• Welcome to AppraisersForum.com, the premier online  community for the discussion of real estate appraisal. Register a free account to be able to post and unlock additional forums and features.

Applied Regression Including Computing And Graphic

Status
Not open for further replies.

Frederick R. Ruffell

Senior Member
Joined
Jan 21, 2002
Professional Status
Certified General Appraiser
State
California
Over the past year or more some of us have been hungry to learn more about regression analysis in residential real estate appraising. Because Austin has been unable to make his particular spreadsheet available I started to investigate other programs. I came across "ARC" a free statistical program available on line from the University of Minnesota, http://www.appraiserdepot.com/ad_regressio...s_software.html.
I have just received the text "Applied Regression Including Computing and graphics" which is the companion to the program. A low level of statistics is required ("one semester of basic statistical methods"). I wanted to see if any of the other forum members would like to start reading the text and doing some of the models together, maybe in a separate forum, to learn this regression thing and maybe some guidance from Austin???? Pam I think you were also interested in some basic regression and I think this (ARC) is the perfict fit so far, free software and ther are several used texts available through Amazon.com
 

Bobby Bucks

Elite Member
Joined
Jan 27, 2002
Professional Status
Real Estate Agent or Broker
State
North Dakota
I downloaded ARC about a year ago and played with it some. It appears the easiest to use are the paid versions. Matt McKissock of McKissock data systems has a version available called Dragonfly which is similar to Realstat4. I would love to see a forum on this subject because like it or not, the business is headed in that direction. Doug Smith wrote a nice review on Realstat and McKissock's version which was very informative. The article is called "Is There a Blue Box AVM in Your Future?" to read it

click on egram at http://naifa.com/
 

Austin

Elite Member
Joined
Jan 16, 2002
Professional Status
Certified General Appraiser
State
Virginia
Frederick:
I would love to do something like that. There are a number of problems I have had with sharing my method. 1. It requires a lot of graphics, regression tables, etc., that can’t be accessed on the Internet. 2. It doesn’t work like you think it does. It is not a model with blanks that you fill in, hit enter, and out comes the price. The whole essence of regression analysis is that it reveals the most basic force in the universe that affects property values and gives you a method of working with it and measuring it. It is called covariance of variables. That means all value factors interact with each other and for that reason you can’t deal with them one at the time. 3. It is really not necessary to get into regression because it is to complex and most people would never spend the time to figure it out, not to mention that I have invented a simpler version for residential use and this about all anybody needs. Same problem though, it ain’t no fill in the blanks out pops the number deal. You have to understand the theory of sequence of adjustments theory, etc., and use it as a tool not a device. The appraising has to take place inside the appraisers head based on sound theory and all of these methods and algorithms are just to do the math. 4. I have tried to send E-mail and disk to people and first off, they couldn’t open the file because there are to many formulas etc., I couldn’t copy my simpler version to disk for some reason (formulas would not copy, and I have found it a waste of my time to send this stuff to somebody and they don’t even make an effort to open it. Look at the book and run and call me a quack for rocking the boat.
When you play with regressions you do hundreds of regression trials to find out what is affecting value and that involves hundreds of pages on the screen to demonstrate how you got to where you are. I will review this program tonight but I will bet I can tell you right now why it won’t work: Because the creator never heard of covariance of variables and has no idea how to deal with that problem. If you look at their list of independent variables and you see more that 5, see things like room count, number of bedrooms, lot size, and general quatative factors that are covariant, you might as well forget it, it won’t work. I read the other day in a course I was taking on GIS that some one did an experiment and 85% of property values could be explained by GLA and age. There is a reason for that. See if you can find my article that was published on AppraisalBuzz about a year ago.
 

Frederick R. Ruffell

Senior Member
Joined
Jan 21, 2002
Professional Status
Certified General Appraiser
State
California
Austin,
You know I really can not figure you out! On the one hand you tout the benefits of statistical analysis and graphical analysis as being so damn better than any of the "Voo Doo" methods the rank and file appraisers use. Then when some of us with a little background in statistics and a hunger for knowledge (because we can see some limitations in the existing methods and possible benefits of statistics) seek out some wisdom from someone like yourself you seem to have myriad reasons why we should not or cannot use statistical analysis in our appraisal process. If you sincerely want to help enlighten us that are willing then by all means join in and guide us around the pitfalls but stop telling us why we should not go down this path, I for one have already started walking.
 

Frederick R. Ruffell

Senior Member
Joined
Jan 21, 2002
Professional Status
Certified General Appraiser
State
California
Also Austin,
The authors introduce covariance on page 60 of 500+ page text. :eek:
 

