Rocket mortgage Fraud?

[Fernando]

Did you miss the recent announcement tha waivers will now be used for up 97% ltv?

I disagree in using waivers for 97% LTV. If they're into risk assessment, such high LTV doesn't make sense.... unless their computer model is so accurate now in prediction.

 

[Fernando]

No, I didn't. The 97% LTV waivers are inspection based. A property data collector must inspect the property to ascertain its physical characteristics. That data is then fed into the computer system and a value generated based on the comps in the system that are physically similar to the subject. It sounds like they have tested the model enough to be comfortable with the potential risk. In order to offer it, I'm sure when the address is entered the computer can pretty quickly tell whether there are sufficient comparable data points to return a result with a high confidence score - basic statistical analysis. That, coupled with an analysis of borrower income, credit, and assets, is what offers the waiver options. Again, it's all a risk assessment. Also remember, the GSEs don't just come out with these programs on a whim - they have to be approved by FHFA and show the testing and quality controls behind them.

If the subject is in a subdivision with a relatively homogenous market area and hundreds of sales in the past 6-12 months, how hard is it for the algorithm to search for proximate sales similar in age, site size, bedroom count, GLA, basement area, rated in say C4 condition? Isn't that basically what we do when we get back to the office and search the MLS? That's not a hard thing to program. It then spits out 20 sales meeting those criteria and performs basic regression analysis on them. If the lender entered a value guesstimate of $200k for a 90% LTV, and the resulting comps within a quarter mile had sale prices from $185k to $225k with the predominant sales having prices from $190k to $210k, don't you think it's pretty easy to figure out where the appraisal would come in at? If you sent three appraisers out there don't you think they should all come in somewhere within that range? If an appraiser went out, pulled those same comps, do you think they wouldn't come in somewhere around $200k? And if three ethical appraisers went out using that data, couldn't we have a 5% difference between the three (say $190k; $200k, and $210k) and generally say that all three values were reasonable? If the GSEs can figure all of that data out without needing an appraisal report, why do they need the appraisal?

I mean, my assistants can generally tell whether the lender's value looks good or not just from the data they pull while setting up my work file. They pull the subject's public record and MLS listing data, and then pull comps based on general parameters I've given them. They don't have an appraisal license and never inspected the property but, just by looking at the subject's MLS photos and the MLS results they pulled, can usually make a pretty good guess at value themselves.

I totally get not liking this trend because it takes assignments away from us, but as a statistics guy who generally understands risk-based assessments, it totally makes sense they would go this route. Orders for those easy appraisals of cookie-cutter tract homes in densely populated suburban areas are going to diminish - it's just a fact. But, there will be increased demand for more complex assignments (rural, manufactured, higher quality, large acreage, etc.) because the GSE models can't handle those well.

I just did a very difficult appraisal because there were no good comps in my area.
I could have used sales in an adjacent neighborhood or city but location was so different. I couldn't do it because I don't like to use location adjustments.
If a computer can think like me looking at all relevant and irrelevant factors, computers could one day replace me. That day is not now.

 

[ZZGAMAZZ]

Waivers and hybrids came about because the GSEs now have millions of data points now from which they can analyze a property against the comparable data they have. After nearly a decade of testing CU's accuracy within a certain confidence level, they feel they can accept CU's model value on low LTV loans.

We have to remember appraisals are for determining risk to the lender. If the borrower defaults on this loan, is the collateral's value sufficient (as of the loan date) enough that we could sell it and recoup the loaned money back.

So, think about it. The GSEs prefer 80% LTV or less loans. Let's say a borrower is only seeking a $100k loan and they think their home is worth $200k (50% LTV), and CU gives the subject a model value of $185k (54% LTV). If their testing shows that CU's model value in this location is typically off by no more than 5-6%, then the risk analysis says even if CU is 15% too high they still have a less than 80% LTV. And if the lowest sold comp price available is say $130k (probably the lowest the subject could be worth), they still have an 80% or less LTV. Why do they need an appraisal to show them what they already know - that the value risk is minimal? Any non-appraiser can look at that data and say there's probably no risk here.

Now, there will be a few outliers where there was an unknown condition that greatly negatively affected the subjects value more than was expected, but that's going to be a rare one-off. A couple of those types happening is easily absorbed when tens of thousand of others perform as expected. Also, if the borrowers continue to pay the loan, then it never goes into default and the GSE never has to deal with the unknown negative condition.

So, if you have a borrower that has a high likelihood of being able to pay on the loan AND there appears to be very little value risk given all the comp data available, why is an appraisal needed to assess a risk easily assessed without one? Why should the buyer have to pay for a full appraisal in that situation?

Yet even in an assignment for which the SCA is populated with excellent/appropriate comps, in a reasonably active market, "condition" that typically provides the primary source of adjustments isn't included in the AVM; so how are they accurate? What proportion of the acceptable 5% - 6% is pertains to condition differences? [just curious, although your description of the LTV process is appreciated...]

 

[jay trotta]

Watched a senate hearing the other day in regard to "AI" and the prevention's that should be implemented to; avoid, limit etc. "data manipulation". Guess who is getting ahead of that curve?
IMO- it just supports the theory, that until something is put in place, the manipulated data is being allowed to flourish, and gaming the system isn't anything new, it's just happening in your sandbox.

 

[NC Appraising]

Watched a senate hearing the other day in regard to "AI" and the prevention's that should be implemented to; avoid, limit etc. "data manipulation". Guess who is getting ahead of that curve?
IMO- it just supports the theory, that until something is put in place, the manipulated data is being allowed to flourish, and gaming the system isn't anything new, it's just happening in your sandbox.

AI: not made to make life more convenient, but to alter how humanity perceives and processes reality.

People will depend on AI to tell Tham what is real and what is not.

AI is a tool of perception manipulation people will not be able to interpret reality without the help from reality.

The best tool the government and corporations have to manipulate and to control the herd.