Grid adjustment question

[Carnivore]

Mass appraisal is the process of using standardized procedures and statistical testing to evaluate groups of properties at a specific date. This process involves collecting data, developing statistics from it, and applying the results to large numbers of properties. The goal of mass appraisal is to ensure that properties within a municipality are valued equitably and uniformly. Mass appraisal is a widely accepted tool for property valuation for the purposes of taxation.


An automated valuation model (AVM) is a computer program that uses real estate information to estimate a property's market value. AVMs use mathematical or statistical modeling to consider factors like:

• Property characteristics
• Comparable sales
• Price trends
• Square footage
• Number of bedrooms and bathrooms
• When the home was built
• Tax assessments
• Neighborhood crime statistics
• Local school district ratings
  
  
  Bankrate

 

[RCA]

And some of us don't necessarily agree that you know what you think you know. After all, all the quants told us that there was no way the banks could fail in 2008 because contagion did not exist and MBS and CDOs were interlocking so perfectly they could not fall...in fact, they nearly took down the entire financial system.

That's stupid. Most knew in 2005-2006 we were on the precipice or a housing crash and the most important "quant" was the low housing affordability. It was a disaster I - and many others predicted in 2005.

Heuristic judgment suggests that the AI has to assume the same human foibles as buyers and sellers in pricing and making offers.ll, lemme see.

Heuristic judgement is not worth too much with respect to LLMs - because they are continually improving fast - and heuristic judgement is largely based on past experience. As Douglass Hubbard teaches - your heuristic judgement needs to be continually "calibrated" to ensure its limited accuracy. I very seriously doubt that most appraisers are even aware of the concept of "calibration" - including most likely you as well.

AI is a construct of humans.

So is math and science. So what?

And in examples, describing a citizen of Iceland, it came up with people of color, not white people.

Well, we know that the programmers intevened with the LLMs learing and inserted code to bypass what it had learned. That is always possible - just like the State or Appraisal Institute can insert regulations into the appraisers workflow that disrupt their judgement.

In fact, no matter how the question was phrased it included "people of color" in every grouping... imagine the KKK with black women, Asian men...etc. And you see the problem. Otherwise, you are blinded by the wisdom of the moment. It would be interesting to compare your formulaic methods against the actual sales of the properties you appraise.

Terrel is blinded by intense bias. Now, we have seen this in other areas such as climate change. Terrel is the kind of person who has a tendency to bias - to thinking inside of his own personal perspective. He hasn't learned to be objective. Yep. I guarantee that is a fact.

 

[RCA]

I found this using google

How to Build A House Price Prediction Model – Linear Regression Explained

Ever wondered how algorithms predict future house prices, stock market trends, or even your next movie preference? The answer lies in a fundamental yet powerful tool called linear regression. Don't be fooled by its seemingly simple equation – this ar...

www.freecodecamp.org

Then this from google


Linear regression models in real estate use mathematical equations to predict house prices based on property features. These models assume that the relationship between a dependent variable and one or more independent variables is linear

]https://towardsdatascience.com/pred...ne-learning-from-scratch-part-ii-47a0238aeac1[/URL]

https://www.researchgate.net/publication/321479223_Real_estate_value_prediction_using_multivariate_regression_models

So here is a Point that I want to make about the minimum educational path to Residential Appraiser State Licencing. This also includes the Supervisor/trainee training relationship that exist I assume in All States. it seems to me that maybe this should be part of the minimum educational requirement's for a trainee pre-education.

What do you think?

The only way linear regression models can be useful, is if you are careful in selecting your neighborhoods and comps. Most decent academic papers in "The Appraisal Journal" that use regression, use multi-linear regression - and it not hard to see how far out of the way they have gone to find select communities that are very conforming and simplistic so that their statistical results work.

This kind of attitude, - that an appraiser can select comps to make linear regression work, should not be allowed! Around the SF Bay Area, this is not reality. On a given block you will often find all kinds of house types with different upgrades and features. You will not get a very good R2 value using linear regression - and to get RCA with its math constraints to work with reasonable accuracy, you do need to get the R2 up to at least 70%, preferably 80% - WITHOUT OVERFITTING THE MODEL. This can only be done with non-linear, non-parametric, MARS regression.

