 All right. Hi, everyone. Go ahead and get started. Welcome to the fifth full conversation on housing race and racism hosted by the Housing Lab. For those of you who are joining us for the first time, the Housing Lab is a place for reflection, innovation, and action. That's led by G-STEP students under the tutelage of Dean Amalia Andreos and IDC fellow Bernadette Baird-Zarrz. Now each of our sessions focuses on a specific theme with an expert practitioner as part of our efforts to explicitly integrate like an anti-racist agenda into our current lines of work. And today's conversation will be focused on fintech mortgage lending and racism with guest Tyler Hopper, who's a PhD candidate here at G-STEP. Now Tyler's work focuses on like social, technological, economic, and regulatory mechanisms contributing to racial disparities, segregation, and exclusion in urban areas. He has particular interest in mortgage lending, housing affordability, homelessness, neighborhood change, and educational equality. And he strives to design studies that inform policy and produce actionable results for legislators, regulators, planners, and advocacy organizations. Today he's currently a fellow at the Columbia Population Research Center and has professional experience in the public education, affordable housing development, and research sectors. Most recently at Skid Row Housing Trust and Columbia World Projects. He holds a master's in urban planning degree from Harvard to Harvard University's Graduate School of Design and a bachelor of arts degree in political science from Pepperdine University. Welcome Tyler and I will turn it over to you. Thank you very much, Eric. Your kind introduction and thanks Bernadette and Juan and others with the Housing Lab for inviting me to speak today. I will switch over to a presentation. Please let me know if something goes awry in terms of what you can see or hear. And I'll go ahead and share my screen before I get started. Okay, how's it looking? Well, thanks again to everyone who has joined and invited me. So the title of my talk today is the Racial Landscape of Bintech Mortgage Lending. Eric mentioned his introduction. And essentially this talk will be a broad overview of my dissertation research, which I am just wrapping up at this time. And essentially my dissertation research looks to be kind of the first set of studies assessing how fintech mortgage lenders, which I'll describe in a moment, differ from the traditional brick and mortar mortgage sector that we're all familiar with, sort of your local bank branch, the loan officer that you go speak to about a loan. There's a new emerging sector of mortgage lending. And so widely thought that it could differ in ways both good and bad from the traditional sector in terms of how fintech lenders distribute their loans to different racial groups, to different neighborhoods, to different metropolitan areas. And my work digs into those questions. So I will move on to, before I get into my results, I just want to give you, to make sure everyone's on the same page, a quick description of the difference between these two lender types, which I'll refer to as traditional and fintech. So the traditional lender is, you know, the kind of old school version that the loan process happens in person. So you, when you seek a loan, you might make a phone call versus schedule an appointment, certainly, but you'll probably meet in person at your local lending or banking branch. Once you provide, you know, a loan officer with your, you know, various documents about your financial history, they will conduct an industry standard underwriting process. They'll look at your FICO credit score. They'll look at your employment history. They'll compare, you know, the amount of the loan to your monthly income. And across the traditional lending sector, this is pretty standard and highly regulated. So most, most traditional lenders are using these same processes. And because this is a very sort of in-person, sort of paper-based process, it can take days or even a couple of weeks to receive an approval for your loan. And that's different from kind of closing on the loan and getting into your house. But even the initial approval, approval can take some time. So contrast that with the fintech lending sector. So fintech lenders, the end product is the same. It's a mortgage for a home. But the process is quite different. So with a fintech lender, you will never have to see a loan officer face to face in person. You'll never even have to speak with one on the telephone. You can go from beginning to end, all on the computer. So rather than going to a local banking branch, you will fill out application materials online. As I mentioned, you'll never speak to a human. You'll submit your documents to some sort of a cloud or a drop box. And instead of, you know, while a fintech lender probably still uses your books at your FICO credit score and your income, they also underwrite loans and assess the risk of each bar we're using machine learning algorithms that are kind of assessing a different set of data points. So it's been found that fintech lenders incorporate things like your social media, friend groups, your spending patterns, what stores you shop at, your internet search histories, your physical location and space. All of these things can be useful proxies for a fintech lender to determine the risk of a borrower. And those are used to supplement the standard underwriting data points. And oftentimes, would you submit all of this information to the online portal? The approval can be instant or within just a few minutes or hours. And so for the rest of this study, I'm going to be talking to you about, you know, results of the lending processes of these two different types of lenders, sort of broadly defined. And just as an example, you might think of a traditional lender, something like Wells Fargo or Bank of America or US Bank and a fintech lender, the most famous one and actually the largest mortgage lender in our country in the United States is Rocket Mortgage. Other prominent examples would be guaranteed rate or better