Welcome to today's webinar, CFPB Authority to Protect Consumers: Reading the Fair Lending Tea Leaves. My name is Jennifer Purdie, and I'm your host, Wolters Kluwer. This is our third and final pre-conference webinar before our 26th annual CRA and Fair Lending Colloquium. I'd like to explain how we got here. As you know, we hosted the last two colloquia virtually, and when we shifted the 2021 colloquium from an in-person event to a virtual event, there were a few sessions that we couldn't quite fit into the abbreviated schedule. Since the content was so rich and the speakers were so remarkable, we decided to keep these sessions in the queue and offer them as webinars. With over 1,100 people registered for today's session, we remain grateful to all of you for your continued support of this event.
This year's colloquium will be held November 13th through 16th in person at Caesars Palace in Las Vegas. Registration is open, and we encourage all of you to plan early to take advantage of the hotel discount. The colloquium is the financial industry's most prominent forum focused exclusively on regulatory trends in CRA, HMDA, and fair lending compliance. We can't wait to welcome you in person in November. For today's session, Tim Burniston, Senior Advisor for Regulatory Strategy with Wolters Kluwer, will moderate a discussion with two partners from the law firm of DLA Piper, Isabelle Ord, and Austin Brown. Austin advises institutions regarding consumer protection compliance, enforcement, and litigation issues. His practice focuses on investigations and enforcement proceedings, class action litigation, compliance advice, examination support, internal corporate investigations, and transactional diligence.
Isabelle is a trial lawyer who resolves complex and high-stakes disputes for national and international businesses and financial institutions. She is the co-chair of DLA Piper's class action litigation practice and the co-chair of the firm's U.S. financial services sector. During their discussion, we'll focus on several of CFPB's recent fair lending initiatives and directives. Isabelle and Austin will also provide guidance on how financial services companies can incorporate the CFPB's latest guidance into their compliance management systems. During the registration process for today's session, many of you submitted questions you would like the panel to address. The panelists are planning to address many of those questions, and what they may not have time to get to today will be added to the colloquium agenda this fall.
At this time, I would like to turn things over to Tim Bernerson, the moderator for today's session and the master of ceremonies for the CRA and Fair Lending Colloquium. Tim?
Thank you, Jennifer, and thank you to all of you who are able to join us today. The CFPB has been very active over the last 18 months and is steadily and frequently issuing policy statements, advisory opinions, circulars, regulatory changes, revisions to examination procedures, blog posts, guidance on a variety of consumer protection matters that are of interest to institutions subject to the laws and regulations administered by the CFPB and their supervision and enforcement, and of course, the public. There is certainly more to come. The forthcoming 1071 rules are just one example, as are proposed rules under 1033 of Dodd-Frank providing consumers with access to their financial records and interagency regulations on quality control standards for automated valuation models, just to name a few.
Let's not forget, the CFPB was created to provide a single point of accountability for enforcing federal consumer financial laws and protecting consumers in the financial marketplace. They are doing their job, and many are glad that they are. At the same time, they're keeping the industry very busy, and their work commands attention. That's really why we're here today. Today, we focus on a few of the many releases relating to fair lending that have either come directly from the CFPB or are part of interagency initiatives that include the CFPB. All of them have a significant impact. All of them are important. All of them have gotten a lot of reaction from the financial services industry and the advocacy community.
Our objective is to make you aware of these developments, cover the important aspects, offer practical advice and guidance for incorporating them into your proactive fair lending programs and compliance management systems. As Jennifer mentioned, we thank you for sending your questions in advance. You'll see that Isabelle and Austin will address many of them relating to the topics we are covering today in their commentaries. There are also some recent fair lending developments as well, some coming from the CFPB directly and others from the broader regulatory community. We will cover them in depth at our colloquium in November, but we may have some time to touch on a few of them today. Our time is limited, so I'm going to jump right into our first topic. Our first topic is unfairness and fair lending.
It will cover the most recent change to the CFPB's examination procedures indicating that the agency will consider discrimination or improper exclusion to also be an unfair and deceptive practice. This is a big change, and it has a lot of sweeping implications. Isabelle, I'll start with you. Maybe you could provide us with just a brief summary of the CFPB's change here, and how would you describe the CFPB's new unfairness exam procedures regarding discrimination?
Thank you, Tim. It's nice to be with everybody today, and thank you to you and Jennifer and the rest of the team for having me and Austin. We're looking forward to this conversation. I'll jump right in here on the new guidance, which is clearly a very broad invitation of expanded authority to address unfair lending issues with a particular focus on discrimination issues. It is sort of a natural evolution, I think, from things that have been out there for a while, but we'll talk a little bit about what the focus is and then some things for you to think about.