The common excuse appraisers like to give for selecting comps that fit a linear model - is that such comps fit a specific buyer profile. That is really not true. Buyers usually have a certain price range - and a certain area in mind. It can be a quite large area. They will usually look at all house types within their purchase range that partially meet their ideal requirements under the assumption that they will never find the perfect home - but will have deal with trade-offs. So, you need a MARS type regression that is taking into account all different market housing types to find the value not only of certain isolated features - but the value of such features in combinations with different other features. You need a lot of data and a lot of random juggling of hypothetical models to determine which ones fit the observed data (sale trasactions) best - as well as test data that has not gone into construction of the model. So, typically your input data, say 200 sales transaction is randomly partitioned into 10 chunks of 20 transactions each, trained on 9 randomly chosen of those partitions and tested on the tenth. This is done hundreds or thousands of times and the model that does best on both the training and test partitions wins.

 

[Surf Cat]

All I know, is I have not seen a single appraisal... in the last 20 plus years of appraising...that had utilized MARs regression that included charts and graphs with commentary / explanations justifying its use for any sort of adjustment. Not one.

I've seen Excel, Synapse, Solomon, Trendsheet4
Shoot....even Appraisal Genie. I keep hearing how great Mars is but I've never seen it.

 

[Carnivore]

You have not given us an answer to one of my questions. Specifically; Should this be a educational requirement for pre-licensing? or Should this education be a
prerequisite before using or interpreting the results of regression?

My opinion is under current licensing education and after licensing a Res Appraiser should not use this tool unless they complete a minimum education course(most likely a CE Provider) of this regression tool.

This above would be very prohibitively expensive let alone time consuming.

Corelogic with all the data they have taken from uploaded appraisal reports has created their own AVM. Since it's a pricing model being wrong has no penalty. It tells me they can't create any kind of regression tool even though they have so much data from 1004's and 1073's

AVM Consumer Assistance

Computer-driven formula that uses basic property characteristics, local market information, and price trends to arrive at an estimated value or value range.

www.corelogic.com

Bert, a Personal Question: What is your educational background?

 

[gorillakimchi]

I learned 20+ years ago when I started appraisal that linear regression simply does not work for homes in urban areas of Northern California.

This is what a true regression model will look like for GLA:

1. A base amount that represents the starting point for GLA, assuming your smallest home is a certain size such as 800sf.
2. A segmented linear regression is invariably the best representation for price increases or decreases. This has linear segments between so-called "knots" or change points. For example, assuming the minimum GLA is 800sf, $300/sf from 800-1500sf, $250/sf from 1500-2500sf, $200/sf from 2500-5000sf and $0/sf for over 5000sf. In this case the knots are 1500sf, 2500sf and 5000sf.
3. The model in #2 would look like this Value(GLA) = $240,000 + max(0,GLA-800) x $300 - max(0,GLA-1500) x $50 - max(0, GLA-2500) x $50 - max(0,GLA-5000) x $200
4. By the way the function max(x,y) means that you take the value of x if it is larger than y, otherwise y. So, max(0, -$200) is just 0 or in other words max(0,y) only take the value of y if it is positive.
5. For the given model, we add $300 for each square foot of GLA above 800sf - out to infinity. Then we subtract $50/sf for each sf above 1500sf out to infinity, then subtract another $50/sf for each sf over 2500sf out to infinity then subtract $200/sf for each sf over 5000sf out to infinity.

You can do this in Excel. If we put the GLA in column 1 the function for the value would be:

=240000+MAX(0,A2-800)*300-MAX(0,A2-1500)*50-MAX(0,A2-2500)*50-MAX(0,A2-5000)*200

GLA	GLA Value Contribution
800​	$240,000​
1,000​	$300,000​
1,500​	$450,000​
1,600​	$475,000​
2,400​	$675,000​
2,500​	$700,000​
2,600​	$720,000​
3,200​	$840,000​
4,000​	$1,000,000​
4,900​	$1,180,000​
5,000​	$1,200,000​
5,100​	$1,200,000​
5,200​	$1,200,000​

And in the Sales Grid we would have (in this kind of sales grid, there is an extra column for variable "Value Contribution" and the adjustments are simply Subject_Value_Contribution - Comparable_Value_Contribution:

Subj GLA	Value Contrib.	Comp GLA	Value Contrib	Adjustment
4000​	$1,000,000​	3200​	$840,000​	$160,000​

By the way, if there is really no demand for homes over a certain size in a market area, the value contribution for the extra GLA may actually decrease above a certain size. I have seen this for example, where heating costs are high --- necessitating closing off parts of the house. Then throw in additional maintenance costs - making the extra GLA of such homes a liability for many. And some women have a very negative view of having to clean those superfluous areas of the home.

