Appraisal Statistics

[J Grant]

Buyers purchase properties as a whole, they don't purchase pieces of a property and add them together for a whole.

An appraisal models itself on what a buyer considers comparable alternatives to the subject and the behavior of the "typically motivated buyer " to find a price they would most probably pay. That price is related to an estimate length of time it would take a typically motivated seller to expose their property to the market to get that price .

Statistics uses math , though derived from real world prices, buyer decisions for res properties are not typically purely math based. Statistics needs a larger amount of data to work, meaning less similar sales are mixed into the data. Still, it works for what it is meant to do..

Imo statistics are best for time adjustments or sf ...but may be no better and perhaps worse than the simpler line item sensitivity analysis of comps for most adjustments.

Market experience lends perspective for an appraiser on an adjustment, or decision to not make one.) . No adjustment can be "perfect" , but they should make sense and be seen in the market .

Considering the non adjusted prices of the comps is a great way to mitigate the fact that adjustments are not perfect..

 

[Evincere]

The skirt length theory is a superstitious idea that skirt lengths are a predictor of the stock market direction. According to the theory, if short skirts are growing in popularity, it means the markets are going up. If longer skirt lengths are gaining traction in the fashion world, it means the markets are heading down. The skirt length theory is also called the hemline indicator or the "bare knees, bull market" theory.

The idea behind skirt length theory is that shorter skirts tend to appear in times when general consumer confidence and excitement is high, meaning the markets are bullish. In contrast, the theory says long skirts are worn more in times of fear and general gloom, indicating that things are bearish.

Although investors may secretly believe in such a theory, most serious analysts and investors prefer market fundamentals and economic data to hemlines. The case for skirt length theory is really based on two points in history. In the 1920s - AKA the Roaring Twenties - the economic strength of the U.S. led to a period of sustained growth in personal wealth for most of the population. This, in turn, led to new ventures in all areas, including entertainment and fashion. Fashions that would have been socially scandalous a decade before, such as skirts that ended above the knees, were all the rage. Then came the Crash of 1929 and the Great Depression, which saw new fashions dwindle and die in favor of the cheaper and plainer fashions that preceded them.

This pattern seemingly repeated in the 1980s when mini-skirts were popularized along with the millionaire boom that accompanied Reaganomics. The pendulum of fashion swung back to longer skirts in the late 80s, roughly coinciding with the stock market crash of 1987. However, the timing of these incidents, let alone the strength of the potential correlation, is questionable. Although there may be a defendable thesis around periods of sustained economic growth leading to bolder fashion choices, it is not a practical investment thesis to work with. Even benchmarking skirt length in North American would be a challenging undertaking. The time spent auditing clothing outlets to establish the length of top selling skirts would take more time than it is worth considering that it is far from proven as to whether the hemline indicator is leading or lagging.

Skirt length theory is a fun theory to talk about, but it would be impractical and dangerous to invest according to it.

https://www.investopedia.com/terms/s/skirtlengththeory.asp

Here is a classic statistical association backed by a proposed cause and effect.

http://www.dailymail.co.uk/sciencet...llection-revealed.html?ico=amp_articleRelated


Google's disturbing vision of TOTAL data collection: Leaked video reveals a Black Mirror-style future in which technology could be used to control the behaviour of entire populations.

 

[Terrel L. Shields]

People just don't buy property like that

Really? In aggregate they most certainly do or there would be no correlation in any methods we employ. All market would be random noise. Nevertheless humans employ certain heuristics to all decision making. While they do become irrational, especially under stress (like the great recession) all in all people make a judgment based on what a market is doing, how much inventory is available, what their budget is vs. what their needs are.

Do they do paired sales? NO. Do they do sensitivity analysts? NO. They determine value by their value in use. Gotta use for a 3 bedroom house? Then they are not interested in 2. Have bad knees? Gonna look at houses with stairs? NO. Quant that.

But the big overlook is the rational expectations of buyers regarding relative COSTS. Cost is by far the most under appreciated aspect of this. These buyers know what they want and have a relative idea of what it costs. They won't pay a premium for a wading pool you call an Olympic sized pool. The know the difference. They know the difference in value and cost between a simple small bathroom and a modern whirlpool tub and 5 fixture bath....but have you ever adjusted for bathroom size? Me neither. Same with appliances. Shops. Garages. Interior trim. Fireplaces. The buyer takes all that in in a qualitative way that ranks them by price, utility, and desire. Regression does that as well as paired sales and cost related adjustments are probably as least as accurate as paired sales despite the near phobia most residential appraisers have over anything related to 'cost'.

Finally quick down and dirty regressions are easily performed with a minimum of data. Done right they are no more time consuming than sensitivity analysis, paired sales etc. I've seen enough paired sales applied without extracting the land and or other major difference in these "matches" to puke a buzzard off a gut wagon. That is a complete hoax. And I bet half the appraisers I know guess at the SF adjustment because it "seemed" to work. And often they are not so far off as to make much difference. They just cannot document the adjustment.

If a larger regression analysis is performed normally that is simply building on an existing data set and so again hardly consumes the whole day. Face it. Appraisers are too cheap to do it right and as a consequence paid accordingly. I would bet 2 of 3 work files are created days before an investigator gets the file rather than created at the time of the report.

