Fannie Mae ‘Census Block Grid’ Adjustment Raises Redlining Concerns
In a recent discussion within the comments section of an article titled ‘The Censorship of Appraisers,’ an appraiser shared a concerning statement from an underwriter about an adjustment made by Fannie Mae’s Collateral Underwriter tool. The underwriter revealed that the tool uses a ‘Census Block Grid’ adjustment (CBG) to evaluate comparable property values. This adjustment, which modifies values based on median home values within specific census blocks, raises concerns for potentially reflecting practices akin to redlining – a discriminatory practice that is now illegal. The underwriter, while often overlooking the CBG adjustment, conceded that it might occasionally offer useful insights, yet also acknowledged the problematic nature of such a methodology and anticipated possible legal repercussions for Fannie Mae.
Little information is available on the Census Block Grid adjustment, given Fannie Mae’s secrecy around the Collateral Underwriter tool. However, if true, this represents a troubling practice – a government sponsored enterprise providing discriminatory adjustments based on census data that appraisers cannot legally consider, while sharing the data openly with underwriters.
The revelation that this adjustment might influence property valuation, coupled with a notable absence of transparency regarding the underpinnings of the Collateral Underwriter tool, raises profound ethical and legal questions about bias in Fannie Mae’s automated valuation tools. Rigorously examining Fannie Mae’s methodologies is critical to ensure equity and rectify anything that might inadvertently perpetuate historical bias, particularly for an entity with major housing finance influence. Greater transparency is imperative.
CBG stands for Census Block Grid. It’s essentially a location adjustment based on the median values of homes in that CBG.
In my opinion, this adjustment the FNMA model makes is pretty much, in my personal opinion, is the equivalent of red lining. Yet they don’t even realize it.
As an underwriter, I usually ignore this line item adjustment that the model makes. Every now and then it can actually be informative. But, I just find it deeply ironic, and keep waiting for the day someone somewhere sues them over it.”
- Outrage Over Connect by ValueLink’s New Monthly “Junk Fee” - November 27, 2024
- ARCC Discussion Exposes GSEs Agenda to Reduce Appraisal “Friction” - November 22, 2024
- FHFA’s Appraisal Waivers Expansion - October 29, 2024
And this is why the state of Maryland is investigating FNMA, thier valuation acceptance and appraisal waivers that leave people in the dark about thier actual property values. Just trust us they say. Did Zillow do this and imploded?
Duh, the fact that this is now only being discussed after decades of including the census tract in the report is the most amazing aspect of this entire discussion. Simply, including the census tract number in the appraisal report opens up the underwriting process to the same if not worse types of racial profiling and biases Appraisers are being accused of. At least the suspicion of Appraiser bias can be challenged but the automated underwriting tools are not transparent enough for any to be challenged or pushed against by the underwriter or anyone else.
If the census tract number is being used strictly for data collection and statistical analysis of housing activity, that is perfectly fine. However, for the census tract data to have any bearing on the valuation process by formulating adjustments, selecting comparables, or measuring the reliability of the appraisal report undermines the objective of fair lending and fair housing as doing so essentially automates redlining and unfair lending.
With permission from the original author, please feel free to use the below comment as is or edit to your liking. The comment was written after an underwriter expressed concern regarding a comp that was outside the neighborhood
(by way of its census tract) and that no neighborhood adjustment adjustment was applied. In short, it was a bogus condition that was cleared, but the hope was that new comment would prevent such time wasting requests in the future.
COLLATERAL UNDERWRITER (CU):
If the lender / client uses such a platform, the following applies.
In the event of varying CU data from the appraisal it should be noted that the appraiser has provided independent analysis / results relating to any and all market adjustments that specifically relate to a single property (the subject), for a specific day (effective date of this appraisal), and has done so without influence from any known CU results.
Considering in part, both lenders and lender agents acting on lenders behalf are prohibited from distributing the CU print report or the submission summary report (SSR), making demands of, or providing instruction to AMCs/appraisers based solely on the CU automated output, or using CU to interfere with the independent judgment of the appraiser, per this scope of work, if asked for post submittal, the appraiser will provide no clarification relating to any provided variances, and will reference this paragraph for review.
Seek the truth.
I made some slight adjustments, but I like it:
COLLATERAL UNDERWRITER (CU):
If the lender / client uses such a platform, the following applies. In the event of varying CU data from the appraisal it should be noted that the appraiser has provided independent analysis / results relating to any and all market adjustments that specifically relate to a single property (the subject), for a specific day (effective date of this appraisal), and has done so without influence from any known CU results.
Considering in part, both lenders and lender agents acting on lenders behalf are prohibited from distributing the CU print report or the submission summary report (SSR), making demands of, or providing instruction to AMCs/appraisers based solely on the CU automated output, or using CU to interfere with the independent judgment of the appraiser, per this scope of work, if asked for post submittal, unless they can provide the methods used in reaching said analysis. If the methods can not be produced or reproduced in a manor that the layman can understand the CU results will bee rejected and potentially bias in efforts to lower risk assessment scores.
