How Many Have Figured Out

[Randolph Kinney]

I think most of us know appraisers well enough to understand the hazards of hype.

BTW, in response to the OP's question: "How many of you actually have figured out, i.e. understand, how I adjust comps? " I will just respond - not me.

 

[Elliott]

So Bert,
What a single car compared to a double car garage adjustment. What's your fireplace adjustment? How much do you adjust for an age of 12 years compared to 16 years?
I promise I will give you my adjustments if you give me yours.

 

[Tim The Enchanter]

Light bulb says,
Bert is a Bot?
I hope not.

 

[Artemis Fowl]

In a nutshell...you isolate the influence of features common to all properties in the sample (call it a base value) and adjust using the residuals via subjective ranking. I think you mentioned another RA analysis using the residuals prior to ranking. YES?NO?

 

[RCA]

My market area does encompass very high end properties and I have appraised them. The beauty of the existing appraisal methodology is it can be applied to all types of properties and price ranges in all kinds of diverse regions and areas, including nationally and internationally. Whether an appraiser has the competence for an individual assignment is another matter. In high $ properties there are more variances and the scale of adjustments is larger. For that reason, typically a lender will order 2 appraisals from 2 appraisers on these properties. Some variance between the appraisals is acceptable but if there is a wide value opinion or other opinion divergence then the appraisals are reviewed and perhaps even a third ordered.

Appraising involves human judgment and human perspective. That is both its strength and its weakness, it is either accepted as a credible model of value or it is not. Appraising property, or appraising anything , is never an exact science, it is supposed to be supportable and credible for how the value equivalence to price (usually market value ) is derived.

In virtually all appraisal, art, jewels, real estate, antiques, horses, boats, historical records of what others have paid for similar/equivalent is the basis for the hypothetical what would they most probably pay for X ( X is the subject)

Imo, AVM's or statistical value models best serve as a check and balance to an appraisal, rather than a substitute for an appraisal. I understand lenders already do that, as does Fannie Mae.

"High-end" does not equate to complex. High-end subdivisions (and we do have them here) can be pretty easy to appraise, as long as there are some sales around. In fact in some areas, where people have money and do a lot of upgrades and remodeling, it is in fact often the homes at the lower end of the totem pole, for which there are no similar comps and all the other comps are so absolutely different, you wind up in no mans land: Estate appraisals of homes that haven't been remodeled in decades (other than say have the roof replaced), in areas where remodeling, especially for sales, is the rule.

The fact that you bring this up, indicates that you don't understand the issue I keep explaining about California. I would think you must have a few of these places around Florida. Maybe not. When I get a chance, I must get into one your MLSs and study the market, as I am getting curious. Parts of Houston are interesting too, like West University Place - particularly interesting.

 

[RCA]

Data qualification is where its at, regardless of which valuation model you use. The 300-unit subdivision scenario with few variations among the individual datapoints simplifies that task to the point that any valuation model will work. So simple even a monkey and a dartboard will work.

Conversely, the use of raw and unqualified data complicates the task; and using larger quantities of raw and unqualified data is only incrementally better.

We already have a ton of experience using an appraisal valuation model (which is what the Fannie grids are), and that model requires the appraiser to spend a considerable amount of time and effort to qualify the data they're using. If you had to break the amount of "appraisal development" time by activity, the analyses/comparisons would probably amount to less than 5% of the total. The other 95% is split between data indentification and qualification. THAT's where most of the appraising occurs, not in the last 5 minutes when we're refining the value range indicated by the most direct comparables that we're presenting. If we made zero adjustments to those comparables we could still emulate what the buyers and sellers are doing in real life because that's what they do.

So really, what the quants are trying to do is to refine the adding machine part of the analysis, which IRL is usually the fastest and simplest step in our process.

It doesn't matter how much time/effort you spend on refining the manner in which you develop and apply line-item adjustments because all that happens with a less-than-optimal combination of adjustments is that the appraiser ends up doing more qualitative instead of less qualitative in their final reconciliation.

ZAIO had a plan where the appraiser nominally qualified all their data prior to using any of it in their AVM. The fundamental concept is certainly valid but the manner in which they were doing it was inefficient. If they had limited their pre-qualification protocols to the properties being listed in the MLS they wouldn't have been wasting so much time/effort.

It is generally true that when dealing with data analysis, a lot of work goes into cleaning the data and data preparation. Like I stated in another post, the pro.mlslistings.com broker database has 400+ standard fields for residential properties, but many of these such as Heating, Kitchen Description, Sewer, are lists of numbers that you have to reference in lookup tables. There are 4500 such entries and you can expand those out into Y/N fields in your grid, giving you 5000 or so fields that you then run through your regression software.

Oh, but then I forgot to tell you, the MLS removes the GPS Longitude/Latitude and removes any census data, - if it is there. So, you need to get that data and merge it. And you are still not done. ....

And with all of that, many features are meaningless value contributors by themselves. In combined groups, they become important, such as a high end kitchen with numerous expensive appliances, a good layout for cooking. While all the things in this kitchen are "real", their combined value is an intangible. Why? Well, for one thing, you are not likely to find any matched pairs. Unless you are in a subdivision, every high end kitchen is different and cannot be compared 1-1 in its entirety. To get at that value, you have to be clever.

