I can’t recall in my many years of readership seeing the Financial Times perform a data-driven study of its own, so the pink paper deserves kudos for that alone. We’ll discuss the findings, that of serious bias in mortgage lending against not just blacks, but even Asians, who are on average higher earners than whites. Even though the study has a serious methodological limit, of the researchers not having access to the critically important FICO scores, they made efforts to compensate for that. The disparity is so large that is suggests there is a there there in terms of non-white borrowers still not getting a fair shake.
As with other charged topics, reactions are often influenced by priors, as the Financial Times comment section makes clear. I hope readers who are in the mortgage business as well as those with strong analytical chops will pipe up.
By way of background, the pervasively-used FICO score was intended to reduce bias in credit decisions. Critics contend it has failed to do so. Black and Hispanic borrowers often have thinner credit files, making it impossible under the current system for them to access normal credit products, or when they can, on more unfavorable terms. Note there are other proxies for demonstrating reliability in meeting payment obligations, like rental histories, that are not currently included in FICO but could be added to give a fuller picture.
On top of that, people of color are often targeted for dodgy products (see the subprime crisis) or given worse terms just because. Note that FICO does not factor in whether a loan has “gotcha” terms in evaluating borrower performance. As the National Fair Housing Organization pointed out:
But many studies and analyses have demonstrated that inappropriate loan products and their components were key factors driving the subprime crisis. Factors including product type, presence of a yield spread premium, distribution channel, inflated appraisals, and prepayment penalties helped significantly to predict whether a loan would fail. Even major credit repositories and credit scoring companies, including Vantage Score and FICO, admit that credit scores declined in predictive value leading up to and during the foreclosure crisis.
Fintech Takes has a longer-form and more current critique of FICO, including the increased lender use of in-house using internal data, supposedly to supplement FICO. But as the piece argues, it is in the process will undermine FICO, as lenders delay or don’t bother buying FICO upgrades and invest in their own models instead.
We see that now in BNPL lending, as discussed in our article.
And the point with respect to the Financial Times’ piece is the whinges about the study not using FICO miss the fact that FICO is increasingly not the key determinant of lender decisions, one assumes particularly at the biggest lenders that have the most payment information.
I have no way of knowing how much people of color are still subject to getting less favorable lending terms. But a big point here is it has happened, and is probably still happening, albeit not on a subrprime lending level. So absent understanding whether that sort of practice is still fairly common, or has become rare, means that one can’t place as much faith in FICO scores as its defenders would like.
And it’s not as if there isn’t plenty of bias against blacks. Studies have found that when the exact same resume is sent to prospective employers, but one with a black-seeming name like Lakisha Jackson or Tyrone Washington, versus white-sounding ones like Sally Smith and Harry Hunter, many recipients stop reading the resume when they hit the presumed black name.
And there’s ample anecdata, such as this 2022 case study in the New York Times, Home Appraised With a Black Owner: $472,000. With a White Owner: $750,000:
Nathan Connolly and his wife, Shani Mott, welcomed an appraiser into their house in Baltimore, hoping to take advantage of historically low interest rates and refinance their mortgage.
They believed that their house — improved with a new $5,000 tankless water heater and $35,000 in other renovations — was worth much more than the $450,000 that they paid for it in 2017. Home prices have been on the rise nationwide since the pandemic; in Baltimore, they have gone up 42 percent in the past five years, according to Zillow.com.
But 20/20 Valuations, a Maryland appraisal company, put the home’s value at $472,000, and in turn, loanDepot, a mortgage lender, denied the couple a refinance loan.
Dr. Connolly said he knew why: He, his wife and three children, aged 15, 12 and 9, are Black. A professor of history at Johns Hopkins University, Dr. Connolly is an expert on redlining and the legacy of white supremacy in American cities, and much of his research focuses on the role of race in the housing market.
Months after that first appraisal, the couple applied for another refinance loan, removed family photos and had a white male colleague — another Johns Hopkins professor — stand in for them. The second appraiser valued the house at $750,000.
With these caveat, let’s turn to the Financial Times study. The methodology:
The FT analysis was based on several approaches to the HMDA data from 2018 to 2023. In all cases, the data was slimmed down to focus on applications for conventional mortgages for home purchase, refinancing or refurbishment, and for owner-occupied single-family unit homes. The analysis looked at all major disclosed factors, from borrower age, loan-to-value ratio and income to the characteristics of areas where mortgages were sought. This data was then examined both year-by-year and in aggregate using two broad approaches.
First, a logistic regression model, used to estimate the size of the racial gaps that could not be explained by other factors in the HMDA data.
Second, a “pairing” approach, where the FT would identify a group of white borrowers who were apparently identical in all respects, other than race, to a group of borrowers of other ethnicities. Comparing refusal rates of these groups provided another way to calculate the differentials in refusal rates. The two approaches gave extremely similar results. The patterns observed in the data — such as the order of the results — were broadly consistent between approaches, lenders, areas and over time. The FT’s results also echo older academic research.
