheet and graphs? I really want to start using regression analysis in my work. I have some understanding of the statistics involved. IF you do not wish to share can you at least reccomend a text that will get me started on the path of setting up my own excell spread sheet?

One other thought, what are your views in using regression analysis with the income approach as it applies to small income properties?