mortgage. But some of the very largest lenders in the country are now fintechs. So any initial questions on kind of the difference between these two lenders that just something wasn't clear before I get into some of my study results? I was curious actually, how long has Rocket Mortgage been around? That's a good question. So the Rocket Mortgage, I believe started in 2012 and sort of it's not quite the same form as today, but really 2012, 2013 was the first year that, you know, fintech mortgage lending and the way we understand it today began. And I believe 2012 is the Rocket Mortgage first year. So it's a fairly recent phenomenon. Burr, yeah. Yeah. Was there any difference or were they very explicit about their initial target audiences, Rocket and others in the fintech or they was it kind of a, you know, we're undervalued grad students like us with low incomes but high capacities? And was there, you know, maybe you'll get into this later, but I'm just curious on their target audiences. Yeah. So it's a good question there. In the mortgage lending sector, there's typically not a big difference in the target audience of a mortgage fintech lender and a traditional fintech lender, where I'm sorry, traditional lender, whereas in like smaller personal loans, you know, a loan for home improvement or to buy a car or something like that, those fintech lenders in those sectors do sometimes explicitly try to target underserved groups. You know, they boast being able to provide credit to people, perhaps with bad credit scores or no credit histories because they're using these other underwriting metrics, but in lending or the mortgage lending sector rather, the target audiences don't differ too much. There may be differences in who is being advertised to on a, you know, broad scale. That's something that I don't believe there's any research on with fintech lenders yet, but there's no explicit target audience difference. Any other questions just on the differences before I move on? So with that kind of defined for everyone and established, I'll give a quick outline of my talk today. So why I'm going to tell you why fintech lending, you know, why this is important as a broad topic for us to understand why now is an important time to understand it before I get into the research questions and the methods of some of my studies. At the end, I'll provide some policy recommendations and broad conclusions. So why fintech lending? Why is this an important thing for us to be talking about and for me to be studying? There are three main reasons. One is that in the broader mortgage lending industry, there has long for decades upon decades been discrimination against borrowers of different races. There's also a lot of evidence on disparities across neighborhoods, you know, based on their racial composition or their location, and even metro areas based on how segregated those metro areas are. And so that's kind of the traditional mortgage lending literature. But layered upon that is a new and emerging literature about algorithmic discrimination. So because fintech lenders are using these machine learning algorithms to assess borrow risk, at least partially, you know, sort of the techno optimist crowd says, well, this is an advanced mathematical tool, it's going to generate more transparency, more, you know, objectivity in decision making. But there's also a big literature now on algorithmic discrimination. And many are finding that these promises of a more objective mathematical decision making process are not coming true. And in fact, you know, the same racial biases that are harbored by human beings are sort of inscripted into machine learning algorithms. So there's a fear that fintech lending will also include these same biases. So really quickly, I'll go through some of the more prominent literatures on these three topics. So, you know, Ken Jackson's Crabgrass Frontier is a classic outline of the federal bias in home ownership programs. The Color of Credit by Stephen Ross is a great book just talking about how not only were historically home ownership programs biased, but even after the civil rights era, those biases remained and continued, and they continue to this day. And finally, importantly, Guy Stewart's discriminating risk establishes that it's not just sort of racist attitudes from a loan officer to a borrower, but also actual discrimination, you could call it, baked right into the underwriting tool. So there are disparities that result from the statistical analyses of loan applicants. And it's not just individual applicants, but it's the neighborhoods they're targeting for their homes that are also experiencing disparities across the lines of race. So a famous recent book, The Color of Law by Richard Rothstein outlines the way that neighborhoods have been redlined over throughout history and no other talks in this housing lab series have also gone over this same topic. And that's important not only because it hurts people's chances of owning a home, but we've found that opportunities are distributed spatially, right? So it's not just sort of your personal financial health that's being impacted, but also your opportunity to access things like jobs or healthy air or parks, things of that nature, good schools. And then lastly, Matt Desmond's very prominent recent book I think really beautifully shows how struggles to stay in housing or to own a home are tied with just your day to day struggles in life and your opportunities in a very intimate way. So on these more traditional sort of literatures about housing, we can add algorithmic discrimination literature. So Kathy O'Neill and Weapons of Math Destruction, sounds like I'm saying that with a lisp. She shows really beautifully that it's not so much that the math behind the algorithms are biased, but human beings are still feeding data into them and that when humans are supplying the data decisions are being made, even before an algorithm is run, maybe determining if a borrower is a risky borrower that can bias the results. We also see Ruha Benjamin's work, race after technology, you know, this discrimination has very explicit