Regression Analysis

[DWiley]

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.

Agreed. My college degree is in math/physics, and I started working in an appraisal firm right when I started college. So, I have been using regression to solve appraisal problems since I was in college in the 1980s. After lots of experimenting I finally was able to apply the math in a practical way. As I noted, that is a big gap in the training - and one that I have shared with most who conduct such training.

When I hear of appraiser who jump into regression after attending one 7 hour seminar (or less) I cringe a little bit I took a very intense course in the design of regression models during my days at Vanderbilt, and I just have doubt that the key points can be conveyed to the typical appraiser in a single day

 

[Terrel L. Shields]

heteroscasticity

I just call it skew although I think you refer to the skew across variable factors is not equal

I have a complex program I rarely use, http://www.realstat.com/realstat.html

I more likely use a simple stat often to test if a factor is actually driving value. "Formulaic" answers usually don't mean much.

This are more useful to me. https://www.business-spreadsheets.com/solutions.asp?prod=57

Danny links one of the best

For adjustments I use a two stage age/condition and SF sensitivity grid that is simpler. Paired sales are meaningless unless you can pair a significant number of them

 

[Michael P Jacobs MAI]

I just call it skew although I think you refer to the skew across variable factors is not equal ... SNIP... For adjustments I use a two stage age/condition and SF sensitivity grid that is simpler. Paired sales are meaningless unless you can pair a significant number of them

LOL see what I mean - heteroscasticity sounds like skew in a way but has nothing to do with skew, or at least not the skew in a trend line. It has to do with the "fray" at the ends of a forecast or observation, like as you get farther and farther from a "typical" land-to-building ratio the less useful the ratio becomes as an indicator... think of the two ends of a frayed rope and how difficult it is to say where the "rope" is as a point estimate as you move into the fray. Well, maybe you do get it... my brain is hurting just thinking about my attempted econometrics graduate degree.

Anyway, your're right about going simpler as long as it does the job. I'll frequently graph land-to-building ratio against $/SqFt of land under certain circumstances but it is worthless as a primary value indicator and just descriptive. Do what market participants do and don't over-think it, right?

I am not a fan of paired sales unless the property is a commodity like a simple tract-home.

 

[timd354]

Paired sales are meaningless unless you can pair a significant number of them

I wish that more appraisers understood this statement. Aside from the statement that "the adjustments were derived from paired sales" being one of the oft repeated lies told in residential appraisal reports, many appraisers don't seem to realize that using one or two so-called paired sales to determine any sort of trend is at best very weak.

 

[NachoPerito]

This are more useful to me. https://www.business-spreadsheets.com/solutions.asp?prod=57

Anyone else purchase this set of excel add-ins for $75? Is it worth it?

 

[Joe Flacco]

I think paired sales is great. I use it all the time.

 

[timd354]

I think paired sales is great. I use it all the time.

I have no doubt that you and many appraisers do use it all of the time, but what makes it great?

I get that is what you were taught to do and and it is quick and easy to do, but what makes it so great?

What does one or two paired sales really tell you about market trends? How is a single paired sale indicative of anything, much less the amount of an adjustment?

 

[Joe Flacco]

I have no doubt that you and many appraisers do use it all of the time, but what makes it great?

I get that is what you were taught to do and and it is quick and easy to do, but what makes it so great?

What does one or two paired sales really tell you about market trends? How is a single paired sale indicative of anything, much less the amount of an adjustment?

Here is a paired sale. It tells me that prices have been stable in this neighborhood between 2015 and 2016. It also tells me C2 vs C1 requires no adjustment when almost new.



Here is another paired sale. It tells me that prices have been stable in this neighborhood between 2014 and 2016. It also tells me C2 vs C1 requires no adjustment when almost new.



Here is another paired sale using sales in rowhouse condo conversion. It shows me that the middle level units sell for about the same. It also tells me the basement level units sell for $100k to 130k less than middle level units. It also tells me the top floor unit with roof deck sells for about $150k more than the middle level units.



Here it is again in the building next door that is new conversion. The units are about 10% larger and is the same developer. It tells a similar story. as the last grouping.


I can do paired sales all day and people can understand. Regression, R value T value blah blah blah. BS.

 

[Salty]

I do not like any regression programs because I never have a large enough sample size of properties "comparable" enough to my subject to use it. I used one offered via TOTAL, but the thing just automatically pulled from a radius. (I cant remember the distance) But it made no sense because that meant is was pulling properties from 3 states and areas that are not comparable. At the time, there was no way to manipulate the parameters. People just want an easy push button way of generating an adjustment. It's human nature, but it can get you into trouble.

I think the best way is to pull comparable sales from the MLS, import it to excel and run the regression there for each item needing an adjustment. Just make sure you know what the data is telling you. It's never a very large sample size, but it is better than nothing and I typically only use this type of method as additional support for the adjustment. I like at least 3 matched pairs followed by a simple regression for support. Add in some common sense and I'm done. However, these days I am making less adjustments due to lack of recent data in any one market segment within the school district in which I am appraising. It is more qualitative. This adjustment process is why it takes me so long to complete a report. Push button adjustments are too risky.

 

[Joe Flacco]

Regression needs a lot of data. With enough data for regression there is too much noise or differences between the properties in the set. Useless.

Let AVM's crunch big data. Appraisers should use old fashioned methods for developing adjustments.