20 Minute AI Appraisals Are Coming

[glenn walker]

Buyer's using an appraisal contingency of course should pay for a full appraisal and on waivers the borrower has the option of NOT using the waiver and paying for a full appraisal.

For the others on purchases no appraisal should be required. The buyer and their agent and their home inspector knows the condition etc. Todays buyers and sellers are pretty good at narrowing final prices down to reasonable ranges of value say 2% to 5% or less.

That's also because today buyers have access to comparable homes interiors on sites like Zillow Red- Fin and others so today it's a much more informed buyer and seller than ones 10 to 20 years ago.

In summary outside of custom or complex properties, most don't need traditional full appraisals. A menu of waivers, desktops, AVM, and full 1004 should be selected by the borrower's
and the lender to see which fits best.

 

[RCA]

You as an indivaul can pay whatever you want
An appraiser is supposed to derive market value. If that differs from a SC price, so be it. Have never read in UPAP where it is about a sale prices being "supported," -

If you are estimating market value, you are estimating the average expected net sale price from a random selection of buyers in the market as of a certain day. Your market will have a mix of buyers in terms of their interest in the subject property. For some buyers the subject will be a good match for others not so much. At the same time, there are likely other properties competing with the subject. Buyers look very carefully at features they're interested in; especially if there are alternative homes and they are having a hard time deciding which to buy. This means that details could be important. Also, you don't know who exactly who the potential buyers are going to be. These aren't the only problems, there are many more. So, you need a detailed price model that covers all features that are significantly important to the potential buyer pool. The best evidence you have are property characteristics and past sales. The best tool around for extracting this information is MARS. ...

 

[J Grant]

Buyer's using an appraisal contingency of course should pay for a full appraisal and on waivers the borrower has the option of NOT using the waiver and paying for a full appraisal.

For the others on purchases no appraisal should be required. The buyer and their agent and their home inspector knows the condition etc. Todays buyers and sellers are pretty good at narrowing final prices down to reasonable ranges of value say 2% to 5% or less.

That's also because today buyers have access to comparable homes interiors on sites like Zillow Red- Fin and others so today it's a much more informed buyer and seller than ones 10 to 20 years ago.

In summary outside of custom or complex properties, most don't need traditional full appraisals. A menu of waivers, desktops, AVM, and full 1004 should be selected by the borrower's
and the lender to see which fits best.

I personally do not think Zillow and Red Fin have made buyers smarter, since both those sites were around during the boom when people paid crazy prices for homes - the borrower does not choose the valuation, the lender does - if a borrower pays cash they can pay whatever they want and do whatever they want.

 

[J Grant]

In the future, and perhaps near future , AI can integrate with applications in sections for analysis, which would be a plus. The programmer would have to teach AI what pattern to look for in data,- not an easy task since data changes constantly and is different in each apprasisal.

 

[J Grant]

If you are estimating market value, you are estimating the average expected net sale price from a random selection of buyers in the market as of a certain day. Your market will have a mix of buyers in terms of their interest in the subject property. For some buyers the subject will be a good match for others not so much. At the same time, there are likely other properties competing with the subject. Buyers look very carefully at features they're interested in; especially if there are alternative homes and they are having a hard time deciding which to buy. This means that details could be important. Also, you don't know who exactly who the potential buyers are going to be. These aren't the only problems, there are many more. So, you need a detailed price model that covers all features that are significantly important to the potential buyer pool. The best evidence you have are property characteristics and past sales. The best tool around for extracting this information is MARS. ...

I respect what you are doing with MARS. And entrepreneurs keep coming out with all kinds of statistical programs.

But the best tool around is the person- the appraiser who has the market knowledge and field experience and can apply it to the appraisal process. No tool can replace what a human brain was engineered to do - synthesize various moving and at times contradictory elements, analyze, and think a problem through. No machine can think;like a human, not even AI trained to replicate thought.

MARS or other programs are superior in their ability to quantify larger data sets. For qualification, of the data, a deep dive into the comps and motivations of parties, a knowledgeable person does it best because a person is going to use the product and take out the 400k loan or lend the $ on the collateral. Markets are comprised of buyers and sellers, some honest, some not, some well informed, some not, some rational, some not, who make decisions that become the various prices.

The reason appraisal is an enduring product is that the MV definition cancels out the noise of the various individuals out there who arrive at different prices. We use their prices of course, but filter it through the assumed hypothetical model "Sale" as of X date. How close that most probable price is to MV is the result of the appraisal opinion -

 

[glenn walker]

Mar's biggest problem is cost and that
it's not accepted by the profession as anything but a sophisticated software platform and not a recognizable substitute for traditional appraisals.

The appraisal-- institute hasn't recognized it nor has the TAF or USPAP. They would consider it as a AVM.

Therefore it has no value to individual fee appraisers or even large institutions like Banks and GSEs because they have their own team's of technology engineers and quants on staff.

Companies like Corelogic and Title Companies already have the systems in place.

 

[RCA]

I respect what you are doing with MARS. And entrepreneurs keep coming out with all kinds of statistical programs.