 

[Eli]

Buyers definitely look at cost for an attribute they like. If a property don’t have the attribute, they look at how much will it cost me to add it or is it even possible.

I use depreciated cost on many adjustments.

You can flip it too like an attribute that has reached the end of its useful or economic life and say what will it cost to replace it or remove it.

 

[George Hatch]

Everybody uses a valuation model, including the buyers and sellers who actually commit themselves to the decisions they're making with their own money.

 

[Randolph Kinney]

http://www.dailymail.co.uk/sciencet...llection-revealed.html?ico=amp_articleRelated


Google's disturbing vision of TOTAL data collection: Leaked video reveals a Black Mirror-style future in which technology could be used to control the behaviour of entire populations.

Google, Apple, Facebook, etc. track everything you do, websites visited, email, texting, posting on blogs, etc.

They do characterize you, your friends, your thinking, what you are buying, selling, looking at, etc.

Control of the masses would be for those that have an online presence.


https://www.nytimes.com/2018/05/16/...h_180517&nl=todaysheadlines&nlid=503291100517

 

[J Grant]

All these posts have an element of truth to them...

Buyer's main "valuation" model comes down to their budget range. Most buyers ( except the super wealthy) are shopping within a budget what they can afford./if financing what mortgage amount they qualify for.

Within their budget range, buyers try to get the best property they can , according to which features are meaningful to them. For some buyers that might be a bigger space, others want newer, others want a view or near urban downtown etc. Many buyers of course want a combination of features but still usually have a few that are critical to them . .

Valuation issues arise because within a similar price range in the same area, different buyers pay different amounts for features important to them , while other buyers will accept a defect to get something they want. A buyer might pay 400k for a 1600 sf older house needing work because it's a block from the beach, vs paying the same 400k for a newer, 1800 sf, upgraded property a mile away inland. If both of these houses show up in a statistical search of 1600-1800 sf properties within the one mile radius, they will give false results for the other. .

 

[Rlong]

I believe overall individual appraisers will benefit significantly from becoming MORE "Appraisal Statistics" literate, AND not perceive of them as primarily to achieve a final value conclusion (whether that be overall value, or an adjustment). Numerous better appraisers than I are proposing utilising quantitative analysis to INITIALLY examine the market to "see what the data might tell you" that you might miss by a typical perusal of MLS sheets. Even if we reject that (or are at, or on the verge of being to old or entrenched to want to learn how, which may include me!) appraisers will benefit from learning the vocabulary and theories when... we end up "FIGHTING THE ROBOTS" that intend to take our jobs!

Fight the Robots!

 

[George Hatch]

Buyer's main "valuation" model comes down to their budget range.

That's how buyers decide what their price range is, but it's not how they actually value the home they're buying. Unless they're letting their broker do all their decision making for them what most buyers are doing is a qualitative analysis among however many prospects they've looked at and ranking the subject in the range. Which is generally what most appraisers (except the quants) do after they get done *refining* (aka adjusting) their dataset. We will rank the subject within that range, the main distinction being that the ranges we end up with are narrower than theirs'.

As for what 1000 other buyers who are buying more disparate properties are doing, that's great; but I don't consider the results of the macro to be as relevant to THIS dataset and what these sales demonstrate when compared to each other. These buyers aren't competing with the entire town - they're only directly competing with each other.

The most defensible adjustment is the one you don't have to make.

 

[RCA]

I think the most such tools out there don't do the job, except with respect to homogeneous areas. That's because they all use linear regression. My understanding is that - that is because they find it toooo difficult to teach anything else to appraisers. Maybe that was true 10 years ago -but maybe no longer, not at this time. The Millennials are coming along and they can supposedly pick up this stuff much easier. Anyway, most of the areas I've worked in, perhaps because I tended I think to get more complex properties other appraisers didn't want to mess with, did not lend themselves to linear regression. That is to say, you would often get R2 values in the 0.30-0.40 range. Or, a bit more specifically, the function given for predicting the contribution to value based on living area only accounted for 30-40% of the variations in value due to GLA. You are still left with trying to figure out the difference; and, if you don't have many comps you can just live with a broader range of adjusted comp prices. But often, more to the point, to get rid of that variation, you try to narrow down your set of comps - but when you do that, you loose the intelligence you could get by analyzing more properties. You will loose getting adjustments for other features. To try to be more clear: If my subject is 2800sf and I stay within the subject subdivision, the contribution to value by living area is pretty simple. If if there are only 3 comps within the past two years - I really want to go outside of the subdivision to get adjustments for market conditions (time), bath count, lot size and so on. But if I do go outside the subject subdivision, my linear GLA contribution function most likely will become non-linear because the price/sf is different between different subdivisions for different ranges of homes. Usually each subdivision focuses an a certain set of GLA and lotsize ranges; often times the breaks in the base functions (or line segments or knots) relate pretty directly to the subdivision, with some overlap). A neighboring subdivision my have houses only in the 1800-2400sf range, but give me some better data on market conditions and other features. An advanced regression program like MARS is searching quite intensively for ways to explain the variation in target variable (usually sale price for appraisers). Give it enough data and it might tell you that tile roof adds so much per sf of GLA to your price - regardless of the subdivision - or it may might even give you an adjustment based on a combination of [GLA, roof-type and subdivision] - although that is getting to be too much. Note, that of course total SF (GLA+GARAGE) would be better than GLA in this latter case.