We're reading the tea leaves today, so we'll talk about what we can pick out of the information that's publicly available, what we can sort of understand from what else is going on with the economy, with the world, and with the CFPB in general as to where some of this is going. We can start, of course, with the words of the director of the CFPB on this issue. He said, "We will be expanding our anti-discrimination efforts to combat discriminatory practices across the board in consumer finance." That's pretty broad, and it's fairly expansive in terms of what their focus will be. Essentially, what the CFPB has said is that this will primarily begin with changes to their supervisory practices.
Here's the quote, "To better protect families and communities from illegal discrimination, including in situations where fair lending laws may not apply." It is the second part of that statement that really gets my attention. This is something that the industry has also focused on, and we'll talk a little bit about where trade groups in the industry are and some of their concerns about this next evolution. This is really a filling of the gap into an area that is not already covered by certain anti-discrimination or financial services laws and may cover things that are either outside of existing law or that are technically lawful. That is why there is some concern about what needs to be done both in terms of compliance and also issue spotting where these trends are going to go.
As I said a minute ago, I do think it's a natural evolution from some of the prior focus on disparate impact and some of the general overarching principles that the CFPB has focused on, principally preventing lasting harm, some of it coming out of COVID-19, but I think more generally we can even hearken back to the 2008 financial crisis, and there's some thinking that there are still consumers out there who suffer lasting harm from what occurred during that time period, also addressing inequities very broadly. We're beginning to see through this new pronouncement how some of these things are going to be put into practice in general. There is some guidance. You know, in the beginning, the CFPB was sometimes criticized for not really giving information about how they would approach certain things.
They have provided us with some guidance and a few examples to give us a sense of at least where we can start with the tea leaves. The beginning place is what the standard is for the Unfairness Act under Dodd-Frank. An act or practice is unfair if it meets three different criteria: if it causes or is likely to cause substantial injury to consumers, the injury is not reasonably avoidable by consumers, and the injury is not outweighed by countervailing benefits to consumers or to competition. The CFPB has told us a little bit about how they think this will go in the context of challenging discrimination. Of course, we talk about substantial injury. It must be more than something that is merely trivial or speculative, and that includes more than just emotional impact.
It really generally is something that involves monetary harm, although the CFPB has now indicated that it may also involve forgone monetary benefit or denial of access to a product or a service. You can see pretty quickly how that might come up in a discrimination type of context. There is a clear reiteration that actual injury is not required. A significant risk of concrete harm is also sufficient, and it may even be enough if there is a small amount of harm to a large number of people who have been deemed to be harmed. The CFPB has further indicated that if there is, putting aside the fact that just an emotional injury is probably not enough, that could come back into play as part of the circumstance, particularly where there has been discriminatory conduct that has impacts on emotional harm or dignitary harm.
Again, it's pretty easy to see how that might come to the forefront where perhaps somebody has been denied credit or has received credit or services that are lesser or different than other folks. Well beyond what we would see in a traditional type of Article III situation, there's a lot going on on the legal side with the U.S. Supreme Court and in the various district courts and courts of appeal about what it requires for Article III standing, what is concrete harm, what to do when it's just a bare statutory violation. This is a lot broader than what you would see in a federal court pleading situation. Keeping in mind, of course, that in a pleading situation, these are just allegations and they can be attacked.
In this circumstance, many of the things that the CFPB has outlined would not be sufficient to create Article III standing in a federal court case, but may be sufficient in the context of their examination and potentially their enforcement criteria. The factor that's most interesting to me is the second one, that injury is not reasonably avoidable by consumers because the CFPB is not hiding the ball here at all. They have said several times that consumers cannot reasonably avoid discrimination. There it is. That's a pretty powerful marking out of a position as to how that will be an injury to consumers. They have also made clear that even if the practice or the issue is something that many others in the marketplace are doing, that's probably not going to be a good enough defense for you.
In fact, that may indicate that the practice probably could not be reasonably avoidable. That is really going to be the area where they focus. The third factor, of course, is that the injury is not outweighed by countervailing benefits to consumers or to competition. This often brings up things that we would raise in a situation about the cost of compliance or the need to vastly expand current efforts. It also focuses a little bit on what will happen in the marketplace. Does the conduct that is at issue actually result in more competition, better access to products, lower prices for consumers, and so forth? I do think it is a little bit hard in the context of discrimination to argue that it is okay because it somehow means that prices are lower or there is more availability of products.
This is definitely a spot to watch, and I do think we'll start to see more around this as it starts to come together.
As this gets rolled out, where do you think the CFPB may focus their attention first? I mean, there's, as you mentioned, applying this in practice raises a lot of questions about what you need to do to prepare, and we'll go there in a second. You wonder about, is this more lending related, deposit account related, other services, everything? Where do you think that the primary area of emphasis and application might be?