And the results:
The FT reviewed 39.5mn mortgage applications submitted to all lenders between 2018 and 2023 using data collected under the Home Mortgage Disclosure Act, a law introduced half a century ago aimed at ending discrimination in the home loan market.
A statistical model that took account of differences in declared income, debt levels, loan size and where people lived was then applied. The results showed Black people were 2.1 times as likely to be denied a conventional mortgage for an owner-occupied home as white applicants; Latinos, 1.5 times; and Asians, 1.2 times.
The Financial Times concedes the outcomes may not be as unwarranted as they appear:
Bank of America said the FT’s analysis “doesn’t include credit history and detailed borrower information that financial institutions use to evaluate loans and are required by the government-sponsored entities like Fannie Mae and Freddie Mac, which guarantee mortgages in the United States”.
The lenders’ argument is bolstered by the close clustering of most of them, with similar rates and patterns of racial disparity.
The few cases with different rates may be explained by their business models. Tennessee-based 21st Mortgage Corporation — one of the biggest providers of loans for prefabricated homes — was 2.5 times more likely to deny Black applicants than white ones.
It said its high denial rate could be explained by the fact that it took online applicants and did not pre-qualify them, meaning it has more speculative applications and inquiries with incomplete or inaccurate information. This is consistent with it having a markedly higher overall denial rate, which stands at 73 per cent versus 20 per cent for the rest of the market.
The Financial Times provided more granular information, as in how the models of various credit agencies scored their data:
The denial rate gap between Black and white applicants with the same characteristics was 23 percentage points under Equifax’s model, 14 points under Experian’s and 13 points under TransUnion’s.
And the article pointed out that limited information about many black borrowers means that some who actually have routinely paid on time won’t be able to demonstrate that:
A 2022 study by the Federal Reserve found that Black and Hispanic applicants tended to be more leveraged and have much lower credit scores, with the average credit score for Black applicants about 40 points lower than that of white applicants…
But the very fact people of colour have disproportionately thin credit files comes from the historic segregation of Americans and the fact that neighbourhoods are still divided by race, according to advocacy group the National Fair Housing Alliance.
Predominantly non-white neighbourhoods are rife with subprime lenders, cheque cashers and payday lenders that were less likely to report positive payments to repositories than the highly regulated banks that tended to be concentrated in predominantly white neighbourhoods, the NFHA said.
A 2021 American Enterprise Institute study, contesting a Housing Center report that used new Federal mortgage data, and had found, like the Financial Times, that blacks with similar characteristics were twice as likely to deny blacks mortgage loans than white, and also had higher-than-seemed-warranted rejection rates for Hispanics, Asians, and Native Americans.
The AEI argued that this result was warranted, using a large dataset of mortgages originated in 2019-2021. Their finding:
In the case of conventional loans, White borrowers averaged a 4.4% E60+ [60 day or more delinquency] through April 2021, while Black mortgage holders on average experienced 6.4%. In the case of FHA loans, the respective percentages were 15.7% and 19.6%.
However, given that this period includes Peak Covid, it’s hard to see the results as representative. Or to the extent it is, it confirms the sorry fact that black employment and pay levels are more vulnerable in severe downturns than those of whites.
We’ve been trying to get a house for months, and we had another one previously. We spoke to several banks and lenders. Credit score is a key determinant at every single one, with the caveat that they actually take the median score between FICO, TransUnion, and Experian, not specifically the FICO score. The difference is typically 10 points or less.
The relevant factors:
– credit score, sets your rate and determines whether or not you qualify.
– debt to income ratio. Depending on the mortgage, lenders will allow borrowers to go up to 45% or 50%, so it’s harder to qualify if you have car loans or credit card or student loan debts.
– down payment, a larger one improves the terms. This is an area where wealthy parents sometimes help, there’s a drastic improvement in terms with a 5% down payment, and then with a 20% down payment. With the latter, mortgage insurance premiums (can be several hundred a month) go away.
Lenders they do things like ask for bank statements, verify employment history, ask for two years of W-2’s, etc. those aspects will be much harder for a population with larger numbers working under the table, or with unstable and/or seasonal employment. The ideal from the perspective of lenders is somebody who’s had a single employer over two plus years that posts consistent paychecks that don’t vary from month to month.
Yes. If you are self-employed, it makes borrowing much more difficult. Even credit unions – I was self-employed for years and when I tried to fire Bank of America and move my mortgage to a local credit union, they refused me. Despite income verification via tax returns, etc. I was mightily angered and only use this credit union for banking convenience (I’m not a profitable customer!) – and routinely give them bad marks on customer satisfaction surveys. B of A refinanced my mortgage without a hiccup, incidentally. Also – I’m not a visible minority except being pretty much a hippie – long hair, etc. Maybe that was it!
In Bernie and Hillary debates Hillary said regulating banks don’t end racism. She wanted Bernie to talk more about identity politics and not economy but even in that she was wrong because Bernie regulating banks would have minimize banks racisim against blacks .