Austin

Elite Member
Joined
Jan 16, 2002
Professional Status
Certified General Appraiser
State
Virginia
Frederick:
First of all, do an archive search using my name and regression. You will find enough reading to keep you busy for at least a week. You don’t need to know a lot of statistics because it can all be done on the computer. I use Excel and it has all of the regression analysis you need. People have been playing with regression since the 1920’s, so it is nothing new. As I have stated before, regression analysis is a mathematical process that works perfectly every time if it is applied to the correct algorithm. By that I mean if you take a lot of random variables out of the blue and stick them in a regression, you will get crap. The thing different about my approach is that I combined regression methods with an algorithm to extract the sequence of adjustments from the market. The problem as I figured it out was that the most significant force affecting real estate prices is that value influencing factors can’t be handled one at the time like you do on a marketing grid. They have to be handled in the aggregate. The reason I harp all of the time about size adjustments being voodoo is that size adjustments are not just size adjustments, actually what you are doing is adjusting out covariance of variables one of which just happens to be size. If you don’t do this in the correct sequence using the correct adjustment extracted from the data you are doing nothing but screwing up the data beyond repair. This is why I have been gripping about the 8 or more possible methods programmed in to most software programs. You can literally come up with anything using one of these methods.
In my view here is what you need to do if you want to get into regression. Get Excel’s program because it is all you need and if you want to discuss this stuff everybody needs to be singing from the same hymnal. You need to learn to work some simple regression problems starting with simple regression graphs then some 3 & 4 independent variable examples.
Then comes the fun part. Which independent variables to use, or more to the point based on my experience, which ones not to use. Here is where my sequence of adjustments methods comes into play. The first step and last step in the process is a simple regression graph of GLA vs. unadjusted sale price and GLA vs. adjusted sale price. There is a reason for doing this. To understand the reason you have to learn a new appraisal theory or way of looking at things. After working with regression for so long, I have an entirely different mindset for solving appraisal problems. Examples of what I am talking about are: Definition of a perfectly adjusted data set; making compound adjustments that most people call a size adjustment; understanding that you can’t import data into a set of comparable sales. Once you pick your comparable sales, everything you need to know is right in front of you. As you can see it is a whole new ball game and it is not something you will pick up in a week or two. If you knew how much time I spent trying to figure this stuff out you probably wouldn’t believe it. The beauty of it is that when I finish I have a graph of the results and can verify the accuracy of the results. Here is a key thing to remember: “The purpose of appraising is to find the equation that best explains the correlation between independent variables (value influencing factors) and the dependent variable price.” There is no question about whose equation is the best. If my equation can predict the actual prices with less variance than your equation does, then I win. That is the standard of verification. That is my biggest grip with the present method of appraising, absolutely no process of verification of results. If you want me to prove it, take your last residential appraisal and graph GLA vs. adjusted prices. Let the program calculate the trend line for your data. I will guarantee you that the trend line will have a slope not significantly differently from the slope of GLA vs. actual unadjusted prices before you ever started. The bottom line is that if the slope of that trend line is not = to zero, you haven’t done anything but shuffle some numbers around. I have never seen another appraisers marketing grid that was perfectly adjusted and I have used some model appraisals from software programs to prove it to myself. Anyway, you get the drift. I guess if you want to stay in this business you don’t have any choice but to learn this stuff but also remember that I don’t make a living teaching. There is an appraiser in Tucson, AZ, that is a statistical whiz. Jo Ann Stratton had him call me a while back. He has taught some AI classes and is writing a course on the subject that should be out soon. Lee Ann and some others have taken his class and can tell you more about it.
 

Frederick R. Ruffell

Senior Member
Joined
Jan 21, 2002
Professional Status
Certified General Appraiser
State
California
Ok Austin, I understand that you do not want to teach. :idea: But we are still going to ask you questions. :idea: OK?

Pam lets all get a copy of the text. http://www.amazon.com/exec/obidos/ASIN/047...2121389-5736954
And download the free software. http://www.stat.umn.edu/arc/software.html
Then I think we all run it like a book club and read a chapter a week or month and use the software as directed from the text!, you see the beauty of this is that the text is written for the software (hence the title "Applied Regression Including Computing and Graphics") and vice versa.
Anyway I got the text today and am going to start reading and will update you as I go. I believe that the authors use some appraisal examples in the text. I am the kind of person who is more motivated when others are involved so I hope some others are interested and will get the text and software and join in. :lol:
 

Eric Boggs

Sophomore Member
Joined
Feb 23, 2002
Professional Status
Certified Residential Appraiser
State
Florida
Hi all,

I am very interesed - count me in. Eric
 

Austin

Elite Member
Joined
Jan 16, 2002
Professional Status
Certified General Appraiser
State
Virginia
Frederick:
Look at those appraisal examples in the book and tell me what independent variables they use. (value influencing variables like size, rooms, baths, etc,) I purchased a book on regression analysis written by two of the leading authorities in the World and their 2nd example was an appraisal problem. These guys know everything about statistics there is to know but their appraisal regression example is a farce. They know nothing about economics or appraisal theory.
They used 14 independent variables and never address, more likely are not even aware of, the significance of covariance of these variables. I took one look at this crap and knew they were out to lunch. Here are their independent variables:
Bedrooms, GLA, fire place, rooms, storm windows (oh brother), lot area in sf, annual taxes, #baths rooms, garage sf, construction type, location 1 & 2 for two different residential zoning categories. Then they make the statement: “It is true that a perfect model is seldom possible.” They should know. We are not talking about mass appraising here, I am talking about appraising a single property to do a conventional appraisal just using different methods and practices to make the equivalent of what most appraisers call adjustments.
This is what I get so frustrated about when people constantly tell me: “Regression has been around since 1920, and nobody has ever made it work.” That is right, nobody has ever made it work, and the reason is that they don’t under the appraisal theory it takes to make it work. My point is that knowing everything about the mathematics of regression is not going to help you one bit. You have got to learn the new theory before you can apply it properly. If you come up with the perfect appraisal model and the regulatory authorities can't understand it, what have you accompolished? This is a case of the tail wagging the dog. The system does not want a perfect model because it can't be manipulated. The pefect appraisal model could not exist in our present environment so don't get in a hurry to learn it. The Mae's would chock on it.
 
Status
Not open for further replies.
Find a Real Estate Appraiser - Enter Zip Code

Copyright © 2000-, AppraisersForum.com, All Rights Reserved
AppraisersForum.com is proudly hosted by the folks at
AppraiserSites.com
Top

AdBlock Detected

We get it, advertisements are annoying!

Sure, ad-blocking software does a great job at blocking ads, but it also blocks useful features of our website. For the best site experience please disable your AdBlocker.

I've Disabled AdBlock
No Thanks