racial outcomes, racialized outcomes, and there's also evidence that poor places and communities of color are being kind of targeted or disadvantaged by these processes. So, you know, this literature really leads me to to study this question, but why now? There are some reasons why right now is especially a good time to be studying fintech lending. So first off, home ownership has a huge role in inequality, just income inequality or wealth inequality rather. There has been a hot legal battle over housing discrimination, you know, slightly proceeding, but especially during the Trump administration. And regulatory agencies have struggled to regulate fintech lenders, really at all. And so just a few data points on home ownership's role in inequality. You know, Sandy Darity, who's a prominent scholar of wealth inequality, has found that if we close the home ownership gap, the racial home ownership gap, we've reduced the racial wealth gap by about a third, which is incredible. You know, we see between renters and owners that the net worth of these groups in our country is staggering. Renters have, you know, almost nothing for their net worth, whereas homeowners on average have, you know, hundreds of thousands of dollars. So you can really see how important it is in the United States, especially to own a home for your financial health. And finally, we're in a really depressing moment where black home ownership is now as low as it was before the enforcement of the Fair Housing Act. So we had this watershed moment in the late 60s when we were, civil rights legislation was, you know, really being enacted left and right. And we're now at a point where the progress that was made in the 70s, 80s and 90s is all the way back to zero and black home ownership rates are just as low as they were before discrimination was outlawed. The legal battle over housing discrimination that I mentioned earlier has mainly been led by HUD, which we usually see as sort of a positive force in urban areas. But under the Trump administration, and we've really seen HUD try to backpedal on not just algorithmic discrimination, but really any sort of civil rights or protection for communities of color. And then lastly, I'll try to go over this quickly for the purposes of time here. But the Consumer Financial Protection Bureau, this is our country's main fintech regulator. Since fintechs really gained prominence around 2012-2013, this agency, which was created, you know, Elizabeth Warren is famous for leading the creation of it, after the financial housing crash was created to protect consumers from, you know, harm in the financial sector, has done almost nothing to rein in any potential discrimination in fintech lending. Since 2013, the CFPB started what was called the Trial Disclosure Program. This was a, they basically asked fintech lenders, fintech mortgage lenders, to voluntarily submit their data and their methods of risk assessment to regulators. And surprise, surprise, they had no takers. So a few years later, they said, okay, if you will disclose your consumer data and your methods for risk assessment to us, we'll give you a no action letter, essentially saying for a number of years, I think it was two years at that point, if you are found to discriminate, there will be legally no action against you like in exchange for being transparent. Again, there were no takers. In 2018, the CFPB said, okay, we will create a digital sandbox where if fintech lenders throw their data into this digital sandbox, we regulators will not even be able to see the data. It'll be totally hidden, but we'll be able to at least run sort of detection, you know, statistical analysis to see if the data would result in biases and in exchange for the no action letter. And I believe there was one taker at that point, but I'm not sure if they even proceeded to fully utilize it. And so in 2019, they said, okay, we'll do a trial sandbox. So your data will only be in our totally anonymous sandbox for just a few months and if you let us analyze your data just for a few months, we'll give you a no action letter saying we can't legally come after you for discrimination. And again, this has been totally unsuccessful, despite the CFPB just bending over backwards to say fintech lenders, you know, we'll give you complete immunity to discrimination if you would submit to any regulatory scrutiny. I would say it's kind of bogus to make this all voluntary, but we can get into that more later. Okay, so with that established, I want to get into my the research questions of my study, right? So hopefully you agree with me that this this fintech lending sector deserves some scrutiny from researchers and policy makers eventually. So I'm essentially asking three questions here, and I'm just going to provide just a snippet of my findings related to each of these three questions. So the first one is, compared to traditional lenders, are fintech lenders discriminating against racial group? So a borrower from a given racial group. Secondly, compared to traditional lenders, do we see disparities across different neighborhoods based on their racial composition with, you know, among fintech lenders? And then thirdly, does metropolitan segregation have an impact on fintech lending? And each each of these questions is really embedded in, you know, a long history of research on the traditional lending sector that has found discrimination. So if what, you know, optimists say is true and these algorithmic methods are more objective than a human loan officer's decision making, hopefully we will see, you know, lower levels of disparities in each of these categories for fintech lenders compared to traditional lenders. Now I'll go over my analytical approach quite quickly. If any of you are in the housing discrimination literature or want to talk more about, you know, quantitative methods, feel free to ask during the Q&A or ping me afterwards. But essentially, for all of these studies, I'm using regression methods to analyze millions of mortgage loans. And conceptually, I'm finding the odds that a borrower will receive a subprime loan while holding