Twenty Seven Billion Buck Risk...You

[Dublin ohio]

Appears AI is considering condition and quality ratings as absolute, which in the real world is not the case

 

[glenn walker]

Agree not absolute but trying to make them to be.

 

[Fernando]

Appears AI is considering condition and quality ratings as absolute, which in the real world is not the case

AI can't think like a human appraiser. It's a warning for lenders relying on computers. Our jobs are safe for the time being.

 

[Tawfik Ahdab]

Let’s not forget it’s fFannie and Freddie who purport that their ratings are absolute.

Aside from the GSE cheerleaders in this forum, there doesn’t seem to be anything like agreement about this among practicing appraisers.

 

[Fernando]

Let’s not forget it’s fFannie and Freddie who purport that their ratings are absolute.

Aside from the GSE cheerleaders in this forum, there doesn’t seem to be anything like agreement about this among practicing appraisers.

What absolute?
Even when C3 between subject and comps, appraisers can still make adjustments. And different adjustments for different comps.
As Trump would say, 4D thinking.

 

[CGinMN]

Looks like with UAD 3.6 quality and condition rating descriptions have been updated with detail to make it more “objective.” Not sure which descriptions the AI was trained on, the old ones or themes expanded ones.

https://singlefamily.fanniemae.com/media/42396/display

A couple problems: Appraisers rely on direct observation (not photos) to determine subject property condition and quality. For comparables, appraisers deliberately go beyond biased marketing photos by looking at disclosures, prior listings, permits, and conducting agent interviews. So to say that computer vision is better because it looked at some selectively framed and edited photos is misleading. Also, adjustments for condition and quality aren’t always made in the grid, but are considered through reconciliation. The study makes no attempt to analyze price data to validate whether their ratings result in a better valuation. So even if the AI can develop what it believes is a more precise rating, if the market doesn’t adjust for the difference and it agrees with the appraiser, then is it actually any better?

I suspect in the future appraisers will get a lot of automated stips because of this.

 

[Mark K]

All of this analyzing and dissecting appraisal reports is nothing more than additional GSE ammo in the war on appraisers in order to remove them from the process. They want to "prove" that appraisers don't have the credibility or sufficient objectivity to evaluate a property.

However, if they would spend half as much effort using AI to determine who is a good credit risk, this wouldn't be an issue. But F/F, et.al., will twist themselves into a pretzel to find a way to loan to anyone with a 500 FICO, no money down, buying a house that costs 10 X their annual income, and six jobs in the past six months.

But...appraisers will be the cause of the next meltdown.

 

[Terrel L. Shields]

No **** 86% are c3/c4. I’d guess 10% are c1 and the last 4% c2

Looks like a standard bell curve that's a bit center skewed to me... I mean since C6 are rejects that don't make the data sample, you will expect the center to "flatten the curve" a mite.

 

[Terrel L. Shields]

Appraisers rely on direct observation (not photos) to determine subject property condition and quality.

To me, quality relates to the original cost (assuming no over-priced builders, no botched plans and starting over, etc.) and that is very subjective unless we have the original cost bids for a property. Condition? We can use the comps as proxy for the subject as long as we are confident the comps are accurate. By extracting the effective age (a concept apparently FNMA wants to do away with) mathematically, we have a defensible number. At least compared to the observed effective age where we simply say "That looks like a 15-year-old house." That can be misleading when a home is obviously from another era (like a 1930s craftsman remodeled with new appliances, floor covering, paint etc.) It still looks like a 100-year-old design. So true "comps" need to be similar remodeled craftsman style, not so much just any old house with similar effective age.

 

[djd09]

the problem is you got some houses with remodeled kitchen and flooring....and others with updated bath and furnace...both rated the same but not quite equal stuff that in your pipe borg