Spencer Paul, review that last sentence and it’s good. There is a spelling error and I believe there is an incorrect word or two. But it’s good.
Awesome, thank you for the catch. More eyes are better.
BIAS!?!?
Don’t you think Fannie will deny, deny, deny?
And what about the ONE MILE radius requirement?
So blatantly inept.
While they don’t have a 1 mile rule, the UW comments regarding such are suggestive otherwise. Yes FNMA will deny anything and everything. Even if this were to blow up and end before Congress, I don’t believe anything will come of it because it would completely erode the media push that appraiser’s are racist and they need that narrative to be pushed to push us out.
Thanks guys. Somebody get this on CNN 😉
They all already know. That is the point of it all. They all know and they are using race as a distraction, or the tool used to nail in the coffin top.
Interesting and concerning.
The process of appraisal modernization began with the CU system and MISMO, and ends with the total elimination of human appraisers in favor of proprietarily managed software.
Ford: Critical analysis of the CU system. Fraudulator 2.0 article.
https://appraisersblogs.com/the-new-improved-fannie-mae-fraudulator-2-is-an-outright-fraud
FHFA & OIG report on AI in mortgage lending.
https://www.fhfaoig.gov/sites/default/files/WPR-2022-002.pdf
Ford: FHFA paper on amc influence on appraisal work quality.
https://appraisersblogs.com/FHFA-working-paper-credibility-questioned
FNMA 2018 on emerging AI. With ‘social trust’. That, or use the CU CBG tool, apparently.
https://www.fanniemae.com/research-and-insights/perspectives/how-will-artificial-intelligence-shape-mortgage-lending
AI/ML-driven “Social Trust” Score:
For thin-credit borrowers, devise a “social trust” score by having machines analyze the individual’s non-financial transactions (e.g., social media activities, internet browsing history, app use, and geolocation data) to help predict loan performance
We’ve known this, the question is why are housing groups attacking appraisers and not the GSEs? Because appraisers are a distraction from the GSEs own culpability.
It’s right out there in the open, even Fannie admits it in their AVM patents.
https://patents.google.com/patent/US20160078573A1/en?#:~:text=Under%20the%20%E2%80%9CTract,in%20valuation%20model.
“Under the “Tract” mode, the comparable analysis application will look for comparable sales in the Census Tract of the subject property, and all contiguous Census Tracts. Because the Census Bureau has tried to identify homogenous areas in the process of defining a Census Tract, this mode is believed to provide an easy but effective method of identifying relevant comparable sales to be used in valuation model.”
From the Census Bureau.
“Designed to be relatively homogeneous units with respect to population characteristics…” i.e. racial makeup.
https://www.census.gov/data/data-tools/survey-explorer/geo.html#:~:text=Census%20Tract,-A%20small%2C%20relatively&text=Glossary%20G%2D11%20Designed%20to,by%20any%20subcounty%20geographic%20entity.
Is this the same AVM that issues waivers? Nobody knows, but we fair housing groups should be encouraged to ask these questions.
It would have to be. If they are using that information to arrive at the CBG adjustments, then it would only be logical that it was being used to locate sales the CU would prefer you use in order to lower the risk score, or in otherwards lower the value. They are the culprits and the one’s that are gaslighting the media about appraiser’s are doing when they are doing it.
CU is not the same thing as the AVM that issues waivers. They may rely upon some of the same data and analysis, but they are different tools.
Oh I do understand that, but I feel I have to assume they are using all the same data. I get that one does the underwriting and one produces the value, but is illogical to view them as working harmoniously in sink? Would this no complete the closed network system they are hoping for, with one major flaw, the closed network would be prone to error with the lack of new data could it not spiral? Meaning the AVM is just there to validate the CU, or the other way around.
The below link; ‘Data analytical models for loan treatment.’ There are new strategies of multi dimensional consideration for every data factor available to assimilate, considerations for lenders which may run beyond the scope of an individuals credit score and an individual home.
Basically if a person is part of a socio economic group of persons whom have higher tendencies to default, be cash poor, spend money unwisely, or somehow be ‘abusive’ to the lending systems (read the stereo type labels they categorically assign people), then that person is treated the same as everyone else around them, even if they do not exhibit those same behaviors personally. They are part of the clustered identification group.
What used to be proprietary behind the scenes operational data utilized by upper tier money managers is now available for public research, because of the way someone monetized the process by way of patenting and selling a software. Want to know the details of how lenders make decisions? And that’s just one find. What else is out there?
Am I the only one who recognizes that there is insufficient qualitative data points in a URAR 1004 to run a reliable AVM. Q3/Q4, C3/C4 is not necessarily consistent (not to mention binary) and may vary depending on comps and no one I know ever makes a line item functional adjustment, and the below/above grade reconciliation is totally contrived to often zero out towards TOTAL GLA; and land value has nothing to do with cost/acre (square vs long rectangle same acreage) – just a few reasons why our data DOES NOT convert to an AVM. Garbage in garbage out. I mean seriously, think about it Lyle!
Why do you think they’re redoing the forms?