I’d racked up some debts in the early 90s, most of which I never paid back. My credit was so bad then that I was once denied an apartment when the landlord ran a credit report, even though I had always paid the rent on time. (Pro tip – if you live with roommates, never put all the bills in your name since they might just disappear and leave you holding the bag). I cancelled the one credit card I had, paid all my bills on time for many years, and never bounced any checks. When I went in for a mortgage in the mid-aughts, I basically had no credit they could find. All the bad stuff had fallen off the reports they ran.
What I did then was take out an overdraft line of credit (ODLOC), which basically just transfers funds from your line of credit to your checking account to keep you from overdrafting. You wind up paying the interest on the transferred funds (basically pennies if you pay it back quickly) rather than an exorbitant bounced check fee. I deliberately overdrew my account a few times and paid it back right away, this to develop a credit history as recommended by the credit union, and a short time later I had a credit score of 800+ and easily got a mortgage. Also, this was right after home prices cratered in the financial crisis, so we were able to do a 20% down payment. We took out the mortgage from a small credit union, not a huge bank.
After reading this article, I’m now wondering if I got approved for the mortgage because I was able to successfully game the system, we got lucky with the timing when prices cratered, or because I’m Caucasian, or maybe some of each, because I sure wasn’t some squillionaire with an impeccable record when I went in for the loan.
I track my (>800) Credit score monthly and keep a spreadsheet record. I can usually tell what causes fluctuations, generally when my credit card balances go up with large purchases. However, there are times when my credit score goes up or down for no discernable reason – my suspicion is that the algorithm is tweaked in order to tighten or loosen credit. Just one tool in the toolbox of managing the credit economy. (=managing the herd).
Back in 2007-8, a friend of mine with a small business had his line of credit summarily cut off by Amex. He had to repay it immediately. It put him out of business, as his revenues were invoiced but his expenses upfront and he used the LOC to manage this. So – more evidence that credit tightening is manipulated – why? As in the COVID op, in order to advance monopoly interests and kill off small independent businesses? I mean, that’s what happened both in 2008 and under COVID.
The COMPAS (Correctional Offender Management Profiling for Alternative Sanctions) system has been at the center of significant controversy regarding racial bias in algorithmic criminal justice decisions. Here’s what research has revealed about this system:
The ProPublica Investigation
ProPublica’s landmark 2016 analysis found that black defendants were far more likely than white defendants to be incorrectly judged to be at a higher risk of recidivism, with Black defendants incorrectly labeled as “high-risk” to commit a future crime twice as often as their white counterparts. This investigation sparked widespread debate about algorithmic fairness in criminal justice.
How COMPAS Works
COMPAS risk assessments are used at various stages of the criminal justice process, including sentencing decisions when considering parole or social services assignments. The system generates predictions about a defendant’s likelihood of reoffending, which judges use to inform their decisions.
The Bias Controversy
The evidence for bias is significant but contested. While ProPublica claimed COMPAS was biased against black defendants, Northpointe (the company that created COMPAS) disputed this analysis, leading to ongoing academic debate. A computer program used for bail and sentencing decisions was labeled biased against blacks. However, even Northpointe’s founder has suggested that race-correlated factors may be at play in the algorithm.
Systemic Issues
Several structural problems contribute to these biases:
Lack of Transparency: Since COMPAS algorithms are trade secrets, they cannot be examined by the public or affected parties, which may violate due process rights.
Historical Data: The system learns from historical criminal justice data, which reflects decades of discriminatory policing and sentencing practices, potentially perpetuating these biases.
Proxy Variables: Even without explicitly using race, the algorithm may rely on factors that correlate strongly with race, such as zip code, employment history, or family criminal history.
Broader Impact
Research shows that COMPAS predictions generally favor jailing over release and demonstrate bias against defendants, though this bias can potentially be corrected while maintaining accuracy. The COMPAS controversy highlights fundamental questions about whether algorithmic tools can ever be truly fair when built on biased historical data, and whether the pursuit of efficiency in criminal justice should come at the cost of perpetuating racial disparities.
I wouldn’t want a tankless water heater if you gave me one for free.
Can I point out that a (successful) DEI initiative at any given bank or lender would have set up a process where backoffice risk analysts or lending officers wouldn’t be able to see name, gender, address, education, etc., on the mortgage application, which would make it difficult to discriminate on those grounds. It’s not hard to do – banks build their in-house UI’s whichever way they want or find helpful. In theory a name or gender should be completely irrelevant.
Consequently, if the FT can find a bank or lender where this has been successfully done, it would be interesting to compare results and would also provide a helpful control.
Were I the author of this study I would have very pointedly asked each institution whether they implement this kind of blind or, alternatively, whether name, address, gender, etc., are shown to risk analysts anywhere in the decision train leading to approval/denial.
And if the answer is yes, then why do they consider this information relevant to the process enough to expose it to risk analysts?
The answer to the question would also tell us if an instituion is walking the walk when it comes to DEI. If they haven’t implemented such measures anything they say about DEI is mere show.
I am highly confident that the belief of US banks is they need to build models for mortgages on FICO (fintech is different because not regulated and so harder to get data to prove discrimination) because FICO is believed to be bias blind. Liability will always trump DEI.