constant there that borrower's characteristics, characteristics of the lender, the neighborhood the homes in and metro area the homes in. That's kind of the big picture here. And so the outcome that you're going to see in all the charts to follow are basically saying, what are the odds that this borrower gets a subprime loan? There's also I also look at whether loans are approved or denied. The good news is fintech lenders do a little bit better than traditional mortgage lenders on how often they approve and deny loans. But we find less favorable results as you'll see with subprime lending. And just to give you a conceptual idea of how, you know, what are the kinds of things that I am holding constant? So, you know, in these quantitative studies, I start with a loan application. These are publicly available applications. The data is all online from the home mortgage disclosure act. And I'm most interested in the race of the borrower. But to make sure when I compare, say, a black and a white borrower, that those two borrowers have a very similar profile. Otherwise, you know, I control for their income, their their loan to income ratio, you know, whether or not a co-applicant signed on the loan, whether it was a jumble loan, all these sorts of things. Similarly, with the lender, I'm always interested if the lender has fintech or traditional, but I hold a few things as constant about lenders to make sure we're getting an apples to apples comparison. With neighborhoods, I'm most interested in the racial composition of the neighborhood, but I also control for the neighborhoods of unemployment, whether or not there was a lot of subprime lending during the housing crisis in that neighborhood, for example, percentage of homeowners, etc. And finally, a similar set of covariates are controlled for at the metropolitan level. And I'm really interested in the level of segregation at the metropolitan level. And so, with all of these things controlled for, my hope is that when I say a fintech lender distributed a subprime loan to this borrower, that the borrower is, you know, essentially, when I consider the race of the borrower, whether it was a fintech lender, the type of neighborhood, all the other characteristics of those borrowers and where they're trying to get their loans are held constant. So we could say these two borrowers are equally qualified to get this loan or to get a prime lending rate, which means they're not a high interest rate, as the one I'm comparing them to. So I'll quickly go over the sample. So I'm looking at all U.S. mortgages, 2015, 2016, 2017. And this is a bit of a mouthful, but for owner-occupied, one to four family first lien loans for home purchase in the U.S.'s largest 200 metropolitan areas. And the reason this is important is because, you know, about half of the mortgage applications in the country are for refinancing. And I don't include these in my studies because the underwriting process for a home that someone already purchases is very different than for a new home. Also, you know, while the top 200 metro areas include 75 to 80 percent of the U.S.'s population, I am excluding some loan applications from more rural areas. So in my sample, I've got around 7 million loans from traditional loan applications, rather, from traditional lenders and over 600,000 from fintech lenders. Of those, you know, subsamples, fintechs and traditional lenders approve about 85 and 90 percent of total loans. And they have very similar rate of subprime lending, in fact, about a little over 7 percent for each of them. Their approved loans include subprime rate. And I should have defined that. Essentially, it's about the home mortgage disclosure act to find subprime as around 3 percentage points above the prime rate. So if everybody's getting a 4 percent interest rate, subprime is 7 or more. A couple of other, you know, charts here just to give you an idea of the wider industry. So we see here that fintech lender lending is growing. So in 2015, the beginning of my study period, only about 6 percent of all loans were originated by fintech lenders. And by the end, it's almost 10 percent. And that trend continues today. So this is, it's a rapidly growing sector. And you see over 3 percentage points, which represents hundreds of thousands of loans in just three years. And disturbingly, in the greater sample size, subprime lending also seems to be growing. Now, after the housing crisis occurred, you know, when we're looking at the years 2010 and 11-12, there was almost no subprime lending. Regulators were increasingly scrutinizing lenders, but lenders themselves really tightened their belts and made underwriting standards much stricter. And it was, you know, really close to zero. But we see today that rates of subprime lending in general seem to be growing, which is a bit worrying. So the results. So as I spoke about before, I'm going to give you three batches of results here. The first will be about comparing applicants of different races, then we'll compare different neighborhoods, and then we'll compare metro areas based on how segregated they are to see if fintech lenders have fewer racial disparities than traditional lenders. So here is the first batch of results. So this is looking at borrowers from different racial groups. And our vertical axis here is the probability that they'll receive subprime loan. So we see, you know, ranging low of about five percent for fintech lenders to Asian borrowers to a high of about 10 percent probability of receiving a loan for Latinos at traditional lenders. And you see, again, I can't emphasize this enough, we've done everything we can to make sure that this white borrower here and that Latino borrower are apples to apples the same incomes, looking in the same areas, the same, you know, employment histories, etc. So these differences are as much as we can control for attributed only to the differences in their race. And if you think stand out, first off, we see that Latino borrowers are by far the most likely of any group to receive a subprime loan compared to equally qualified borrowers from other groups. And we see