Author: David A. Braun, MAI, SRA (President, Braun & Associates, Inc.) has been actively engaged in real estate appraisal, review, and consulting since 1976. In this period, assignments have been performed throughout East Tennessee for numerous clients
This is the final part of a three part article on "market modeling". The purpose of this series of articles is to introduce real property appraisers to the concept of market modeling in a way that helps remove the mysterious cloak surrounding it.
In Part I we discussed that market modeling covers AVMs, matched pairs, regression, artificial intelligence, and expert systems. The article pointed out that appraisers perform market modeling every time they fill out a sales grid when performing the Direct Comparison Approach. In theory, even the most complex AVM program can be duplicated by hand calculations.
The climax of Part I was that all of the methods represent one science; as their purpose and nature are all similar. In fact, appraisers know a lot more about market modeling than they think. A regression application that I programmed in Excel was provided (Excel utility) so that appraisers could get some hands on experience with automated modeling.
The video tutorial was designed to give the appraiser enough information to use the application; but was not intended to be a "how-to" for regression analysis. Based on feedback from many of you, the application that I provided was very functional and much easier to use than many other programs available to the appraiser. My goal was for appraisers to see that applying a proper regression analysis intended to extract adjustments was much more labor and time intensive than we had been told.
In Part II we focused on regression analysis and laid out some requirements and goals that would be necessary for any market modeling system to be used in everyday appraising. These were:
- Produce a more reliable (accurate) value opinion without adding too much time.
- Shorten the time to perform an appraisal of equal quality.
Click here to continue reading . . .
Our discussion must stay focused on these two requirements as I am not qualified to discuss regression analysis in any other context. We discussed that the functions of extracting sales and finding a value were two different things.
It was my opinion, based on my own limited knowledge and testing, that regression analysis for the purpose of extracting adjustments was not practical for the typical appraiser.
I believe there is a method of analysis other than regression that will enable the appraiser to include it in a reasonable time period and aid in the selection of comparables and in for their own application. It will result in a more defensible value and will reduce the time to perform the appraisal by evaluating large numbers of sales, moving them to the grid, automatically inserting adjustments, and inserting the explanations for the adjustments. This requires that the appraiser maintain a residential database such as the one in WinTOTAL and in many competing appraisal forms packages.
While I do not believe regression analysis meets the criteria we listed above for extracting sales, it is very suitable to provide final value opinion. This would be a low end appraisal product that could be performed by appraisers and be USPAP compliant. It would work much like the application that we made available in conjunction with Part I of this series of articles, except it would have an expanded report that includes all USPAP requirements.
Because it works as a "Black Box" it is not necessary to keep working the analysis until the model looks perfect, you are done as soon as it predicts reasonable sales prices for the sales as compared to their actual sales prices.
The application is very practical for forming value opinions associated with many assignments that do not require a high degree of accuracy. Some examples would be low risk mortgages, to assist in the review function, to provide preliminary values, etc. There are features included to help the appraiser ensure that the subject property is similar enough to the sales used to produce a credible value opinion. It can be used in conjunction with a desk-top or curbside data gathering process.
For a residential desk-top (Although the product is not limited to residential properties), it should take from 30 to 60 minutes to research the subject property, download sales from the MLS, run the regression analysis, and complete the form. The Desk-top version of this application is intended to be a 24 hour service.
It's designed to reclaim much of the Broker Price Opinion business and some AVM business that we as appraisers have lost over the past few years. While this is a low fee product, it has a lot of potential for a high dollar per hour earnings. It is my goal for every appraisal office to have the ability to provide this service and to report it on a standard form. I want to thank a la mode for their support and encouragement on this product and the use of the Labs to get the feedback necessary to produce a functional product.
SERIES CONCLUSION
Let's conclude our discussion of Market Modeling. We discussed in Part I that a model is an explanation of an outcome, and that its usefulness is in its ability to predict some event. It is crucial to understand that the model itself may or may not be based on correct assumptions.
The value of a model, whether based on correct or incorrect assumptions, is measured by its predictive power. That's not to say that the validity of the model's premises is not important, but the accuracy of the model is easily tested against the sales used. Often, the real actions of a market are too complex for computer programs to deal with, resulting in poorer results than a much simpler model.
This doesn't seem so weird if we stay focused on the fact that the automated part of the model is purely mathematical. This concept is indeed contrary to what appraisers have been taught about valuating a property.
For example, in the real world a regression model may be more accurate when it considers fewer property characteristics. This may be because of inaccurate data or because of some covariance between characteristics (an independent variable). If a model is considering 50 sales it may not be practical to begin verifying each sale for inaccuracies.
When the fit of a model to the data set is accurate, the model is still useful even if it is based on bogus assumptions as long as the subject property is very similar to the sales data used. However, such a model will produce an absurd value for the subject property if it is not similar to the sales data. This is why it is imperative that the measure of an AVM's confidence rating consider the conformity of the subject property to the sales data used.
Perhaps the most important thing that we've learned as appraisers about the mathematics of the adjustment process is that all of the techniques and computer modeling methods work best when a human appraiser is involved in the process.
It surprises me that many appraisers define the "Holy Grail" as some fully automatic system that works infallibly without any human intervention. I suppose they are looking for a way to mute the "bad" appraisers.
Certainly, I have not provided any extensive mathematical theorem to prove that appraisers are needed. The greatest evidence is that if appraisals could be done without the appraiser we would already be gone.
The theory of market modeling must be further explored by appraisers. At this point numbers alone can give a decent final value, especially in stable markets. However, their accuracy is limited because of the subjective nature of human behavior within real property markets.
The primary edge that appraisers have over automated market models is not in interpreting the numbers, but in knowing why buyers and sellers behave the way they do.
This knowledge is imperative in rapidly changing markets. Appraisers must utilize all of the technological advances available to them. To date there seems to be two distinct creatures; appraisers and automated market models. Perhaps as we learn more about automated market modeling and learn how to use tools like regression analysis the distinction between the two will begin to diminish.
This article was written by a practicing appraiser who is not an expert in regression analysis. Hopefully, this information will bridge the gap between what appraisers currently believe about regression analysis and what regression and similar automated modeling tools can actually do for appraisers in real world situations.
Related Articles:
Author: David A. Braun, MAI, SRA (President, Braun & Associates, Inc.) has been actively engaged in real estate appraisal, review, and consulting since 1976. In this period, assignments have been performed throughout East Tennessee for numerous clients
This article was republished with permission from a la mode. The a la mode labs site and its contents are © copyright 2007 a la mode, inc.; but articles are © copyright 2007 their respective authors. Opinions expressed on the labs site are those of the authors only and do not necessarily reflect the views of a la mode. Return to the labs home page by clicking here. Click here to sign up to receive the labs newsletter.
Recent Comments