These are still early days, but we can look first to the words of the CFPB, and here's what they said. The CFPB will scrutinize discriminatory conduct that violates the federal prohibition against unfair practices, including closely examining financial institutions' decision-making in advertising, pricing, and other areas to ensure that companies are appropriately testing for and eliminating illegal discrimination. When you take just that statement on its own, again, it's quite broad, but it also appears to be impacting the entire customer experience, not just the effects of potentially a practice. When we start looking at advertising and pricing and other things, those are kind of gateway experiences for a customer. I do think it is a much more holistic approach to this. Where is it going to come up? Certainly, it will come up in the examination process, and the CFPB has already said that.
We will talk a little bit about some of the areas they are going to look at. I also think this is going to come up separately in a couple of other ways that are worth keeping an eye on now and which may ultimately roll back to the CFPB. The first, of course, is complaints. Whether those are complaints that are made to the CFPB or another regulator, certainly those are the canary in the coal mine. If there is a lot of complaint about a particular practice or a particular actor in the industry that could circle around to discrimination, I think that is going to get attention. Certainly, if your financial services institution is receiving those types of complaints about something that is an issue that you would want to put a lot of attention on.
I think we may see it independently through the class action litigation bar. Certainly, we know that one of the CFPB's mandates is to work with the plaintiff's bar and to see what's going on in the class action space as well. To the extent some interesting lawsuits are brought that focus on discrimination, I do think that may draw attention as well. In terms of where this could come up, the CFPB has given us some guidance both in terms of examination, and I think that rolls over also into the litigation and the enforcement side. The examiners are going to review products or services and look at what risk of harm would come up particular to those activities. Kind of low-hanging fruit here, of course, mortgage originations, certain types of consumer lending.
We'll talk about appraisals a little bit more later, and particularly products that are already sort of in the crosshairs of regulators or disfavored products, things like small-dollar loans or high-interest-rate products. I think, too, what we're seeing here is kind of a reactivation of the disparate impact approach. This is not something that the word is not being used, but the effect of it is basically the same. We need to be thinking about what kinds of facially neutral conduct we're engaged in, and then how does that play out through the entire experience, both the results and also the initial contacts that we have with the customer and how they're handled throughout the entire experience with our institution.
They've also told us to think about products that have a combination of features or that can make it more difficult for the consumer to understand the cost or the risk and therefore to understand how the harm could result. Certainly, things that are opaque or complicated, and many financial services products are by their nature, no matter how hard we try with our disclosures to explain things and make clear where the issues may be. I'll talk about that more in a minute. Things that come to mind, certainly teaser rates, products or services that have changing obligations that perhaps are not as well disclosed or clear as they should be, combined offerings or offerings that are targeted at a particular group.
We do not have to guess about that because the CFPB has already marked out certain groups that it has particular interest in, like older Americans, military lending, students, and so forth. These are some spots to watch as well. Just a couple more thoughts here on what this may look like, in particular in terms of whether something is misleading or likely to mislead a consumer. The CFPB has told us to look at the consumer's interpretation and whether that is reasonable under the circumstances. I do a lot of class action litigation, and I can tell you for sure that there are a number of cases out there where the consumer's alleged interpretation is not at all reasonable, and we manage those very effectively at the motion-to-dismiss stage through Article III standing issues and plausibility and so forth.
Because those rigorous standards are not going to be in play here, it's going to be a lot broader, I think, in terms of what could be considered reasonable for a particular consumer. We have been told to evaluate the entire context of the transaction or course of dealing. Again, it's no longer just what is the result and is that potentially discriminatory, but it's everything from the advertisement through the entire consumer process to determine whether the overall net impression is misleading or deceptive. That's pretty important to know that it's going to be the entire process. There's also been some pretty clear marking out of territory about written disclosures and when those may not be sufficient.
I mentioned before, and I know all of you spend a lot of time on your terms and conditions and the disclosures that are provided around consumer products and services. This is very much a hot area of litigation in the consumer space as well. The CFPB is indicating to us that a misleading statement may not be sufficiently cured, even if you have stuff in the fine print or other disclosures later. An obvious example, you know, outside of the financial services space would be if you buy a product and it has a representation on the front, for instance, that the product is healthy, and then you flip the box around and it has a whole bunch of qualifications about how that product is healthy only when it's consumed as part of a healthy diet or so forth.
The CFPB is now saying that that may not be enough, particularly in a circumstance where the consumer is either directed away from reading the fine print or told that it's not necessary to read it, and that having better and more robust disclosures in other places may not be enough to cure the initial alleged misrepresentation. Here's the other real big bell ringer to think about here. Technical compliance with the law may not be sufficient. What we're already doing in terms of being really careful about what the different obligations are under fair lending and consumer statutes may not be enough. In other words, even if it's fine legally, it could still have a discriminatory effect throughout the process, or it could fall into that gap that the CFPB is seeking to exert authority over. That's where it gets a little more fuzzy.
