Grid adjustment question

[Surf Cat]

Seems a simple question, we do grid adjustments for every appraisal all the time. For GLA and lot site, if we use regression method, every sq ft should be same.

I noticed you mentioned GLA & site area right off the bat.
Make sure your sequence of adjustments is correct.

1. Property rights conveyed
2. Financing terms
3. Conditions of sale
4. Market conditions
5. Location
6. Physical characteristics

If you make adjustments in the wrong order (GLA and site which should be last) and some adjustments are based on percentage, the adjusted value of the comparables could be skewed and....misleading.

Regression is just another tool. I calculate the linear regression of my comps on my HP12c all the time before I do any adjustments. Lets me know where I'm at before I start.... cutting a line right through the most comparable data.

Sensitivity is just as important of an adjustment tool depending on what you have for comps.

 

[Terrel L. Shields]

Thanks all!

Another thing, in my local area, most single home are around $2.0M ~$3.0M. Seems hard to do the pairs analysis for Air Condition and fireplace adjustment since even two similar comps, one with AC or fireplace, another without. Maybe there is about $10K ~$20K sale price difference, it is hard to say this $10K ~$20K sale price difference is because AC or fireplace or because GLA or lot size difference since no two comps have exactly GLA and/or lot site. So just don't do the adjustment or $0 adjustment? Some peer told me use cost method to apply $5K adjustment.

You can't basically. Most things we do are based on a straight line aka linear regression. In fact, many things are not linear, rather they are based on power law or logarithmic metrics. In land sizes, as an example, per unit, 1,000 acres normally sells for a lot less per unit that 40 acres, or 4 acres, or 4,000 SF - the relationship on a short basis will appear nearly straight line, but the full big picture is a rapidly increasing curve upwards or downwards.

As for GLA adjustments. I see a lot of those where appraisers work from the sales price without subtracting the land value. Land is not subject to depreciation. NEVER. If the land is impacting the value, then an obsolescence is at work. And obsolescences are applied to the improvements, not the land. So, to make any sort of adjustment to the GLA, first subtract the land value out. Since so few appraisers seem to know what the land is worth, they are off base before they even start. That GLA adjustment is only for the improvements that are GLA. Everything else needs removed from the sales price. Garages, pools, buildings, land, site features, etc.

 

[sputnam]

Appraisers usually adjust for the residual, extracted contributory value of $ per sf of a differential, rather than raw cost per sf to build, or dividing the sale price by sf which is what RE agents do

I don;t use regression to derive adjustments, like anything else it is how it is applied, mechanically as a formula vs market derived reaction -

Regression analysis is a useful tool. The issues that I have with it are that some appraisers think it's the cure all... useful and valid for deriving any and all adjustments.... and... some appraiser don't understan regression analysis. They use software tools to grab data sets and to do the calculations. If you don't know how to do regression analysis by hand, using pencil and paper... you shouldn't be using it in your appraisals.

 

[J Grant]

I noticed you mentioned GLA & site area right off the bat.
Make sure your sequence of adjustments is correct.

1. Property rights conveyed
2. Financing terms
3. Conditions of sale
4. Market conditions
5. Location
6. Physical characteristics

If you make adjustments in the wrong order (GLA and site which should be last) and some adjustments are based on percentage, the adjusted value of the comparables could be skewed and....misleading.

Regression is just another tool. I calculate the linear regression of my comps on my HP12c all the time before I do any adjustments. Lets me know where I'm at before I start.... cutting a line right through the most comparable data.

Sensitivity is just as important of an adjustment tool depending on what you have for comps.

Right - if you grid your comps ( you can always add more to the exercise) and make the adjustments properly, in order the physical characteristics are last, and if you adjust for the other key features, such as a pool and a view, and leave sf for last, it extracts the SF adjustment price per sf as a contributory value. And THAT is how buyers more or less decide on what they pay -
I have never, ever heard of a buyer for residential properties sitting down and doing a regressionor using statistics and paying accordingly. Sure, some nerd buyer might do it, but most buyers don't. Appraisal is supposed to replicate how buyers (and sellers ) think and act. Unless a regression can be applied as an overlay or check on how the market actually thinks and acts it can produce, IMO, misleading results.

Another problem with regression is that it is difficult or impossible for a reviewer or party to replicate and see if it was viable or not, whereas using the comps right there on the grid itiis very easy for anyone to see it and determine if the adjustments are reasonable - by reasonable that is, what the market is generally doing as seen in price reaction and marketing times, and what is proportionate ot the price range and properties.

200 sf in one property can be worth far more or less than in another and it is into just because of quality or cost to build. For 1000 vs. 1200 sf house or condo, that 200 sf is hugely important - it can be an extra bedroom or family room and makes a 3 bedroom house vs. a 2 bedroom house or cramped vs. liveable. Whereas 2500 vs. 2700 house that 200 sf is not noticeable and worth is much ( typically). And then there is an over improvement where a property is simply too big for its area and price range.