whites and Asians similarly have very low rates of subprime loans with black groups somewhere in the middle. But beyond just these kind of overall trends, we're interested in the disparities between a white borrower and a similarly qualified non-white borrower. And we will hope that fintechs would do better. So we would hope that a fintech lender would close that gap a bit. However, when we compare, for example, white borrowers to black borrowers, we see that for traditional lenders, that gap is about 19%. So a 19% growth in subprime loans between white and black borrowers. And we see that corresponding gap for fintech lenders is about 10 percentage points higher in terms of growth. And this, this is not good. Fintech lenders, you know, with their sort of overtures about objectivity, are actually found, let me admit, never mind, are actually found to, you know, increase disparities in subprime lending between white and black borrowers in particular. The disparities between white and Latino borrowers are about equal for both groups, but it is a problem that the disparities remain at all. So a few takeaways from this comparison of racial groups. First off, we see that both white and Asian applicants are advantaged by rates of subprime lending for both groups of lenders. We also see that, you know, both fintech and traditional lenders are kind of similarly, disadvantaged Latino borrowers, but fintech lenders disadvantage black borrowers, even more than traditional lenders do. And just to give you a sense of the magnitude here, this is, you know, this last bullet point is just an analysis of my subsample, which is probably only a third or so of the total mortgages each year for various reasons. If white applicants receive subprime, subprime loans at the same rate as comparable black applicants, they'd receive about 7,000 more subprime loans each year from fintech lenders and about over 50,000 additional loans, subprime loans, you know, from traditional lenders. So you can see over time, we're talking about, it's not only disadvantaged borrowers of color, but we're advantaging white borrowers comparatively, right? So you see that we certainly don't want anyone who doesn't deserve it to receive a bad interest rate, but you can see cumulatively, we're talking about hundreds of thousands of white households that are getting great interest rates compared to similarly qualified black and Latino borrowers. Okay, now let's move on to see how neighborhood racial composition impacts the story. So this chart is a little busy, so I want to orient you to it. Same as the last one, our vertical axis here is the predicted probability that a borrower will receive a subprime loan. Across the bottom, we have each racial group of borrowers, but the bars represent different racial categories for each neighborhood. So the darkest bar on the left for each group is a neighborhood with less than 20% non-white residents. So this is a white neighborhood, essentially. And then the lightest colored bar on the right hand side is a neighborhood with more than 80% non-white residents. So essentially we're moving from the whitest to the least white neighborhoods. And this chart I should also note, this is the traditional lender chart. So we see here that for traditional lenders, as a neighborhood gains non-white residents, the neighborhood is also more likely to assign subprime terms to a borrower. We also see kind of the same pattern of white and Asian borrowers with very low rates of subprime loans and black and Latinos, especially Latinos with higher rates. And just to kind of exemplify some of the things going on here, well, I'll say this first, keep this pattern in mind. Keep this general pattern. And we'll switch over to the FinTech chart, which we hope would be a little flatter, right? We wouldn't see that increase in subprime lending as a neighborhood gets less white. However, the chart looks much the same. So this is the FinTech chart here. We do see interestingly that the least white neighborhoods over on the right do a little better than kind of the 60 to 80% non-white neighborhoods. But these differences for most groups aren't really statistically significant. And you can see with the error bars there. But a few things stand out here. First off, this is eerily similar to the traditional lending pattern. It seems that even though a FinTech lender maybe has their headquarters in San Francisco and they don't have a loan officer, you know, in your neighborhood, that might say, oh, that neighborhood over there, for whatever reason, is a risky neighborhood. It's stigmatized in my mind. It's not that they would say that out loud. Still, something in the underwriting is disadvantaging neighborhoods of color. And we see here, as I put this kind of red bar on, you know, you can really see how race and a borrower and race of neighborhood interact. So a Latino borrower who's seeking a loan in the whitest neighborhood is still more likely to receive a subprime loan than a white borrower seeking a loan in the least white neighborhood. So we see this compounding of disadvantage, or you could look at it as advantage to white and Asian borrowers here. You know, the race combined with the neighborhood is especially troubling for black and Latino borrowers. So a few takeaways here just to kind of reiterate. For both lender types, the probability of receiving subprime credit increases as the neighborhood gets less white. Latinos, again, as with kind of our just individual borrower-based analysis, they're also, they're still most likely to receive subprime credit in every neighborhood types. And as I said, you know, the Latino borrower seeking a loan in the least kind of diverse or the most white neighborhood is still more likely to get a subprime loan than the equally qualified white borrower, no matter where they're shopping. Okay, the final graphic here, we're going to look at the segregation of a metropolitan area. So, you know, many scholars have assumed that the reason we see neighborhoods of color disadvantaged is because they are in very segregated metro areas. And so there are clusters of these