Just one more point to make for you quickly. They have also pointed out to us a couple of places that they think are going to be relevant to helping them understand both on the examination side and, I think, later on the enforcement side what kinds of issues there may be. One of them is complaints. Both complaints that you receive at your own institution, which will be reviewed, and also complaints lodged against subsidiaries, affiliates, or third parties that you do business with. Are you monitoring what is happening with those vendors and third parties and affiliates, and are those complaints being addressed? Certainly, if you're getting something from a state attorney general or the FTC, those are probably being handled promptly. What about third-party resources like the Better Business Bureau? They have also flagged the Ripoff Report and other sources like that.
Something to think about in terms of potentially broadening your scope of what you're looking at. Lastly, they have flagged chargebacks or refunds as an indicator of a problem. I'm not sure that that has so much direct impact in the financial services context, but definitely they have said that that would be a red flag to them.
Okay. Thank you. We covered a lot. I may come back to you on a couple of other questions if we have time, but I want to move to our second topic, which is the recent Regulation B advisory opinion on adverse action notices. On May 9th, the CFPB issued an advisory opinion affirming that ECOA prohibits lenders from discriminating against customers after they've received a loan, not just during the application process, and that lenders are required to provide an adverse action notice to a borrower when there's been an unfavorable decision made against the borrower. Maybe I'm kind of missing some here, but I thought this was really settled law, and the CFPB opinion is consistent with what my understanding has always been. Something must have prompted the CFPB to issue this opinion and raise the industry's consciousness about it.
Austin, I'm going to turn to you for this one. Why do you think the CFPB issued this advisory opinion regarding ECOA coverage after borrowers have applied for and received credit?
Thanks, Tim. I want to echo Isabelle's thanks to you and to Jennifer and everyone at Wolters Kluwer for inviting us. I mean, first, let me say, Tim, that I think, you know, I agree with you that while the bulletin describes what I think is generally considered settled law, I do think it was timely and appropriate. Here's why. In my own practice, I've seen instances where institutions were not aware that Regulation B applied to actions taken after credit was issued. We spend a lot of time in the fair lending industry talking about pricing and underwriting decisions. You know, right now, as we'll talk about a little bit later, a lot of the fair lending scrutiny is on exceptions in the origination context.
As the bulletin says in a few more words, Regulation B applies to any aspect of a credit transaction from beginning to end, including when credit is revoked and when there's an unfavorable change in terms. Those are really just examples. Those examples, the ones that I just mentioned, revoking credit and change in terms, really apply to open-end credit such as credit cards and HELOCs. There are other examples, I think, in the closed-end context, such as where you're providing or considering offering a repayment plan to the customer. You know, query, are you evaluating race or gender-neutral factors or factors that cause a disparate impact when making that decision? Or when you're repossessing a vehicle, are you making the determination about whether to repossess or when to repossess or how to repossess without respect to a prohibited basis?
Same for foreclosure, for that matter. Regulation B prohibits discrimination in those circumstances. There may be instances where taking the initiative to monitor fair lending risk in those areas would be warranted. I think the bulletin was targeted to reminding creditors and servicers of the application of Regulation B to those post-origination activities.
Okay. What do you think some of the practical consequences of the advisory opinion are?
Yeah. I think the primary practical consequence is for supervised entities at which examiners will be looking at the issue more closely. I think those institutions should anticipate questions like, what are you doing to monitor fair lending after the loan is made? The inquiry might be much more specific than that, and actually probably will be, because it's always been true that the regulators have looked at servicing practices from a fair lending perspective. I don't want to suggest that the bulletin is a watershed moment in post-origination fair lending. As I said, you know, I think there have been certain instances where institutions haven't been as conscious about the application of Reg B to those activities than to the origination. Maybe regulators will choose to get more granular in their analysis of post-origination activity. I used the repossession example earlier.
Maybe instead of looking at fair lending and how you decide to repossess or foreclose, for example, they may start looking at when you repossess. How do you decide the timing for repossessing a car, let's say? And are there factors in that decision that might not be, you know, race or gender-neutral? Or are there factors that go into that decision that have a disparate impact, such as potentially the type of car that the borrower is driving and having that affect when you repossess the cars? For open-end credit issuers, which can revoke the credit or change the terms, I think those institutions should expect a closer review, particularly of the adverse action notices and the like.
Let me just ask you. I mean, you just covered a number of different key things that financial institutions should probably take away from that opinion and how they're thinking about this as it relates to their lending programs. What are some key things that financial institutions should do to try and mitigate risk here?
