Regression Analysis

[Deleted member 130081]

Like most tools in the appraiser's tool box, regression analysis is only as good as the user. However, far too many residential appraisers reflexively dismiss regression analysis because they don't understand regression, they don't have even a basic grasp of statistical analysis, they have never used regression, and it does take some additional work and skill to be utilized properly.

No. Regression is only as good as the data you feed it.

 

[timd354]

I have opened my horizons, maybe too far even. It would be great to go back to the bliss of ignorance.

No, I do not claim to utilize matched pairs, as they are almost never available and for the same reason, regression analysis of adjustments is also not possible, at least not in any credible manner as far as I am concerned. Like you said, regression and paired sales for adjustments are essentially the same thing, where regression is just on a larger scale. If the data to do matched pairs isn't out there, how do you populate your regression model? Garbage in garbage out. A car is a great invention, but it will only run properly on gas, it wont run on plastic. Regression is no different.

I have never stated that regression analysis is appropriate in every circumstance. In some areas, sufficient data needed to run a good regression analysis is difficult or impossible to find, although sometimes the data is there although it needs to be cleaned up and cross checked (which can be a tedious and time consuming task). By the way, contrary to those who believes regression only works in areas that have a very homogeneous housing stock, nothing could be further from the truth.

From your posts, I take it that you probably work a rural area with limited data, which means that most any method used to determine adjustments is difficult and probably not all that reliable. Just curious, but what method(s) do you use to determine adjustments in your markets?

 

[timd354]

No. Regression is only as good as the data you feed it.

How is that different from paired sales or any other method that you utilize to determine adjustments? And how is that different from the sales comparison grid in your appraisal reports?

Part of the required skill in using regression is knowing how to check data and normalize the data by doing things like reasonableness checks, correcting for know errors, reconciling multiple data sources, etc. in order to best eliminate unreliable data points and obvious outliers and correcting the usable data to the extent possible. .

 

[George Hatch]

Sensitivity analysis using more comparable (and better qualified) data is way simpler to use and way simpler to explain as a means of getting to "meaningful to intended users" . IRL Quality > Quantity.

Most of your valuation occurs when you pick your comparables, not when you tweak the range with adjustments. Aside from the quants, who else would really care if a bedroom adjustment is $10k or $15k so long as it effectively helped refine that range of adjusted value indicators? You're still going to end up ranking your subject somewhere within that range when you reconcile for the one and only "value conclusion".

The only thing an inefficient combination of adjustment factors does to one of these analyses does is compel the use or more subjectivity in the reconciliation.

 

[Gobears81]

No. Regression is only as good as the data you feed it.

The property that I was trying to use Excel for a regression analysis/ graph was 1,300-acres. The largest sale that I could find of that type of property was only 500-acres. I was showing the relationship between size and price, but since an across the board adjustment was necessary, I wanted to show what the sales would imply for a value of the subject since there were no paired sales, etc. I didn't put full emphasis on the regression analysis and adjusted the sales by a lower level than that which the regression implied, but it was nonetheless useful support. Garbage in, garbage out is not the most appropriate phrase, considering that the sales were actually pretty good outside of the discrepancy in size.

 

[timd354]

Sensitivity analysis using more comparable (and better qualified) data is way simpler to use and way simpler to explain as a means of getting to "meaningful to intended users" . IRL Quality > Quantity.

Most of your valuation occurs when you pick your comparables, not when you tweak the range with adjustments. Aside from the quants, who else would really care if a bedroom adjustment is $10k or $15k so long as it effectively helped refine that range of adjusted value indicators? You're still going to end up ranking your subject somewhere within that range when you reconcile for the one and only "value conclusion".

The only thing an inefficient combination of adjustment factors does to one of these analyses does is compel the use or more subjectivity in the reconciliation.

I agree that sensitivity analysis is a very useful tool, especially in residential appraising and I certainly don't think that regression is necessary for most residential appraisals. That being said, regression does have its place and can be particularly useful in analyzing market reaction when the use of significant across the board adjustments is necessary, especially when those across the board adjustments are made to items that really drive value such as significant differences in GLA, site size, location, etc.

 

[Deleted member 130081]

Like most tools in the appraiser's tool box, regression analysis is only as good as the user. However, far too many residential appraisers reflexively dismiss regression analysis because they don't understand regression, they don't have even a basic grasp of statistical analysis, they have never used regression, and it does take some additional work and skill to be utilized properly.

No. Regression is only as good as the data you feed it.

I have never stated that regression analysis is appropriate in every circumstance. In some areas, sufficient data needed to run a good regression analysis is difficult or impossible to find, although sometimes the data is there although it needs to be cleaned up and cross checked (which can be a tedious and time consuming task). By the way, contrary to those who believes regression only works in areas that have a very homogeneous housing stock, nothing could be further from the truth.

From your posts, I take it that you probably work a rural area with limited data, which means that most any method used to determine adjustments is difficult and probably not all that reliable. Just curious, but what method(s) do you use to determine adjustments in your markets?

