Bias in Automated Valuation Models
The CFPB is reviewing bias in Automated Valuation Models (AVMs). The proposed rules are a joint effort by the Consumer Financial Protection Bureau, the Office of the Controller of the Currency, the Federal Deposit Insurance Corporation, the National Credit Union Administration and the Federal Housing Finance Agency. These agencies are concerned AVMs may reflect bias in design and function. The mathematical models rely on biased data resulting in inaccurate valuations. Basically the agencies are stating historical data going back to redlining is built into these models and do not reflect current market data. Remember markets are not static and are always changing.
VaCAP shared Ted Talks by Cathy O’Neil and Tricia Wang back in 2019 in our “Big Data? Thick Data? Human Data?” post. Both of these professionals have a clear understanding of the pitfalls of algorithms in all aspects of life, not just valuation models. Even if you have listened before, it is well worth listening and sharing again.
Don’t be fooled they are listening; the proposed rules are mandated in the Dodd Frank Act as clearly stated in the introduction section of the proposal. Dodd Frank was enacted on July 21, 2010. The real question is why did it take 12 years.
The Consumer Financial Protection Bureau (CFPB) today outlined options to ensure that computer models used to help determine home valuations are accurate and fair. The options will now be reviewed to determine their potential impact on small businesses.
“It is tempting to think that machines crunching numbers can take bias out of the equation, but they can’t,” said CFPB Director Rohit Chopra. “This initiative is one of many steps we are taking to ensure that in-person and algorithmic appraisals are fairer and more accurate.”
…The CFPB is particularly concerned that without proper safeguards, flawed versions of these models could digitally redline certain neighborhoods and further embed and perpetuate historical lending, wealth, and home value disparities.
- It’s Just Responsible Journalism! - February 21, 2024
- Limitations for Damages Against Appraisers - January 9, 2024
- AMC Engagement Letter: Appraisers Do Not Include Your Invoice - July 28, 2023
As long as people are involved there will be some bias, who programs the AVMs? It would be nice if someone in the group knew something about appraisal
Chuck Minzenberger damn biased computers!!!
Ernie Ramos blaming appraisers for racial bias is like blaming a thermometer for the temperature!
Both the GSEs AVM are racist. They are programmed to undervalue minority communities.
The Dude!!!! Now there’s a guy who doesn’t have a biased bone in his body.
Well, the AVMs likely apply the same “bias” to sales in the neighborhood of the subject property. Go figure! Why would they do that?
So lets see here. Appraisers are biased for using the data we use everyday. you know the same data these algorithms and AVMS use every day with the exception that an appraiser visiting the site has much more info in hand than an AVM.. The same data that appraisers must decipher due to poor agents and tax data collectors. Of course the data will be biased and stupid. So yes lets stop using actual people to see, smell and hear whats going on at the property and replace that with a computer. What could go wrong? If the powers to be want change… start looking at the data that agents and more input and stop looking at appraisers as the bad people. Bad data input equals bad data outputs. Enough with this already. The data has been baked for years and years and that’s an issue of local governments, cities and the country.
The data is all screwed up. Fannie Mae expects regression analysis for adjustments regardless of the bad data. Statistics need good data. Realtors measure homes inside out. Appraisers measure how’s based on exterior measurements. Appraisers have to use condition ratings based on six groups. Realtors rate house based on who owns them the bank or person. Banks were giving homes away that needed cosmetic work. Meanwhile a similar home was move in condition. Now appraisers have to take this data and unravel it. Changes made after 2008 have done nothing. Realtors, appraisers and the assessment office are most often not on the same page. I can write a book about how inconsistent it is. We need to all work with the same rules and held accountable.
Regression Analysis (RA) is so unreliable it has no usefulness. All of the RA’s I have seen don’t include/consider several factors of differences, particularly condition and quality differences/adjustments which can be large. The skewed results are meaningless. Ask a real statistician and they will quickly point out that the data sets for appraisals are considerably to small and typically don’t recognize all pertinent variables to produce reliable results. Smart attorneys and appraisers can easily dismiss this method of deriving adjustments. Unfortunately, the AVM side can spew their nonsense because it seems impressive and intimidating until you break the components down.
AVM’s are unreliable. Their only consistency is being all over the map value wise except for where they should be. Nobody in their right mind should rely on these – ever.
Think about this: Under “the rules”, an appraisal is biased if it concludes a value less than the contract price. That, of course, raises (at least) two questions: (1) why is the AVM always “right” and the appraiser, thus, always wrong? And (2) how will the self-appointed arbiters of bias conclude an appraiser is biased if the appraisal is for a refinance? The Chinese have a curse: “May you live in exciting times!”. Clearly, appraisers are living in exciting times!