Yeah. I think the first thing we should take from the advisory opinion is really what it says on its face, which is that you can't approach questions from the CFPB regarding post-origination activities or even the adverse action notice requirements by saying, "We didn't know that." I mean, you know, that's really the point of the bulletin, quite honestly.
Not a good.
Yeah. I mean, it's not a good. I mean, and you know, I think in the past, I think it has occurred, or even we disagree. I mean, one of the good things that the bulletin does is lay out, not just say, as we'll talk about, I think, in the adverse action and complex algorithms notice, not just say, you know, the law applies to you. You know, it explains why Reg B and the adverse action requirements apply to the circumstances that we've just discussed. It lays out the legislative history. It lays out the text of the regulation. It lays out, you know, the supervisory guidance that's been provided in the past relating to it.
I think the first thing is make sure you know what the bulletin is saying and, you know, incorporate into your compliance management system, you know, an accounting of the fact that Regulation B applies to the post-origination activity. That is the first thing. It is the most basic. You do not need a lawyer to tell you that, certainly. Often basic is the most important, quite honestly. I think the second thing is that from the CFPB's perspective, institutions are not thinking enough about adverse action notices when they revoke credit or change the terms of an existing credit arrangement. That, I think, you know, in this particular economic environment where there is probably going to be more in the way of credit card accounts being revoked or adverse changes in the terms of the credit arrangement, I think that is going to become more and more important.
I think the timing of that guidance was important. I think the third thing is that there could be greater scrutiny, of course, of practices after credit has been granted under fair lending, but I think other laws as well, right? The minute a regulator starts sniffing into a particular area under one law, it could easily turn into a review under another law. I mean, certainly we've seen a lot of UDAP scrutiny in the post-origination context without as much public fair lending scrutiny. I think the heightened scrutiny under fair lending laws could lead to even more under UDAP, particularly given the intersection of fair lending and UDAP that Isabelle just discussed. It is worth sort of, I think, thinking about where the regulators might focus their attention in the fair lending context and then thinking about, hey, you know, fair lending is important.
Let's look at that, but have we surrounded, for example, the UDAP issues relating to that issue as well?
Okay. I would like to turn to our third topic, if I could, which kind of includes another element of adverse action notices here, but this time in connection with complex algorithms. Our third topic is artificial intelligence and the use of alternative data. We have all been following this for many years. There are a lot of benefits here. There are also some risks that need to be understood and managed. We saw in December of 2019, the agencies came together, issued an interagency statement on alternative data focused on underwriting and potential risks and benefits. We have not seen a whole lot recently, but we know there is more to come. We saw a request for information in March of 2021 asking for information on the use of artificial intelligence by financial institutions. That comment period was extended out till July of 2021.
There is a lot going on here and a lot of thought being given to this topic. Austin, I'll stay with you here. The CFPB and the agencies have suggested for quite some time that the industry should expect more from them on the use of artificial intelligence and alternative data. It's been pretty quiet lately. Why do you think that may be?
Yeah. Let me first say that I think, personally, this is one of the more fascinating areas of fair lending right now, no question. AI, machine learning, artificial intelligence, and alternative data, I should have said. It's also, you know, I think one of the most complicated. To top it all off, I think it's probably the most frustrating because from a lot of people's perspective, we're not making as much progress in figuring out how to assess fair lending in these areas as many of us would like. We're looking, everyone's looking at each other to provide answers. You know, on your question, Tim, about, you know, maybe why we haven't made as much, regulators haven't made as much progress in telling us what to do as some of us would have liked. I mean, let's start with AI and machine learning.
I think it's really hard to develop an approach that works for every institution in every instance. I mean, first, let's face it, there's the fact that not every institution conducts fair lending testing or monitoring of even the most basic underwriting and pricing models, right? There may be circumstances where that type of testing may not be necessary. Against that backdrop, I think for complicated models like the one where we're looking at artificial intelligence or machine learning, it's difficult for regulatory agencies to say, you know, you have to conduct fair lending analyses of those models. If they say that, the question becomes how. That's not an easy question to answer, as everyone on this webinar probably knows.
The bottom line is that there's no single approach to using or testing AI for potential discrimination, which, you know, from my perspective, means we might want to be careful what we wish for when it comes to regulatory guidance. By that, I mean that if you're in the category of looking for guidance from regulatory agencies about how to approach AI and machine learning, a quick response may be too prescriptive, and what is prescribed may not be what you're currently doing or even what you think is appropriate, which could make things even more difficult. To answer your question, patience and deliberation, I think, are key here, particularly in an area that obviously is new and, you know, coming into greater use and popularity. I think that's the approach that the regulators are taking. I applaud them for that, quite honestly.
Alternative data, you know, I will just quickly will say is a little bit easier. I think the approach for reviewing and testing alternative data as opposed to AI and machine learning is more straightforward. That really doesn't require a pronouncement from the regulatory agencies to accomplish. I will discuss, you know, what can be done there in a little bit.