neighborhoods that a loan officer can see and kind of identify as, you know, because of a racist attitude that that's a neighborhood where we don't want to lend to its too risky. And so segregation on a metropolitan level has been found to have a big impact on how often a loan is given subprime terms. So here, again, same vertical access, probability of subprime rates. Now along the horizontal axis, these are percentiles of metropolitan level segregation. For those who know the segregation literature a bit, this is the dissimilarity index. I've run similar tests with multiple other indices that yield similar results. But we see that in the 10th percentile, so a very non segregated metro, this might be San Francisco or Seattle, pretty similar rates of subprime lending for both lender types, about almost 6% for traditional lenders and 6.7% for fintech lenders. But as we get to more and more segregated metros, you could think of that 90th percentile as a Chicago or a Milwaukee or a Buffalo. We see a huge drastic increase for the traditional lenders in their rates of fintech lending. So almost a 4% increase for an equally qualified borrower, think of it as, you know, in Seattle versus in Chicago in terms of how likely they are to receive a subprime loan from a traditional lender. And finally, some good news, our fintech lenders do much better. So that curve is much flatter. We see only about a percentage growth in subprime lending to more segregated metros than least segregated metros. So this is good. This means that maybe because fintech lenders don't have loan officers on the ground that might be biased against large areas of, you know, concentrated groups of color that they are less likely to sort of discriminate in their loan distribution against these sorts of metro areas. So a couple of takeaways before I get into my policy recommendations here. Again, that curve is much less steep for the fintech lenders. So for some reason, you know, we're seeing fintech lenders treat very segregated metros much better than traditional lenders. And if you were thinking, how could a fintech lender kind of discriminate against, you know, neighborhoods of color at the same rate, remember, our second set of charts as the traditional lenders did, but they do so much better in metro areas that are, you know, more segregated, you know, it could be because in the underwriting algorithms of fintech lenders, they don't have a person on the ground who might be able to observe a big cluster, big segregated cluster of neighborhoods, but there's still some proxy in their underwriting that does disadvantage a neighborhood of color, even if it's not in sort of a big segregated area. So it's not completely illogical that those two charts don't add up. Okay. Just got a couple more minutes here. So I'll quickly go over these results, just your key takeaways, and then talk about policy for a minute or two before I field some questions. So big picture results, right? When it comes to how individuals and different racial groups do at each of these two lender types, essentially fintech and traditional lenders have the same rates of subprime lending across racial groups, but actually they disadvantage black borrowers even worse than traditional lenders. So it's bad news there. Similarly, when it comes to neighborhood level disparities, we see fintech almost mirroring the traditional lending market. And then lastly, the good news, when it comes to metro segregation, and again, we're talking about tens of thousands of loans in each metro, this really could structure the advantage and disadvantage in different neighborhoods. Fintech lenders don't seem to be lending with great disparities between more and less segregated metro areas. And that could have a positive impact on how those populations within those areas are able to move around space and access different neighborhoods. So to the policy, a few ideas here. One is that all of this sandbox and voluntary regulation stuff I talked about at the beginning has just got to go. Like every big bank or lending entity that is in the traditional lending market and has to be regulated by the FDIC or other big regulatory agencies, fintech lenders, they have to be regulated by someone that will require them to submit their data for scrutiny. It's just ridiculous that this is voluntary, given that this 2017 was 10% of the lending market representing tens of millions of loans a year. Secondly, as I mentioned before, all this data, most of it was from the Home Mortgage Disclosure Act. But a big problem is although the HUMDA data requires lenders to say the race of the applicant, many of them don't. And there's no enforcement mechanism to do this. And this hurts our ability to run studies like this one. The vast majority still are reporting race, but there's this unknown category that I didn't talk much about in this presentation. But there's really no penalty if the race is not collected. And secondly, credit scores, maybe the most important factor in determining if the loan is given or not, are not publicly available. However, in my study, I should note I was able to kind of create a mathematical instrument that proxies for a credit score and has actually been found to be more conservative than the actual credit scores in these types of studies. So it may mean that my results actually under put some of those discriminatory or disparity related findings. Lastly, the Community Reinvestment Act, we're all planners, we should be familiar with that or at least people interested in neighborhoods. A great way to, I think, regulate these fintech lenders would be to fold them into the CRA. Now, the hard part about that is the fintech lender has an office in Detroit for, you know, Rocket Mortgage or in San Francisco for others. But the CRA regulates banks based on where they're located. So that's not going to work because that fintech lender in San Francisco is, you know, giving out loans in Florida and Texas and anywhere else. So the CRA needs to be updated because increasingly we live in a digital world and the days of the brick and mortar, you know, shaking