Sadly, no

By Kathy G.

Responding to this post, a Certain Someone writes:

Credit is one of those weird areas where there is a lot of belief in discrimination, but as far as I can tell, not all that much evidence.

[. . .]

Yes, I've seen the research arguing that people in black communities get worse loan terms than their credit score suggests. As far as I can tell, this research failed to control for some pretty major factors, like assets. . .

[. . .]

Most of the aggregate research I've seen fails to reject the null hypothesis that there is no discrimination in loan markets . . .

Would that that were true. But actually, the bulk of the academic literature on this subject suggests that there is a significant degree of racial discrimination in loan and credit markets.

To be fair, it's not easy to determine the extent to which discrimination occurs. The available datasets are incomplete. Researchers don't necessarily know which variables the lenders and creditors are looking at when considering credit or loan applications, or how those variables are weighted.

Nevertheless, researchers have been able to get their hands on some unusually rich datasets, which they've examined using the most plausible specifications as to the lending criteria.

There are basically two types of credit discrimination that occur: discrimination based on the race of the applicant, and discrimination based on the racial composition of the neighborhood where the applicant resides (the latter type of discrimination is known as "redlining").

In terms of discrimination based on the race of the applicant, the best evidence we have is a series of studies based on data from the 1991 Home Mortgage Disclosure Act, with supplementary data supplied by the Federal Reserve Bank of Boston. Among the factors looked at were individual applicants' financial, employment, and property background variables -- and yes, this data contain info about assets, both liquid and total (which is contrary to the assertion made above that factors like assets are not controlled for in these studies). The original paper by Alicia Munnell et al., which was published in the American Economic Review, the premier academic journal in the economics field, found that applications from blacks and Hispanics were significantly more likely to be rejected than similar applications from whites.

Now, as the economist Kevin Lang points out in his invaluable book Poverty and Discrimination -- a book, I might add, that has received strongly favorable reviews by such right-leaning economists as Arnold Kling (who wrote, "I heart Kevin Lang") and Tyler Cowan Tyler Cowen --  Munnell's original study has been subject to "extensive reanalysis." Critics have pointed out that even though the dataset used in the study is probably the richest we have on the subject, it doesn't, in Lang's words "fully capture all the information available to the banks." This is true;  but based on chapter 3 of this book, which reanalyzes the Boston Fed data, Lang concurs with the assessment that the Boston Fed studies "create a presumption that discrimination exists in mortgage lending."

Lang, citing the discussion of audit studies that occur in chapter 2 of the Urban Institute book, also notes that audit studies of the pre-approval loan process "confirm that blacks receive less encouragement to apply and are more likely to be encouraged to apply to other lenders."

There's more: a review of the literature on racial discrimination in credit markets in the current issue of the Annual Review of Sociology concludes that, although the number of mortgage loans given to blacks and Hispanics has substantially increased over the past two decades, significant discrimination persists:

Nevertheless, the evidence indicates that blacks and Hispanics continue to face higher rejection rates and receive less favorable terms than whites of equal credit risk.

Thus far, the best research that has been done on the subject of racial discrimination in credit markets concerns mortgages. But studies on other forms of consumer credit have just begun to emerge. In a review of this emerging literature, economist Gary Dymski writes that initial findings suggest that "specific minority borrowers are significantly more likely to be sold a subprime or predatory loan than are nonminority borrowers with similar risk profiles." A recent, very interesting paper by the Federal Reserve Bank of Boston looks at credit card redlining; here's a summary of the results:

Using a unique and proprietary database of credit histories from a major credit bureau, this paper links location-based information on race with individual credit files. After controlling for the influence of such other place-specific factors as crime, housing vacancy rates, and general population demographics, the paper finds qualitatively large differences in the amount of credit offered to similarly qualified applicants living in Black versus White areas.

Again, these findings must be taken with caution. Even though the dataset used for this study is unusually rich, it is far from complete, and the researchers did not have information on the criteria by which these financial institutions made their lending decisions. On the other hand, the researchers did have access to thousands of individual credit reports from a geographically stratified random sample, and the results are suggestive.

Even acknowledging that the available data is far from perfect, the evidence is overwhelmingly clear: there is a significant degree of racial discrimination in credit markets. Honest brokers who have reviewed this literature, such as economists like Lang and Dymski and the sociologists in the Annual Review of Sociology acknowledge the ambiguities but come to the basic conclusion that the evidence on this score shows that discrimination exists.

The argument that this kind of discrimination can't happen because it would mean markets are behaving irrationally is dubious at best. Human beings and the institutions they create are not always rational. Norms, institutions, and psychological factors play a role in economic decisions, and sometimes they cause economic actors to behave in ways that are not entirely rational. 