I'd like to also just ask you on this topic about disparate impact comes to mind right away when you start looking at these particular issues. I wondered whether or not you had any sort of suggestions on testing for disparate impact risk with these kinds of models.
Yeah. I mean, I'd be happy to discuss that. You know, first, I do want to say at a basic level on AI and machine learning, I do believe that doing something to monitor for fair lending puts you ahead of the curve, so to speak, when it comes to mitigating fair lending risk in this area. I think with all the uncertainty about, you know, how you should test for fair lending and given how complicated and expensive it can be to test those models, doing something really is a big mitigation step, I think. I'll talk about, in specific answer to your question, Tim, what they can do. On alternative data, I think more is expected. Let me start with alternative data. Alternative data, I mean data that's not found in a credit report or that's typically associated with an applicant's credit profile.
There, I think the expectation is pretty well established that institutions should look at least initially at the individual variables to assess whether they are facially discriminatory or a proxy or a prohibited basis. If yes, they should be eliminated immediately, of course. If in doubt, you know, I mean, a guiding principle that I follow in all things is leave it out. I mean, that's easier said than done, but you know, when in doubt, leave it out. That helps eliminate the disparate treatment risk. The question becomes disparate impact. On that point, I think the question is whether the variable, while neutral, is correlated with a prohibited basis group and causes a disparate impact. That can be tested in a number of different ways, including dropping the variable from the model and seeing whether the impact changes.
When doing that type of an, and that can be hard if you're talking about a whole ton of variables, right? I mean, dropping each individual variable and then rerunning the model and seeing what the result is, although, of course, with computers these days, maybe that can be done in nanoseconds, but it's not as simple as it sounds. Let's put it that way. We sometimes recommend starting with the variables that are least explainable when it comes to determining creditworthiness. For example, you know, I'll just use an egregious example. The type of cell phone you use, whether it's Apple or Samsung. I'm not familiar with that variable being used in the States, but I've heard it might be used outside of the States in credit decisions.
There, I might start with that variable in the disparate impact analysis because candidly, I can't imagine trying to explain how the type of cell phone you use is relevant to your creditworthiness. On the issue of, you know, testing for disparate impact with respect to AI and machine learning, honestly, Tim, I was sure that you were going to tell us the answer to that question.
It's on me.
Let me first say that I, you know, I know there are a lot of people on this webinar whose job is to think about that very question, you know, in-house, who, you know, have a real expertise with respect to machine learning, with respect to AI. You know, my hat's off to all of you. I will acknowledge upfront, I'm just an attorney. I'm not, you know, a data scientist. I'm not someone who, you know, crunches data all the time, and we typically hire experts to do that type of work. Let me take a stab. One method that has been used more and more, in my experience, in underwriting is the adverse impact ratio, which looks at the relative approval rates of prohibited basis group and non-prohibited basis group applicants. That can be used when testing artificial intelligence and machine learning models.
That AIR, as we call it, tests whether there is equality in outcomes between groups using that model. If each group, the prohibited basis group and non-prohibited basis group, is equally likely to be approved, the AIR is 1.0. If the prohibited basis group has a lower approval rate, it's below one. A similar option is what's called the marginal effects test. One problem with both is that they don't account for the accuracy of the model in predicting risk. It's assessing the fairness of the model, but it's not assessing the accuracy or taking into account the accuracy, which, of course, is a major safety and soundness issue for banks and, of course, you know, for major business consideration for others as well. When making loans, of course, avoiding default is a major consideration, but so is fair lending.
Sometimes there can be tension between striving for equal outcomes and avoiding default. As an extreme example, you know, what if, the way I think about it is what if an institution achieves an adverse impact ratio of 1.0, which means both groups are going to be declined at, you know, precisely the same rate, but the default rate puts the company out of business. You know, that's not what we want to accomplish. You know, I think what's happening more and more is that the regulators are looking for the institution to see if there's a less discriminatory alternative to a model that has a low AIR but high, you know, accuracy to see if it would, you know, have less of a disparate impact. Let's put it that way.
I mean, on that point, keep in mind that, you know, the burden's on the plaintiff, here the government, if you show that there's a business necessity for your model, in other words, that it's really good at being accurate, the plaintiff must show that there's a less discriminatory alternative. The regulators are sort of pushing back to the institution to do that these days. What I want to do right now, actually, just very briefly, if I may, Tim, is put up a screen on the screen. Let me see if I can figure out how to do it. I know I can do it, but let me just make sure I can do it quickly. I'm putting up on the screen a chart that Ric Pace at Pace Analytics had prepared.