your hands with your local bank branch officer are coming to an end. And we can't have these regulatory mechanisms that tie the location of a building to a regulatory zone. So the CRA needs to find a way to encourage fintech lenders to land equitably to all communities, even if their office is on the other side of the country. And with that, I would like to say thanks for your time. And yeah, I'd love to answer any questions. Well, that was, I don't know really how to respond. That was just really kind of eye-opening. I had no idea about some of that, especially with, it was called a HMDA, correct? Yep, the Home Mortgage Disclosure Act. So race is not like, race is sort of on there, like on the applications, but they're not required to disclose it or? Well, they're required, but it's not enforced. So the vast majority of applications do include the borrowers race. But what we're seeing is that fintech lenders, especially, you know, you might have an online form that people are filling out and they choose just not to fill out the race column. And those loans are still, you know, approved or denied. So it's, you could think of it as something like 85 or 90% of the loans do have race, but that 10% is millions of loans. And it's just crazy that this isn't enforced in some kind of way. And you said credit scores are not publicly available. I mean, what is the benefit to publicly making credit scores public? Obviously, to get more, like, more accurate data, correct? Yeah, so, you know, oftentimes, you know, in response to findings like mine or dozens of other research studies, the banks will say, well, if you saw the credit scores of all these applicants, you would see that the disparities that you're detecting are in fact attributable to their poor credit histories. And then so researchers say, okay, well, publish the credit scores and we'll, we'll determine if that's right or not. And bank lobbies have bought that tooth and nail for decades. So, you know, they've claimed that the reason is because if you put a credit score out there, it could kind of compromise the privacy of a borrower. But all of this data is anonymized. And, you know, many people smarter than I have argued that it won't compromise their anonymity in any way to have credit scores posted. But like I said, there are kind of ways to proxy for a credit score, which, you know, I do include in my studies. Hopefully, I'm being more conservative even than I would be if a credit score was included. But yeah, it's a big challenge. You can purchase credit scores, by the way. There's one other group of researchers that are over at UC Berkeley who are kind of going on a similar wavelength to me here. And they, I called them and talked to them about how they got their credit scores. Can they share them with me? And they could not. And they said that to buy five years of credit scores from Xperia and cost them something like $140,000. Out of reach for most researchers. Yeah, most definitely. And if anyone has a question, I'll let you go. But I had a follow-up question. Should I stop sharing my screen, by the way? Or I don't know what you're talking about. I just want to give everyone an opportunity to ask a question, if they would like, before I do my follow-up. Yeah, no, I had one question. Tyler, thanks so much for your presentation. I feel like I've heard what the topic of your dissertation is for a while, but this is the first time I'm hearing it come to life. So it was great to hear. So I was curious whether do you have any hypotheses surrounding the kind of counterintuitive finding that rates of subprime lending were slightly lower in neighborhoods that were 60 to 80% non-white versus 80 plus percent non-white? Yeah. So there's sort of, I think that there's an urban to rural thing that can go on with those racial percentages where when you're looking at the most white neighborhoods, sometimes, depending on the geography you're looking at, those can actually be very poor rural white neighborhoods. And so increases in diversity are just non-white residents. Sometimes have the opposite effect that you would expect. That 80% and above versus 60 to 80%, to be honest, don't have a great hypothesis of why fintech lenders in particular would kind of disadvantage them slightly more than the very non-white or above 80% non-white neighborhoods. But I will say the difference there is not statistically significant. If you predicted how likely a borrower was to get the subprime loan, your error overlaps quite a bit. So statistically, those two groups are about the same really. But it's a good question. Thanks. Okay. I'll go ahead. Yeah. So I was just really curious as to your method for calculating the proxy credit scores if you for those that you do not have access to. It's a good question. So I'll try not to get too much detail with the data. But basically, in that home mortgage disclosure act data, there when a lender denies a loan, they can check a box that says the reason why they denied it. And there, I think there are seven or eight of those reasons like poor job history, you know, loan to income ratio was off. One of those reasons is poor credit history. And what you basically do is you take a sub-sample of the total data and you run a regression to predict like how likely borrowers to have a poor credit history. And then using the estimates from that data, you then kind of like use the estimates and predicting your full model for each borrower, how likely it was that a lender would deny them for that credit history box. So the way it works is, you know, it's a proxy for credit score. It's really like how likely someone was to be denied because the bank said they had poor credit. It's not an actual score. It's a range of likelihood. So that's how that works. And it's not perfect by any means, but it, like I said, it could actually be a little more conservative than including credit scores, which better than, you know, making false claims about disparities. It's incredible. Yeah, Bre. Yeah, say thank you so much. I mean, we've been talking so much at the lab and in GSEP and you know this about, you