Well girls and boys, it sounds like participants in the appraisal industry are coming together in a “grass roots” movement. This is exciting and promising and I hope the momentum continues. Yes., we are mad as hell and not going to take it any more! Rather than complaining, it sounds as if we are using the tools available to us and their own words and ridiculous policies to our advantage. This is the only road to achieve fruitful results. Hopefully this will continue and will result in a class action lawsuit at some point. They can’t have it both ways.
“THE FORM” is the same all same all with just a different entry protocol. In order to be even close to useful you’d need 4-5 well defined mid C/Q entry choices, BUT, and this is the point – our use of the grid with $ adjustments is nothing more than our saying we can narrow the bracket to reflect a more meaningful range which we then might rank. Can you say “Subjective” – I mean really!
Great find Bill. The CU system is a mass assessment hedonic regression tool.
https://en.wikipedia.org/wiki/Hedonic_regression
Corelogic/FNC patent on appraisal report auto review. Expanded patent history, they share the tools with lenders.
https://patents.google.com/patent/US8244563B2/en?q=(Current+Assignee:+fannie+mae)&oq=Current+Assignee:++fannie+mae+inc
Systems in place to predict borrowers behavior based on statistical data of millions of loans, thousands of factors, including activity not related to lending, and geographic data. Corelogic. ‘Data analytic models for loan treatment.’
https://patents.google.com/patent/US8775300B2/en?q=(~patent%2fUS20160078573A1)
_________________________
The classifier stage may be characterized as personalization modeling. When making personalization models, the developers model the actual players, not the target events. In the real estate transaction, that means the borrower, the property, the loan product and the local real estate market are modeled, not the results provided in the second stage (e.g., default, prepayment, or loan modification failure, per se)./ Personalization models provide information about behavior of the player, such as the borrower. Whereas a typical score predicts only one kind of target event (e.g. FICO, default), the personalization models simultaneously predict a portfolio of target events across a wide range of borrower, asset and local market behaviors.
Once the factors have been established, the factors may be used to segregate borrowers into groups, each with a specific profile, containing borrower’s with that profile. Since each factor is an information rich vector describing a particular dimension of the borrower’s behavior potential, the position of any specific group of borrower’s within the array of factors conveys a rich, multi-dimensional view of a specific kind of borrower
The valuation metrics are used for calculating the raw net present value and the revised net present value. A raw net present value is calculated solely upon the outputs of the model and the economic assumptions relative to the borrower’s expected behavior and the property’s projected values.
____________________________
And now we know why lenders deny more loans among some socio economic groups of people and not others. This is a social credit score system. The concept of ‘default rate clusters’. Image attached.
This tool identifies fraudulent activity. The term; ‘specific geographical data’, does that mean the census tract?
https://patents.google.com/patent/US20110173116A1/en?q=(~patent%2fUS20160078573A1)
The mortgage application data may include property valuation information and geographic information. For example, the model or models 115 may be configured to output scores and/or other risk indicators based in part on geographic default risk information.
Can’t wait until they start trying legal cases using their algorithms. Better yet, performing heart and brain surgeries using robots to perform the procedures. Yes, folks, coming soon to your neighborhood legal and medical practices that don’t actually require human beings.
Special life hack; Do not talk to robots.
Especially when holding a scalpel or during a closing argument.
Census, zip code and big data. Can someone explain to me how City of Trenton and neighboring Hamilton Township NJ are supposed to be valued the Same? Thats where Fannie is taking this. The 2 locations well they share multiple zip codes yet Trenton is a decades year old capital of New Jersey and Hamilton is a suburb mostly built mid-century. They share zip codes, but they don’t share schools, services, etc. Markets change from decade old housing to mid current century build from block to block. But the zip codes don’t…. The 2 markets are quite different in every aspect. How exactly is Fannie or whatever to distinguish. Literally people in Hamilton get mail which was supposed to be sent to Trenton property with similar address. I lived at 41 Colonial Avenue in Hamilton Township 08810.Property Routinely got mail from 41 Colonial property In Trenton with same 08810. Incorrect mail was received at the home daily. The mail was incorrectly sent because of the mailing address. Hmmmm. mailing address. Which is UAD compliant. 41 Colonial Avenue in Trenton was being sold and buyers were literally showing up at 41 Colonial property in Hamilton. Same address but different property………. Big Data. Secondary market lenders. Really? Appraisal used to be a profession. Now it’s not. Thanks appraisal institute. and others which collected dues and did nothing. while census tracts and zip codes overcame common sense.
The Institute has been “knee-jerking” their way to obscurity. Shame their leadership is so arrogant they can’t see they’ve been driving themselves and the profession off a cliff. What they are trying to do now is too little, too late.
The letter that wrote might as well have been on some obscure blog that no one reads. It was not shocking and didn’t “do” anything. Just my port right here. It conveys a message but nothing from it will directly happen.
Just got more confirmation that this is really real from one of the more modern architects of the valuation acceptance, appraisal waivers and PDC crap. They are in fact using the CBG in all of these products and in all underwriting.