I think it does a good job of illustrating because I need illustrations to know what I'm doing, of illustrating this problem. I mean, if you look here on the y-axis, the model accuracy, that's assessing whether it's good at predicting default risk. On the x-axis, the fairness, let's say it's the AIR, this blue one being the original trained model, you can see here it has high model accuracy. It's very good at predicting risk or default. The fairness is fairly low. The AIR, let's say, is low, let's say below, you know, 0.8. The question becomes whether there is an alternative that achieves greater fairness while not giving up too much accuracy.
That is what this green box is illustrating, is different, you know, variations on the way the model is trained through machine learning to see if we can get to a sweet spot where there is both, you know, greater fairness and you are not giving up much accuracy. This green dot in this case is one where you can sort of see there is pretty significant, you know, lift in fairness without giving up much in model accuracy. That is the kind of thing, let me stop sharing, that is the kind of thing that, you know, I think we are trying to address with respect to disparate impact.
That's very helpful. I would like to just jump to another part of this topic, which is a recent CFPB circular that ties right in. This was the May 26th circular from the CFPB that confirmed that an applicant must receive specific reasons for why an application is denied or when other adverse actions take place, even if the creditors are relying on credit models using complex algorithms. I mean, to me, what this basically says is the reasons have to be provide reasons, and they need to be specific. You can't just say, "Well, our model turned you down." Not very helpful and not consistent with what Regulation B would require.
In some ways, this shouldn't have been a big surprise, but what do you sort of make of it, and what do you see as kind of the key takeaways and anything that you think that institutions might need to do to adjust or comply here?
Yeah. Let me say that, you know, I've been very complimentary of the regulatory agencies thus far, but this one, I was very disappointed with this circular, quite honestly. I don't think it advances the ball in terms of compliance very much. I just, I don't, I candidly do not understand why it might have been issued. That being said, I think there are a couple of key takeaways. The first is, you know, that you need to comply with the adverse action requirements of Reg B, even if you're using a complex algorithm, period, full stop, right?
No question.
I think that most people would agree that was settled before the circular was issued. Now, you know, in as much as there are institutions out there who are suggesting that the adverse action requirements are less stringent where you're using AI or machine learning or complex algorithms, that's been put to bed, right? I mean, no, apparently not. The second key takeaway, I think from my perspective, is that the CFPB is suddenly skeptical whether an institution that uses complex algorithms can comply with the adverse action requirements. You know, in other words, the circular seems to suggest that complex algorithms such as machine learning shouldn't be used because you can't comply with those requirements.
Some have said that, and I disagree with that conclusion because, you know, I think to reach that conclusion would mean that the CFPB had suddenly changed entirely its view of machine learning and AI because, you know, as you know, Tim, you know, July 2020, the CFPB issued a blog post that said AI has the potential to expand access by enabling lenders to evaluate creditworthiness of some of the, you know, millions of consumers who are unscorable. In the 2021 request for information, just last year, it said that the interagency request, it said the agency support reasonable innovation by financial institutions that includes the identification and management of risk associated with the new technologies.
You know, I think the bulletin reflects a pretty profound tonal shift, in my opinion, but I'm hopeful that it does not reflect a, you know, an actual shift in the agency's view that AI and machine learning can be very helpful and can be used by institutions when making decisions. In terms of how the institutions can comply, I think, you know, I think you need to be able to explain what the factor was that your model considered. That factor should be interpretable by the person that receives the notice. I think a lot of institutions are using the Shapley method right now, which essentially, you know, lists every feature in the model and a contribution to the outcome. The ones, the features that had the highest contribution are the factors that are disclosed.
The CFPB drops in a footnote in the bulletin that methods like, they don't specify Shapley, but they say post hoc, I mean, methods like Shapley won't work in all cases. I think that, you know, institutions need to continue to be diligent in ensuring that they meet the ECOA and FCRA requirements. To that end, I think it means that they need to be laser-focused on striving to make sure that the reasons are specific, interpretable, and actionable.
I think that, you know, CFPB representatives have made it very clear in public forums over the last year or so that they were working on clarifications here and that it was an important issue that they were going to opine about. Indeed, they did keep their word. I think it becomes incumbent upon those subject to the adverse action requirements under Reg B to adjust and, as you point out, be sure that folks are being provided with the right information. We have about 10 minutes left and maybe about five to seven minutes of that all talk. I wanted to just turn to the next sort of grouping here, which was basically other hot topics, so to speak. We can then just throw a few things out here to get us started with.
One of the things that comes to mind right away, at least for me, is appraisals. Fairness in the appraisal process has been, I think, a really strong part of the Biden administration's emphasis here in the last 18 months or so. We saw the creation of the PAVE Task Force on Property Appraisal and Valuation Equity. These were pretty high-level folks in the government involved with this from across the government, the regulatory community, various cabinet-level departments. You issued a report recently that outlined a number of actions that the government will be taking over time. Maybe on this one, Isabelle, I'd ask you to, I know you can't cover everything that's in the PAVE report in the time we have today, but what did you see as the key takeaways?