It is called market value, not math value and not statistical value for a reason. Math and statistics can be used to get results and are, of course, but they are in service of supporting the market-derived value, not the other way around.

Buyers adjust with their wallets - they pay more for positive features and expect to pay less/get a discount for negative features. Any method we use should be following what they actually do as seen in in prices and what are the most similar, competitive properties they would buy instead of the subject. ( which are your comps of course or sales that could be comps )raThe same 200 sf in a

 

[J Grant]

Here is the problem with pure math applied to residential properties -

in the world of pure math, math for the sake of arriving at numbers, it all makes sense if kept to its own realm and is consistent.

4+4=8 is a pure math formula. The same formula is used to count and arrive at a solution, whether it is 8 apples or 8 gold coins. But we realize that 8 gold coins are worth a lot more than 8 apples ( unless one is starving and the 8 apples are the last food on earth ). So the value of something, anything, can differ from the math count. In normal circumstances, when food is plentiful, 1,00,000 apples might be worth 2 gold coins. Which seems counterintuitive; more is more, right? Not always. (the famous less is more saying, )

Furthermore in residential, it is often not 4+4=8. It can be, for example, 4.3 +3.7 = 8 . That is because, to the buyer, one feature is worth a more ( 4.3) then the rest of the property (3.7)

 

[Terrel L. Shields]

Another problem with regression is that it is difficult or impossible for a reviewer or party to replicate and see if it was viable or not,

Just a screen shot of the input and plot is necessary for them to attempt it themselves.

 

[Terrel L. Shields]

That is because, to the buyer, one feature is worth a more ( 4.3) then the rest of the property (3.7)

I don't think you understand how regression works.

 

[RCA]

Right - if you grid your comps ( you can always add more to the exercise) and make the adjustments properly, in order the physical characteristics are last, and if you adjust for the other key features, such as a pool and a view, and leave sf for last, it extracts the SF adjustment price per sf as a contributory value. And THAT is how buyers more or less decide on what they pay -
I have never, ever heard of a buyer for residential properties sitting down and doing a regressionor using statistics and paying accordingly. Sure, some nerd buyer might do it, but most buyers don't. Appraisal is supposed to replicate how buyers (and sellers ) think and act. Unless a regression can be applied as an overlay or check on how the market actually thinks and acts it can produce, IMO, misleading results.

Another problem with regression is that it is difficult or impossible for a reviewer or party to replicate and see if it was viable or not, whereas using the comps right there on the grid itiis very easy for anyone to see it and determine if the adjustments are reasonable - by reasonable that is, what the market is generally doing as seen in price reaction and marketing times, and what is proportionate ot the price range and properties.

200 sf in one property can be worth far more or less than in another and it is into just because of quality or cost to build. For 1000 vs. 1200 sf house or condo, that 200 sf is hugely important - it can be an extra bedroom or family room and makes a 3 bedroom house vs. a 2 bedroom house or cramped vs. liveable. Whereas 2500 vs. 2700 house that 200 sf is not noticeable and worth is much ( typically). And then there is an over improvement where a property is simply too big for its area and price range.

It is called market value, not math value and not statistical value for a reason. Math and statistics can be used to get results and are, of course, but they are in service of supporting the market-derived value, not the other way around.

Buyers adjust with their wallets - they pay more for positive features and expect to pay less/get a discount for negative features. Any method we use should be following what they actually do as seen in in prices and what are the most similar, competitive properties they would buy instead of the subject. ( which are your comps of course or sales that could be comps )raThe same 200 sf in a

No, it is not necessarily right. The crux of the matter of sequencing calculations is your definition of "Net Sale Price" - your measure of what the house actually sold for that will work across all comparable properties. It is typically "Gross Sale Price" minus the value of unusual financing terms minus the value of unusual conditions of sale. Then if your regression runs off this derived Net Sale Price, the value contributions it estimates for all input variables will be based on the target variable "Net Sale Price". In such cases Date of Sale and Location have no special priority over physical characteristics in the sequence of calculations.

Of course you are using an outdated method of calculations from 50 years ago ... which the appraisal organizations still teach.

 

[andrew81]

50 years of knowledge superseded by an avm that says location has no special priority.

Your experience is absolutely wrong. One can drive 10 minutes in any direction my market and the same exact home can have 10% to 30% difference in value.

 

[RCA]

50 years of knowledge superseded by an avm that says location has no special priority.

Your experience is absolutely wrong. One can drive 10 minutes in any direction my market and the same exact home can have 10% to 30% difference in value.

Well, of course, such posts as this are not worth answering. My goodness, .... so absolutely backward. This forum has been a waste of time for me, it appears. --- Except, the AI has finally started making some moves. ... Now the ball is rolling -- I think. Squish squash.

BTW, I have nothing to do with AVMs - you apparently don't even know what they are, although I am sure you feel like you do, but you don't.