know, the wealth gap and how that's compounded through our practices and through the practices we inherit. And to have something so spot on is really wonderful. So thanks for, for showing us and connecting it back also to the previous conversations. One kind of small question and then a speculative question. What do you think would happen if you looked at refinancing data or just real areas, assuming world real areas are generally like super white, could exclude them, but refinancing, what do you think the dynamics might be there? Might it be ameliorated somehow? And then secondly, speculatively, like the world we live in is imbued with white supremacy and racism, thus the data that we have to work on even machine learning that sucks up a bigger or can, can engage with a wider swath of data and its movements is also reflects those structures and reflects those the ways that we live as a society. And so the results of stuff are going to be problematic too. I know there's, you know, what's, what would be if you could suddenly turn into an activist scholar with your own, you know, center for change? Elizabeth Warren 2.0 with the right political support in place. How would you even start to untangle that? Or, you know, is there, are there kind of quick and easy fixes in the, in algorithms themselves that might, you know, just to be able to correct for those things? Or, you know, would you start it at the building blocks of what data and how, and how, you know, how they're coded? I don't know, where, where would you, where would you go? Yeah. And I'm expecting this from you in like three years, by the way. Yeah, super easy question. Let me just give you the answer in a nutshell. Now, so let me answer your first one first. So on refinancing. So I have done all of these same analyses with refinancing loans included in the analysis. It, you know, essentially refinancing loans are much less likely to receive a subprime rate, because the person already owns the home and they've been making payments on it, expensively for years or months. So lenders are, are much more likely to approve a refinancing loan and to give it a prime rate. But with that said, you still see the impact, especially a borrower race on the refinancing loan market. Not so much of the neighborhood race has a little bit less of an impact, because I think the asset's already there. And there's a history of, you know, the neighborhood not having an impact on the, the financial health of the asset, I guess. But, you know, the patterns aren't so different, but the rates of subprime lending are much lower for refinance loans. And so I have a paper coming out in Housing Palsy Debate, hopefully within the next, I don't know, however long it takes them to correct the proofs where I talk about refinancing loans a little bit. In terms of your second question, yeah, it's so hard, right? Like, all this big data and it's reflective of the existing biases in our society. And you know, I think that it's not so much that an algorithm itself is biased, right? An algorithm is just a, you know, powerful regression, essentially, that's identifying patterns, but it's that the data that's fed into the algorithm reflects biases. And the way you use the algorithmic results to kind of inform your decision making can, their computer scientists are doing fascinating studies on that question. But in terms of just the data inputs, you know, there are ways, even right now, that people, that people are sort of accounting for this. I think a good example, the sad one, is with COVID right now. So regulators have introduced legislation to say that at least during the time of the pandemic is happening and the economy is shut down, the credit score aggregators, you know, experience, trans union, et cetera, should exclude a person's financial history during this time. Because if they missed a payment or were evicted or whatever, it wasn't because they were not a good borrower or, you know, it's out of their control. And so I argued in trying to get an op-ed published, which is harder than you think, that basically says it should be the same for fintech lenders. You know, they may not use credit scores to the same degree, but they should have to disclose to regulators that, hey, everything in this applicant's data history between February of 2020, and whenever the end of this thing is, was not factored into our assessment of their risk. So there are things, tangible things that I think can be done to make sure the data that's fed in doesn't harm a borrower who doesn't deserve to be, you know, have their likelihood of getting a loan harmed. So that's, you know, that's kind of one example, getting at what you're asking. But I agree. It's an insane web of big data out there now. And when you start factoring in a social media friend group analysis to a loan application, I just think you're bound to find proxies for race, you know, like they actually like lenders, I don't think so much in the mortgage lending, but fintech personal loan entities have measured the aggregate credit score sort of of your Facebook friend group profile to determine, you know, kind of a birds of a feather flock together thing. And so, yeah, you could say that that's, it's just based on who you're associating with, but you can't help but see racial patterns there, depending on, you know, how people group themselves with friends and colleagues. So I think regulators need to have the opportunity just to see this data at the very least. Does anyone have some additional questions? Okay. Tyler, this this has been absolutely incredible. Thank you so much for your time. And for actually, I just printed your paper that Bernadette posted in the chat. So I'm looking forward to reading that momentarily. Oh, great. Yeah, absolutely. Again, thank you so much for your time. It's our pleasure, our pleasure. And I would encourage everyone to register. I believe Juan posted the registration form in our chat for everyone who is not registered. Please do so. You'll be updated with our next conversation. And we'll see you all soon. Thank you so much. Happy Friday. Thank you. Bye bye.