In the same post where McCardle argued that credit discrimination doesn't exist, she nevertheless admitted that "the evidence for discrimination in the labor market seems strong--nay, nearly incontrovertible." Well, surely racial discrimination in labor markets is irrational behavior as well -- a firm would almost certainly be better off by hiring the most productive job applicants, regardless of race. Why would firms act irrationally when it comes to employment discrimination, yet behave completely rationally when it comes to issuing credit?

If racial bias is strong enough to cause firms to act in irrational ways when it comes to hiring decisions, it seems highly plausible to me that it might cause them to act irrationally when it comes to credit decisions as well. And indeed, the evidence strongly suggests that racial discrimination is no less absent in credit markets than it is in so many other features of American life.

Comments

[Gabriel]

"CowEn". Never "Cowan".

[bob]

What are the chances that a certain someone, like other hacks in service of the status quo, asserts that racism to be a thing of past.

[Susan of Texas]

I'm sorry, Kathy, but Megan has said loan discrimination is impossible because computers decide who gets a loan, and we all know computers are not people and therefore can't be prejudiced.

Megan also said that most arguments are too boring for her to refute, so I gues I'll just have to take her word for everything.

[Ninja Zombie]

In hiring, you have an actual person making a judgement call based on another person sitting in front of them.

In issuing credit, you have a bunch of statisticians gathering data, putting it into their computer, which then spits out a formula designed to optimize profit. Then, when it comes time to run a credit check, the SSN of the applicant is typed into the computer, which then runs the formula on their data.

In short: humans making hiring decisions may have subtle biases. Computers making credit decisions don't.

[Susan of Texas]

"The National Community Reinvestment Coalition recently studied mortgage information from 100 metropolitan areas in the U.S. Some observers say the results have been quite shocking.

For instance, the study shows that, in Hartford, Connecticut, minorities were much more likely to own a high-cost mortgage than Caucasian homeowners. This was the case no matter what the income of the people being studied. Interestingly enough, the difference was actually greater for middle-income blacks than for low-income blacks.

According to the study, middle-income blacks in Hartford were 2.9 times more likely to have a high-cost loan than a middle-income Caucasian borrower. Meanwhile, poor blacks in Hartford were 2.1 times more likely to have a high-cost loan than poor white borrowers.

Hartford ranked 11th worst in the nation for racial differences in high-cost loans. Since blacks and Hispanics comprise 80% of Hartford's population, it's not surprising that there have been so many foreclosures in minority communities."
http://www.rebuild.org/news-article/minorities-targeted-for-subprime-loans/

[Mike]

==
In issuing credit, you have a bunch of statisticians gathering data, putting it into their computer, which then spits out a formula designed to optimize profit. Then, when it comes time to run a credit check, the SSN of the applicant is typed into the computer, which then runs the formula on their data.
==

That sounds like magic. It also sounds like your FICO score (plus maybe some correlations matrices conditional on the bank's holding - not the applicant). Your FICO score isn't the only thing they look at when you decide to get a loan.

That 'data' you refer to, by the way, doesn't give you anywhere near a clean of an answer as you might think it does. Also her arbitrage-based argument for why loans would converge to an optimal risk/reward setting has problems as well, if only cause there are massive transaction costs with home loan searching.

[Bloix]

Imagine that there are two types of widget - one red and one blue. The defect rate among red widgets is 3% and the defect rate among blue widgets is 4%. There is no test that will detect a defect before it fails during use. If the effects of failure are sufficiently severe, purchasers will pay a premium for red widgets. This is so even though 96% of blue widgets function just as well as the best red widgets.

This is why discrimination is so difficult to eradicate. Small differences among populations do correlate with race (and gender, religion, family background, place of origin, and every other stereotype you can name).

Even if the overwhelming majority of individuals do not conform to the stereotype, employers/lenders/renters who acknowledge that the stereotype provides some information and use it to price risk will do better in the long run than those who do not.

So, for example, if I am told that a team of all women is going to play soccer against a team of all men, chosen randomly, and I must bet $100 on the result, I will be advised to bet on the men. Of course the women may turn out to be the US national team and the men may be the Walt Whitman High School team, but if the bet is repeated many times, then I will profit most by selecting the men every time. In order to profit, all I need to know is the sex.

Similarly, if I am a lender, I do not need to know whether the borrowers have well-off grandparents who can help out in a jam; or whether they are at risk of lay-offs in a slow economy; or any other of many factors that I really can't quantify. All I need to know is the race of the applicant. Since race correlates, however weakly, with these unknown factors, I can make more money by taking race into account than if I ignore it. I really don't care if I am being fair to the young couple across the desk from me. Fairness is not part of my job.