LOL. First of all I only make dollar adjustments on mortgage appraisals that require them - I use qualitative notations on all other work. My method? Trends analysis. I work urban, suburban and rural. I find the data to be the same in all three, which is a lack of isolative data. I have also worked analyzing national data, where I discovered real estate data is fairly similar anywhere you go, which is, that homes are not similar, so therefore neither is the data (again, isolative data does not exist). Dollar adjustments are a farce in most cases, at least in the 26 line item adjustments found on the 1004. I understand the theory behind them and why the method was subsequently "recognized" and in theory, its great. The problem is there is no reliable method to determine adjustments in a manner than can be supported by presenting market data, not one. If there was, we would all be using it instead of scrambling and arguing about what to do. Dollar adjustments on a mortgage appraisal were dreamed up by the client, and the appraisal powers that be fell in line. Its one of those things that sounds great, and sure would be neat if we could in fact solve that problem, but is based in fantasy. I think our mortgage clients have known this for a long time, or are becoming keenly aware presently. For a very long time, no one cared. It was not an issue until quite recently. This explains why appraisers never bothered to say anything about it, though we really should have been all along (USPAP states we must reconcile all methods and techniques). The reason adjustments have come to spotlight recently, is the FNMA Selling Guide, which brokers have learned the hard way must be followed. In the very small appraisal portion (sorry Marion I don't know it offhand - B4 or something), there is a list of unacceptable appraisal practices. Now, these are not speaking to the appraiser mind you, rather to the underwriter who uses the selling guide to determine if the appraisal (and loan) qualifies or not. However, as has always been the tradition of underwriters, instead of testing an appraisal for compliance themselves, they simply forward the test list to appraisers and demand they produce appraisals to those specs. This is a form of influence that was never intended, nor has it ever been stopped either, and accounts for all the ridiculous things we think are required, like sales within a mile etc. Anyways, one of these unacceptable appraisal practices states an appraiser must make adjustments derived from market data, and that the adjustments must be supported. The funny thing is, that's all it says. There is no written guidance on how an appraiser might go about accomplishing this, which makes sense when you realize there is no way of accomplishing it. But, there it is. Any appraisal that does not support its adjustments is not illegible for sale to FNMA. That's a mountain of truth if you really sit back and think it over. So, if an appraiser claims they have employed a certain technique to determine their adjustments and the truth is they have not, they are participating in mortgage fraud, as the underwriter doesn't care about your adjustments, just that you somehow state they are supported and you sign your name to that. Be careful is what I am saying, because mortgage clients don't want supported adjustments because they give a turd about the actual adjustment, rather they want you to sign your name stating that you did it. USPAP says we must develop and report our conclusions to our clients in a manner that is meaningful and not misleading. So, do you suppose an appraisal that cites matched pairs or regression for adjustments can really pass that test? Why would we instead not just tell our clients they ask too much? The answer is because it would bring down the mountain. I say protect yourself from the avalanche and word your reports carefully.

 

[Deleted member 130081]

The property that I was trying to use Excel for a regression analysis/ graph was 1,300-acres. The largest sale that I could find of that type of property was only 500-acres. I was showing the relationship between size and price, but since an across the board adjustment was necessary, I wanted to show what the sales would imply for a value of the subject since there were no paired sales, etc. I didn't put full emphasis on the regression analysis and adjusted the sales by a lower level than that which the regression implied, but it was nonetheless useful support. Garbage in, garbage out is not the most appropriate phrase, considering that the sales were actually pretty good outside of the discrepancy in size.

I get it my friend, I get it. I just wanted to point out that regression should be used with caution. I wrote a very long post on adjustments in general that you might find interesting - no good solutions to offer, but might shed some light on why we are being asked for support more than ever before.

 

[Michael P Jacobs MAI]

Econometrics is rocket science - literally. Without training we can use statistics for illustration purposes but nobody without a degree in it (or maybe if a true genius) should be using it as a primary method. That said, the more you understand the better you will be able to describe value parameters. If you like statistics and have the grades, take a couple of graduate level courses over a year... it cured me, but at least I can use the vocabulary (heteroscasticity is my go-to concept explaining why my comparables might suck under certain circumstances lol). We used Limdep, which I understand now can not only do the math but geographically determine the sample to use as well, but you would still have to know the problems inevitably present in every data set. Good luck!

 

[Michael P Jacobs MAI]

<SNIP>...First of all I only make dollar adjustments on mortgage appraisals that require them - I use qualitative notations on all other work...

Just remember that the dollar (or percentage) adjustment method we're all using - without empirical proof - is simply another qualitative adjustment. The only difference is that we're giving some indication of the magnitude or impact of the positive or negative "direction" of the adjustment. We're "nudging" the value one way or the other, and using numbers does not make it a true quantitative method. Anyway, that's my story and I'm sticking to it as Hal Smith would have said.