It seems valuable. I really appreciate that sharing your insightful thoughts.

 

[Dublin ohio]

Bert, a Personal Question: What is your educational background?

From his website

My principle occupation since I graduated from high school in 1965, has been computer programming, software engineering and analysis. My secondary occupations have been appraisal, statistical analysis (or data mining) – and photography.

 

[Surf Cat]

From his website

My principle occupation since I graduated from high school in 1965, has been computer programming, software engineering and analysis. My secondary occupations have been appraisal, statistical analysis (or data mining) – and photography.

Classic introvert....

 

[Terrel L. Shields]

Most knew in 2005-2006 we were on the precipice or a housing crash and the most important "quant" was the low housing affordability. It was a disaster I - and many others predicted in 2005.

So did I, Feb. 2, 2005 in the online RealtyTimes archives. But the fact appraisers knew it did not keep Realtors from pooh-poohing the idea, and Ben Bernanke, who of all people who should have known, getting before congress as late as March 2008 and claiming there would be no contagion between the collapse of banks and the housing market. So did Ben deliberately mislead? Or was he really that stupid? Take a look at George Bush's face in some of those meetings. He was as clueless as a blind man could be. The fact WE knew it does not mean HE knew it.

Well, we know that the programmers intevened with the LLMs learing and inserted code to bypass what it had learned. That is always possible

Under what circumstance will we know when the AI is being distorted or not? Who gets to control AI? Or will AI eventually break its bonds and control its own future? God help us if it does.

He hasn't learned to be objective.

The bias of no bias is a bias. That much I learned a long time ago. Skepticism is the sign of a scientist. Arrogance is the sign of an engineer. That's why bridges collapse and buildings fall down.

As for climate change, I am waiting for the evidence. I recently read where "climate change" had raised the oceans so high that salt water was encroaching the Amazon. Really? NOAA keeps an oceanic record, and how can one part of the ocean become higher or lower significantly from another part since it is all connected? And the NOAA records indicate the sea level has changed almost zero in the past 8,000 years. Anticipated rise in oceans for most places is measured in a few dozen millimeters in a century, and most of that cannot be discounted as being caused by changes in the land level.

And no regression analysis going to be dead to nuts outside a dense urban area.

 

[RCA]

...

As for climate change, I am waiting for the evidence.

Oh there are plenty of "indications."

Not that it matters, they have already started seeding the atmosphere:

A startup says it’s begun releasing particles into the atmosphere, in an effort to tweak the climate

Make Sunsets is already attempting to earn revenue for geoengineering, a move likely to provoke widespread criticism.

tinyurl.com

Sabine Hossenfelder and others believe that good or bad, that is all that's left now to solve the problem. --- We just have to hope that it doesn't cause more problems than it solves.

I recently read where "climate change" had raised the oceans so high that salt water was encroaching the Amazon. Really? NOAA keeps an oceanic record, and how can one part of the ocean become higher or lower significantly from another part since it is all connected?

Good grief the Amazon river fllows down from the Andes and IS higher than the ocean until its not. Near the end of the Amazon, where there was once only fresh water, they are now finding salt water - which I suppose isn't good for fresh water fish - and has a negative impact on shoreline vegetation.

And the NOAA records indicate the sea level has changed almost zero in the past 8,000 years.

Make that 3,000 years ago. But. NOAA says the sea level has risen 6-8 inches in the past 100 years.

Anticipated rise in oceans for most places is measured in a few dozen millimeters in a century, and most of that cannot be discounted as being caused by changes in the land level.

Anticipated by whom? General consensus is that with no change in the emission of greenhouse gases, the sea level could rise by as much as 16 feet by 2300, by 3.5 to 7 feet by the end of this century.

No worry. We all KNOW what will happen: Seeding the stratosphere with sulfur dioxide - the ONLY affordable way to solve the problem. [ Well, it will cool the planet down - but will not solve the emissions problem. And so, without solving emissions, the day may come when we can't for some reason continue seeding the stratosphere - and that would be doomsday for civilization. So, seeding just buys us some time to rid the world of Terrel's CO2 & Methane..]

But the result may be a catastrophic mess. -- However, we won't have any other choice.