Anti-discrimination laws are necessary because discrimination is rational; the market will never eradicate it.

[Megan McArdle]

You misread the post. I didn't say that there was no discrimination because it was irrational, and the "gotcha" point you've offered was in fact used to make . . . exactly the same point you do: the arbitrage argument is obviously imperfect.

Rather, the point was that I think the evidence for current discrimination is pretty thin. (1991 is now almost 20 years ago). There are two ways to test for discrimination: look at the characteristics of the applicants, and then look at the performance of their loans. The evidence of discrimination in gross studies of racial classes is suggestive--but when you test that theory on loan performance, it falls apart. If borrowers were irrationally discriminating against minorities, their loans should outperform whites in the same loan class. They don't seem to. There have been multiple studies of this subject, and they tend to conclude with "we could not reject the null hypothesis".

Partly, of course, we have to decide what we mean by discrimination. If living in East New York is correlated both with being a poor credit risk, and being black, is it discrimination to deduct points for that? Having a low income and a bad credit history are also correlated with being black, (and with each other), but we don't tell banks they can't consider those factors. Moreover, poor neighborhoods are, in fact, a higher mortgage risk, because those are the neighborhoods where the houses tend to lose value fastest. If I buy a house in Bloomingdale, as I would like to, the bank will be taking a higher risk on me than they would on a similarly priced housing unit near Dupont Circle. Even for the front end studies, my understanding is that once you control for things like house price risk, asset ownership, and sundries such as length of employment/time at address/length of credit history, I'm told the effect looks a lot less interesting. Which is not surprising, because in today's mortgage market, it's very hard for lenders to find out what the race of an applicant is, while obviously most employers know right away.

But I don't think it's impossible that, say, mortgage brokers are discriminating--I'd never claim that it can't happen because that would be irrational. My claim is much weaker: it doesn't seem to be happening, and the fact that it doesn't seem to be happening isn't really surprising, given how impersonal the whole process is these days.

[Ninja Zombie]

Mike:

In addition to your credit rating, they also look at your current circumstances, and what you want to do with the money. Virtually all of it is done by computer.

And the data doesn't necessarily give a clean answer (though it is typically locally unique and stable), but that's not the point. The point is the process: gather data excluding race -> run optimization routine on it -> get formula not including race -> have computer apply formula to data.

The process is automated so there is no unconscious bias (unlike hiring). Same input -> same output, no hidden variables. And race isn't one of the non-hidden variables (to minimize risks of lawsuits, the opre guys are not even given racial info).

[L2P]

Virtually all of it is done by computer.

Note the "virtually." Fail.

[Mike]

Ninja Zombie,

I don't think it is possible to enter the data for "what you want to do with the money" (say, to start a business) and have it give a risk assessment without a user-input on something. I mean you could, but I doubt that happens in practice, cause it would be an odd estimate. (Source on the 'stable' condition? Have you been paying attention to contagion in the credit markets?)

But whatever; the issue is if the FICO cutoff for an Alt-A is (say) 700, and a black candidate has a 710, if he is given a subprime loan is that discrimination? A white candidate, 690, given a Alt-A loan? Cause that certainly looks like what is going on...

Megan:"If borrowers were irrationally discriminating...their loans should outperform whites in the same loan class."

Kathy, I know this is a UChicago thing, could you discuss this idea further? It strikes me as a sufficient but not necessary condition for discrimination in general. Above, if the black candidate with the 710 goes to the subprime pool, he gets aggregated among other blacks who, let's say, are defaulting a lot. One can immediately say "see, black candidates are doing poorly."

If you condition it on his FICO score, which is what I assume they mean, well that's exactly the discrimination we wanted to prevent ex-ante; and it can't be compared to the white with the 690 and the Alt-A beacuse the subprimes pays (a lot) more in interest that is more risky itself, and the returns are bounded on the top end anyway (you can't overperform, a loan, just not-underperform it).

[Bigbalagan]

One way to look at this problem is to consider the imbalances between neighborhood assets that are held in banks (deposits, etc.) and how much of those assets return in the form of mortgages. It's surprisingly out of balance for non-white urban neighborhoods.

When you look at this as a macro picture, as opposed to issues like credit-worthiness and redlining, it's pretty clear that, in many situations, urban deposits are funding suburban housing growth. In a certain (twisted) logical sense, you could argue that banks would not be fulfilling their fiduciary responsibilities to their (even poor, urban) depositors if they weren't investing in places where growth was the most lucrative (i.e., not urban rehab). But posed as a deposit vs. mortgage question, one might be inclined to look beyond the underwriting problem to the more fundamental issue of how do we get economies of scale is creating affordable housing? A more specific form of the question is: why should poor black and hispanic depositors place their money in banks that don't give them mortgages?