But the best tool around is the person- the appraiser who has the market knowledge and field experience and can apply it to the appraisal process. No tool can replace what a human brain was engineered to do - synthesize various moving and at times contradictory elements, analyze, and think a problem through. No machine can think;like a human, not even AI trained to replicate thought.

MARS or other programs are superior in their ability to quantify larger data sets. For qualification, of the data, a deep dive into the comps and motivations of parties, a knowledgeable person does it best because a person is going to use the product and take out the 400k loan or lend the $ on the collateral. Markets are comprised of buyers and sellers, some honest, some not, some well informed, some not, some rational, some not, who make decisions that become the various prices.

The reason appraisal is an enduring product is that the MV definition cancels out the noise of the various individuals out there who arrive at different prices. We use their prices of course, but filter it through the assumed hypothetical model "Sale" as of X date. How close that most probable price is to MV is the result of the appraisal opinion -

You seem to think that MARS replaces the appraiser. When I use MARS it is just a tool like linear regression or a "method" like matched pairs analysis. I still do inspections with measurements per ANSI specifications. So your above arguments don't make any sense. MARS creates the model that results in adjustments and RCA applies constraints to the total adjustments allowed for subjective measures on features lacking measurement such as condition, quality, function utility and view. You would have to understand the math and logic behind these methods. They are a burden EXTRA to your usual work as a traditional appraiser. You can't make much of a living with MARS and RCA unless you are charging over $1.5K per appraisal or have a very efficient system set up, and some kind of guarantee you will be getting at least 2 appraisals/week on average at $1k-1.5K each. The reason I am doing them is more or less R&D. Although I did occasionally get SFR assignments for $1.8K - often from architects who wanted a super good value on some house they did, especially after upgrading some multi-million dollar house to get the value after renovations. In every case, it was an architect who wanted to know how much the house HE lived in was worth after extensive renovations. In one case, the architect had modified an SFR in Sausalito into a 3 unit, where he lived in the largest unit. Another case was an architect who had bought a large home in 2012 and spent 5 years upgrading it, - which he also lived in. I valued it around $5M as I recall.

Now, on the other hand, many appraisers could do much better with just using MARS, bypassing the full RCA method. I've been using MARS since 2004 - and that was a good experience. You first need to learn MARS, including recognizing and avoiding overfitting. And understand, I am not sure it would be worth learning if you are working in the middle of Montana with few comps to analyze. Also, if you make any code changes to your R code, you should test it on some old data and see if it creates the same model. You don't want to create a model and then find out, weeks later, that there was a bug in your code. For some appraisers it would be too much stress to use MARS, RCA and R and still get the speed they need.

 

[J Grant]

You seem to think that MARS replaces the appraiser. When I use MARS it is just a tool like linear regression or a "method" like matched pairs analysis. I still do inspections with measurements per ANSI specifications. So your above arguments don't make any sense. MARS creates the model that results in adjustments and RCA applies constraints to the total adjustments allowed for subjective measures on features lacking measurement such as condition, quality, function utility and view. You would have to understand the math and logic behind these methods. They are a burden EXTRA to your usual work as a traditional appraiser. You can't make much of a living with MARS and RCA unless you are charging over $1.5K per appraisal or have a very efficient system set up, and some kind of guarantee you will be getting at least 2 appraisals/week on average at $1k-1.5K each. The reason I am doing them is more or less R&D. Although I did occasionally get SFR assignments for $1.8K - often from architects who wanted a super good value on some house they did, especially after upgrading some multi-million dollar house to get the value after renovations. In every case, it was an architect who wanted to know how much the house HE lived in was worth after extensive renovations. In one case, the architect had modified an SFR in Sausalito into a 3 unit, where he lived in the largest unit. Another case was an architect who had bought a large home in 2012 and spent 5 years upgrading it, - which he also lived in. I valued it around $5M as I recall.

Now, on the other hand, many appraisers could do much better with just using MARS, bypassing the full RCA method. I've been using MARS since 2004 - and that was a good experience. You first need to learn MARS, including recognizing and avoiding overfitting. And understand, I am not sure it would be worth learning if you are working in the middle of Montana with few comps to analyze. Also, if you make any code changes to your R code, you should test it on some old data and see if it creates the same model. You don't want to create a model and then find out, weeks later, that there was a bug in your code. For some appraisers it would be too much stress to use MARS, RCA and R and still get the speed they need.

Thanks for the explanation. I hope you can get it synced with ACI or Wintotal software integration.

 

[RCA]

Thanks for the explanation. I hope you can get it synced with ACI or Wintotal software integration.

 

[RCA]

I am not planning on doing that at this time, - because that is for GSE forms and GSEs DO NOT KNOW what value contributions are. So, I am not going to waste my time integrating with their forms. But of course, I could extract and aggregate information from my forms to Excel spreadsheets that someone else could upload to WinTotal. That's doable. But I have too many other things I would rather do than get more tied up in interfacing to GSE needs. - I would leave that to someone else.