 

[RCA]

So Bert,
What a single car compared to a double car garage adjustment. What's your fireplace adjustment? How much do you adjust for an age of 12 years compared to 16 years?
I promise I will give you my adjustments if you give me yours.

When I started appraisal as a trainee in 2002, my mentor gave me adjustments for these things. Like $25/sf when it cost $200/sf or more for construction. Of course, if you have perfectly balanced comps, it doesn't really matter. Or if the comps have the same square footage. Hmmmm. Sounds like tract homes they give trainees. .... Are you still back the horse and buggy days? Really?

 

[Tom Woolford]

How many of you actually have figured out, i.e. understand, how I adjust comps?

I consider the method very clever, if I don't say so myself, on par with Eratosthenes indirect measurement of the circumference of the earth 2300+/- years ago.

I was discussing this with one of the statisticians over at Salford Systems, and I'm not sure he understands it yet. But, then he is not an appraiser.

The clue is I use statistics to predict the value contribution of the tangibles, subtract if from the actual sale price to get an indirect measure of the intangibles. I do this for maybe 300 recent sales in a neighborhood. I rank the residuals form lowest to highest and give them scores 0.0-10.0 in 0.5 increments reflect the percentage of residuals of lower value, using an Excel macro, then replace the macros with the actual scores (so I can re-sort and they won't change), then create a function that maps the scores to the residuals and thus, all comps in your grid have the adjustments for tangibles based on the regression model, and a total adjustment for all intangibles, based on the difference between their residual score and the score the appraiser gives the subject for percentage ranking of the subject compared to all of the sales in the neighborhood. You can split that score adjustment however you want, but it doesn't change the total adjustment for the intangibles.

The method I use, can be used as a general method for measuring the value of almost any object with both tangible and intangible features.

In other words it is an exact measure of the total worth of the intangible features of a sales comparable; but done in such a way that the appraiser has things set up to do a fairly precise and meaningful estimate of the value for the subject property's intangible features. And, simple math says, you you can split those total adjustments up however you want between the condition, quality of construction, functional utility, view, aesthetic appeal - it won't make any different on the total adjustment to the subject.

Plenty of clues. See if you can understand the method.

No, I don't understand it. Why don't you write a paper, submit it to the AI for peer review. All you have presented here comes off as boasting about how smart you are.

 

[RCA]

No, I don't understand it. Why don't you write a paper, submit it to the AI for peer review. All you have presented here comes off as boasting about how smart you are.

Of course. If you read any of the existing books on the Sales Comparison Approach you will find they are a rehash of the same ol' same ol' that's been out there for decades.

Except you may find some simple linear regression statistics. Parametric no doubt.

The AI statistics courses are pure parametric statistics. In non-parametric, we tend not to make assumptions about underlying populations distributions, in particular anything like the normal distribution. Real estate is non-parametric. - In non-parametric, we use median rather than average. There is rarely talk about standard deviation or variance. We use binning and other techniques. We use non-parametric regression. And that works really great with the small sample size we get with certain kinds of property. We assume as little about the distributions as possible. Our distribution may be lumpy, like when you cross the boundary into an area with a superior school district. It's a different world.

There are no peers in this world --- yet --- well in any case they are certainly few and far between. Hahn should publish a paper and tell us what he is doing or Dell. .... But I haven't looked at their latest output.

And this is the fact. I am pretty sure they will say, like some others I know, - there is no way your average appraiser is going to even begin to understand this stuff.

But, it doesn't hurt to try ......

 

[RCA]

J


Just remember, in anything that you do in any appraisal, in the event that your appraisal is reviewed you will be accountable for what your peers would do under the same circumstance. Is your methodology accepted practice? Good luck, I seriously doubt you could get past any Daubert-, Robinson- or Frye-type motions if you ever ended up in court.

USPAP Page 13 Lines 378-387

SCOPE OF WORK ACCEPTABILITY

The scope of work must include the research and analyses that are necessary to develop credible
assignment results.

Comment
: The scope of work is acceptable when it meets or exceeds:
• the expectations of parties who are regularly intended users for similar assignments; and
• what an appraiser’s peers’ actions would be in performing the same or a similar assignment.

Determining the scope of work is an ongoing process in an assignment. Information or conditions
discovered during the course of an assignment might cause the appraiser to reconsider the scope of work.

An appraiser must be prepared to support the decision to exclude any investigation, information, method, or
technique that would appear relevant to the client, another intended user
, or the appraiser’s peers.

Of course. But so far I haven't come across a single appraisal or method that works in most of the market I am in, unless there just happen to be good comps. The typical support is not in the report. If I can believe others, standard linear regression is used, averages taken, but in difficult situations there is to date no solid approach. If you know one, please present it.

The problem with your comment, is that if someone were to attack the method I use, they should be prepared to logically support their adjustments ... in some very difficult situations.

I would appreciate very much if someone could point out the best method available that works in the SF Bay Area for custom homes and provides %5 accuracy. I have open arms to new methods, if they have any salt.

Let me elucidate a bit. I have been in arguments with people about using linear regression. They keep insisting on using sales from homogeneous subdivisions in the mid-west - and just turn their head the other way and pretend the markets here in SF Bay area don't exist. I don't think such people will get anywhere in court, if they can't answer direct questions. You can go up and down the line, and as far as I can see it's the same story.