As the housing business has migrated from the status quo as late as the late 70's, based on savings banks, into the commercial sector, it has gotten increasingly impossible to identify the problem in this simple way. But by now it should be obvious to everyone what the endstate of that (d)evolution has been: mortgages have moved *through* the commercial banks, to the investment banks, where they have been bundled into collateralized investment vehicles which take out the whole mortgage system when they implode from their own complete lack of transparency.

[Susan of Texas]

"Which is not surprising, because in today's mortgage market, it's very hard for lenders to find out what the race of an applicant is, while obviously most employers know right away."

I don't see how that is so. An employer hiring chooses from a stack of applicants, just like a loan officer. He or she does not interview each applicant; there could be dozens or even hundreds. The application process at first is not that different from a loan process. And as already noted, a computer alone doesn't make the final decision for loan processes either.

For an interesting study on this see the University of Chicago's School of Business study on job applications with African-sounding names versus Anglo-sounding names. http://www.chicagogsb.edu/capideas/spring03/racialbias.html

[Barry]

Ninja zombie:

"In issuing credit, you have a bunch of statisticians gathering data, putting it into their computer, which then spits out a formula designed to optimize profit. Then, when it comes time to run a credit check, the SSN of the applicant is typed into the computer, which then runs the formula on their data."

Mike: "That sounds like magic. It also sounds like your FICO score (plus maybe some correlations matrices conditional on the bank's holding - not the applicant). Your FICO score isn't the only thing they look at when you decide to get a loan.

That 'data' you refer to, by the way, doesn't give you anywhere near a clean of an answer as you might think it does. Also her arbitrage-based argument for why loans would converge to an optimal risk/reward setting has problems as well, if only cause there are massive transaction costs with home loan searching."

It's also wrong - the data ain't gathered by statisticians; it comes into the system through a lot of means.

[Barry]

Megan: "Rather, the point was that I think the evidence for current discrimination is pretty thin. (1991 is now almost 20 years ago). "

Well, we have a cut-off for any cite that Megan uses. If it's 1991 or before, throw it out. Boy, it certainly gets rid of those dusty old classics from U Chic, doesn't it? Any argument based on efficient markets and (Chic blah, Chic blah, etc. blah) could have and certainly was made in 1991, 1981, etc.

Standard right-wing line about discrimination: (a) It doesn't happen (efficient markets), (b) if it did happen, the Invisible Hand would smite, (c) it did used to happen, but oh so long ago, and finally (d) government interference would be evil.

Actually, not 'finally', because right-wingers just cycle back through (a) again, or, as Megan did, to (c), since Kathy G blocked her use of (a).

Meanwhile, from Amazon, the book 'Poverty and Discrimination' was published in 2007. That suggests to me that some of the data juuuuuuuuuuuuuuuuuuust might be from a bit later than 1991.

In fact the Urban Institute's study is dated 1999. The cites are:
"Two of these studies find no evidence of redlining, but a third, which accounts for the relationship between redlining and private mortgage insurance, finds redlining against low-income neighborhoods, which in Boston are largely black (Tootell 1996a; Hunter and Walker 1996; Ross and Tootell 1998). "

Other cites are: "Most studies focus on outcomes by census tract, while one attempts to isolate the role of lenders (Schill and Wachter 1993; Phillips-Patrick and Rossi 1996)."

In the review article in 'Annual Review of Sociology', a casual skim has cite after cite from the 1990's.

[Walter]

Megan: "If borrowers [sic] were irrationally discriminating against minorities, their loans should outperform whites in the same loan class."

Megan's logic escapes me. Don't we call it discrimination because equally qualified minorities--individuals who would be expected to perform at least as well as their white counterparts--don't get into the same loan class as whites. How can exclusion of any significant portion of these minorities lead one to expect that the rest would outperform whites? If one assumes the process is less than perfect--so that most members of the class are fully qualified but some less than qualified people leak through--the exclusion of fully qualified minorities would increase the percentage of less than qualified minorities in the class and tend to lower performance.

But the real problem is the blithe assertion that comparing "characteristics of the applicants" with "the performance of their loans" is an adequate test for discrimination in the loan application process. Isn't it obvious that a myriad of unforeseen real world factors--eg. catastrophic illness, sudden job loss etc.--invisible to the process of determining the "characteristics of the applicants" can and do lead to performance difficulties? Even without positing that racial discrimination is a measurable factor in the relative incidence of these life crises, there is no reason to assume that the causes of under-performance across racial lines accord precisely with the pre-loan ranking of creditworthiness.

