3.6 Crunch Time

[ilsindaco2628]

FRAME IT!!

A relic from the "Great 3.6 Catastrophe of 2026"

 

[Alsie35]

A relic from the "Great 3.6 ABOMINATION Catastrophe of 2026"

Fixed it for ya ...

 

[TerryRohrer]

3.6 isn't a theoretical concept, it's a form overhaul. 3.6 is the new form, a form which is supposed to present the data analysis in a credible and accurate report. A report that must be produced in a timely manner. Wasn't this the ultimate goal of 3.6? We've gone through the training, and not much has been said as to how the appraiser was actually supposed to provide the client with the completed product.

None of that relates to anything I had to say. To each their own opinion on the matter.

Our effective age comes from an analysis in which we take a significantly statical dataset (No less than 30 comparables) within a specific geographic/economic market. We then individually go through each comparable and assign an effective age based upon the observations and run it through a regression analysis program. Out of all the variables that we ran, only effective age consistently demonstrates a p-value that is relevant (<.05). We've have a couple of instances where we could use reproduce a coefficient for a garage or a bedroom, but have never successfully seen a relevant p-value from bathrooms.

That description seems to imply that you are doing a single-variable regression on each variable? Not as meaningful as multivariate regression. The other issue is that effective age, in this and almost every other approach, is being created and used as a fact, but without any basis in or weighting of how features impact prices. It transforms unrelated variables that might have dramatic differences in their impact on prices (probably related to cost) into a hard number that appears to be a measured, concrete fact.

From your other post, I would say a significant part of your problem is an insistence that only "comparable" sales be included in a regression ("to get to a homogenous data set"). Given that there is no universally agreed-upon measure or definition of what constitutes "comparable", there is no basis for that. The less variability there is among the sales you analyze, the less likely regression can perform...it is designed to explain variance, not the lack thereof. The old saying that the three most important aspects of real estate are location, location, and location is meaningful. Many appraisers are happy to go to a different area than the market area they describe as relevant to the valuation of their subject to find "comps" or derive adjustments, and ignore the difference in location, or ignore the fact that different amenities impact prices differently based on location, or simply make up a location adjustment to equate that irrelevant data to the relevant data. Efficiency is maximized by making up adjustments without analysis. Try doing multivariate regression of all or most of the sales in your subject market area. I almost always get useful results using regression in rural areas, and most of a group of 7 or 8 variables almost always have p-values of less than .05.

While the usual goal is a low p-value, it can be abused by being overly emphasized. A regression class I took 10 or 12 years ago (Regression Modeling: Why Bad Happen to Good Appraisers), they were targeting less than 10%. But, even if you are at 25%, what is the alternative? Run the model without that variable and if it makes no difference, leave it out. Otherwise, decide if an estimate, based on all the data you have available, is better than one from just 2 sales (paired) or one fabricated through depreciated cost.

As you see in this thread, there are those who have never used or tried to use regression who know all about why it can't work. There are also appraisers who have learned how it works, and when it works, and why it works, and understand how to achieve very good results with it in almost every type of market. It is funny that some who don't know what it takes suggest things like, "You really need 10 to 20 data points for everything variable.", when the usual rule of thumb by those with a clue is a minimum of about 5 datapoints for each variable in the model, with 10 to 20 being the typical preference. Yet, the same folks will make 10 adjustments based on 4 sales without any measures of reliability or significance and feel superior about their experience.

Another aspect being ignored by those without any knowledge of statistical analysis is multicollinearity. Almost every variable in real estate is related to another...number of baths or bedrooms or garages to GLA, etc. In a regression model, highly correlated variables might be indicated to be significant, even if the coefficients are not "usual". Their merit can be tested by leaving out one, then the other. But, if both are included, their "overlap" is negated. Yet, every day, appraisers are measuring adjustments from different data sets and different pairs and simply assuming there is no multicollinearity. Similarly, the relationship between GLA and prices might be linear, but it might not be. Paired sales analysis and sensitivity analysis of three sales in a table simply assume it doesn't exist. Efficiency is maximized by making up adjustments without analysis.

 

[djd09]

UAD 3.6 Appraisal Software: What’s Ready & What Appraisers Need to Know​

By John Dingeman

Lenders have been asking Class Valuation what appraisal software platforms support the new Uniform Appraisal Dataset (UAD) 3.6 standards, because their appraiser panels are asking the same question.

UAD 3.6 Appraisal Software: What's Ready & What Appraisers Need to Know

Explore UAD 3.6 appraisal software options and discover which platforms are ready for the new mortgage appraisal standards right now.

www.classvaluation.com

all the software companies were verified by the gse's...

 

[Doug Wegener]

Nov. 2 is the federal UAD 3.6 mandate- but the real deadline is October 15, lenders need time to process.------
at least according to the text I just received.

 

[Lil Rhody]

Nov. 2 is the federal UAD 3.6 mandate- but the real deadline is October 15, lenders need time to process.------
at least according to the text I just received.

anybody who orders conventional appraisals, should begin ordering them on or even before Oct 15th as UAD 3.6 assignments. The November 2nd date is in regard to UCDP submission date and is not related to the loan origination or the appraisal engagement date. If a lender engages an appraiser on say October 28th for an appraisal assignment, that appraisal needs to be developed and delivered to the lender so they can submit it to UCDP before Nov 2nd. Once 11/2 rolls around only appraisals that have already been submitted through UCDP in 2.6 format can be resubmitted in that same format. All others need to be in UAD 3.6 format.
I'm sure that there will be some appraisers that will accept an appraisal order before 11/2, complete the assignment on a legacy form and then have the client ask them to redevelop the report as a UAD 3.6. That could lead to an uncomfortable discussion about a fee increase.

 

[chad hampton]

I’ve now heard of three software companies that said to do this the right way it will take at least another year. They said more off the record regarding their thoughts on the GSEs so I don’t want to post that here.

 

[chad hampton]

UAD 3.6 Appraisal Software: What’s Ready & What Appraisers Need to Know​

By John Dingeman

Lenders have been asking Class Valuation what appraisal software platforms support the new Uniform Appraisal Dataset (UAD) 3.6 standards, because their appraiser panels are asking the same question.

UAD 3.6 Appraisal Software: What's Ready & What Appraisers Need to Know

Explore UAD 3.6 appraisal software options and discover which platforms are ready for the new mortgage appraisal standards right now.

www.classvaluation.com

all the software companies were verified by the gse's...

Class valuation is the last place I’d go for advice. Maybe just ahead of the GSEs.

 

[Fernando]

I remember in the last major change in the form, I would date my appraisal before the Fannie edict date.
If Nov 2 is to be the abomination day, then I would date my report like Nov 1. Would that work to continue with 2.6?

 

[TerryRohrer]

I remember in the last major change in the form, I would date my appraisal before the Fannie edict date.
If Nov 2 is to be the abomination day, then I would date my report like Nov 1. Would that work to continue with 2.6?

Try it and report back to us. Maybe no one will notice an effective date of November 15 with a report date of November 1. Or will you just lie about both?