I find it interesting that Megan is concerned that the research indicating racial discrimination in credit approval "failed to control for
some pretty major factors" yet suggests a test for determining the existence of such discrimination exists that "falls apart" on the same exact grounds.

[Ninja Zombie]

Mike: for starting a business, perhaps not. I don't know how business loans work.

However (according to a mortgage broker I asked about this today), the process for a home mortgages is completely automated. The broker gives basic info on the applicant (SSN, income, etc) and the property to be purchased (address, square footage, etc). He suggested family homes are good, bachelor pads are bad (based on his personal experience, he doesn't know the formulas). The mortgage issuer says yes or no, and gives loan terms.

Information on race does not pass from the broker to the issuer, and therefore can not play a part in the decision of the issuer. To discriminate, you need to know the race of the people involved. Loan issuers and their computers don't.

[Barry]

Susan:

"For an interesting study on this see the University of Chicago's School of Business study on job applications with African-sounding names versus Anglo-sounding names. http://www.chicagogsb.edu/capideas/spring03/racialbias.html"

Oh noes!!!!1!!1!! Teh Nobl Laryetz r pol korekt!

[Barry]

Ninja Zombie: "Information on race does not pass from the broker to the issuer, and therefore can not play a part in the decision of the issuer. To discriminate, you need to know the race of the people involved. Loan issuers and their computers don't. "

They don't have race information, or things which they could use to infer race?

Riiiiiiiiiiiiiiiigggggggghhhhht.

Also, it's becoming more and more clear in this mortgage meltdown that the clean, whiteboard description of the system is definitely not how it functioned.

[Ninja Zombie]

Exactly. The broker does not tell the mortgage issuer the race of the applicant. There is a form, with many textboxes: SSN of the applicant, address of the house to be purchased, etc. There is no race textbox/dropdown/etc.

Next thing, you'll be telling me amazon discriminates (even though their website only asks for name, cc and address).

[mike]

the problem is that the algorithm doesn't kick in until midway through the process. Ill happily grant that the formulA is colorblind (neglect whether fico rewards a certain habitus or not), but please grant back human beings are involved in the process at some point. Many banks recently could care less about optimal reward risk since thet were selling the mortgages on the secondary market.

Also im not making this up. There's already tons of data saying that blacks were offered subprime loans when they qualified for better term (same with whites) check a portfolio.com link from Megan's page - subprime looks like a redzoning map. Megan thinks it is the result of not enough controls - a very smart guy I discussed this with from the fed thinks that they were classified however the secondary market wanted to buy them (which were subprime which there was high demand for) though they were given fair terms as with their true rating, the subprime type just a title. Both are tesible hypothesis, and people are digging into the data as we speak - I'm not too optimistic about what the results will be.

Also I got my example about average outperformance as a sign of racial backwards above - I still think it is shady but need to think about it more.

[Barry]

"Also im not making this up. There's already tons of data saying that blacks were offered subprime loans when they qualified for better term (same with whites) check a portfolio.com link from Megan's page - subprime looks like a redzoning map. "

That's why I've come to not give a flying f*ck about what right-wingers think, because we keep pointing out evidence, and they keep pointing to a theoretical ideal of what should happen. It's like somebody pointing to an advertisement, to disprove reality.

[Anon]

This thread has probably already passed away, but I thought I'd ask:

Ninjazombie has been pointing at automation as a bulwark against racism. But I thought the history of redlining was similarly impersonal. Banks simply carved out areas that were too risky to invest in -- these often just happened to be neighborhoods occupied by minorities. Difficulties in obtaining loans (home, business) then led to a vicious circle. I seem to recall Jane Jacobs discussing this in the context of urban renewal -- "Blight" zone maps would just happen to coincide with "redlining" maps, with no formal coordination between the banks and the government required. [I'd be appreciative if someone more knowledgeable than I had something to say about this history.]

To test whether this issue is relevant here you'd have to relate the degree of discrepancy in loan quality with some measure of the degree of segregation of the local populace. I'm not familiar with the academic literature enough to say where this might be happening. But the system inputs for assessing the neighborhood of the property being considered had to come from somewhere.

[I'm also setting aside the abuse of mortgage instruments like NINA and NINJA loans, though the effect on automated systems would be the same -- garbage in, garbage out.]

Anon

[Approval loan]

Thanks for sharing your time and information with us! Tomas vaalue.com