Sure. I'll be very brief so we can touch on a few other trends as well. This is an interesting example of that intersection between the potential unfairness and discrimination analysis and AI because the way that AI features into this discussion is both as the problem and as the solution, right? You can have models that have inappropriate features and ultimately end up creating a discriminatory result or one of the takeaways from the PAVE report and something that the bureau continues to work on. The PAVE report is in early days, so there will continue to be developments here.
One of the takeaways is that perhaps AI is also the solution because if you can come up with a tool that takes some of the subjective criteria out of appraisals, then you may be able to eliminate disparities that exist for communities and borrowers of color. In particular, there are certain objective criteria. How old is the house? What shape is it in? Does it still have grandma's 1960s kitchen, or has it been fully remodeled? Those are fairly easy to quantify, but there are also subjective criteria, including what comps are used to evaluate the value of the house. Sometimes even the views of the individual appraiser himself or herself about the neighborhood where the property is located and various subjective issues can produce these disparities. An AI tool might be able to even sum of that out.
It is, of course, a problem both on the overvaluation side because you end up with a situation like the 2008 financial crisis and on the undervaluation side. Again, this kind of harkens back to the CFPB's view that we should be looking at ways to prevent lasting harm if ultimately over time, the ability to build equity in a home or to use that equity for other things is diminished because of unfair appraisal processes or you cannot sell your house for what it should be worth. These are all issues that potentially affect particularly communities and borrowers of color. This is definitely a space to watch. Let's keep on with that one there. I know Austin has a couple of things he wants to flag for us too as we close up here.
The other kind of hot topic on my mind is the ESG area. Again, very early days with regulators. It is still with the SEC, and there is quite a bit going on there. I do think a lot of what we have talked about today could potentially fall in the ESG bucket too. That is sustainability, environment, social, and governance. Certainly questions about discrimination and how companies and businesses are managing those risks and disclosing what they are doing could be an ESG issue. Maybe that is just a rebranding of things that are already out there, but definitely a spot to watch, particularly in financial services.
Yeah, good point. Austin, got a couple for us?
Yeah. Let me raise that, you know, a lot of us have spent a lot of time trying to figure out what BISG is and applying it to the non-mortgage fair lending context. It turns out that the CFPB in April 2022 noted that it's considering something different called the BIFSG, which also brings in first name in addition to surname and geographic components. For those of you who feel like you've mastered BISG, and I know there are a few of you out there, I'm not one of them, but a few of you out there now, we've got BIFSG to think about. That's one thing I wanted to raise. The other thing is to raise just, we haven't talked about exceptions in pricing and underwriting and servicing.
I mean, I think all of us who do fair lending work know that one clear focal point for the regulators right now is exceptions, whether they be exceptions for, you know, competition to acknowledge competitive offers, whether they be exceptions for relationships with an institution that gets you more favorable pricing or more, you know, flexible underwriting, or just exceptions where, you know, a loan officer or someone decides that they're going to waive a fee or something like that. I mean, that, in my experience, has been a very significant area of fair lending scrutiny. It's not one that really has been the subject of these, you know, the specific subject of these bulletins, but one that I think that all of you on the line should be thinking about because it's one that, at least in my experience, is a big focal point.
I'm just going to throw one more out there before I turn this back to Jennifer. I'll give you guys 30 seconds, which is always a real treat. We've got kind of a really challenging economic environment right now, rising interest rate environment, inflationary spiraling, a lot of issues affecting consumers, both here domestically, global issues that we're working on, and a lot of regulatory activity. What's kind of the one big thing that we may have to think about given the current economy as we go forward and try to work to protect consumers and keep our institutions compliant?
I'm happy to start with that one. I think in an environment like this, particularly where average families and those that live paycheck to paycheck are feeling really pressed and it's costing more when they go to the store, this is the type of environment where you start to see the consumer complaints. People begin to become delinquent on their consumer loans. There's a lot of money that has now come out of the system from COVID relief. If they had savings, there was an article the other day that says a lot of that has now been dissipated due to inflation. I think we're at the beginning of what will result in more regulatory scrutiny and in general, a lot more busyness on the default and workout and bankruptcy side as some of this pushes its way through the system unless the economy improves.
Yeah, I'll echo that. I mean, Isabelle, you stole my idea, Isabelle. I'm kidding. I think that, you know, higher interest rate products that are targeted to customers with lower creditworthiness who may be looking for, you know, money to fill the difference between their living expenses and their paycheck. They've been scrutinized in the past, but sort of the cycle of debt type products has been something that the CFPB has focused on for a long time and I think, you know, may be brought back into light due to the economy.
Good points. We have about two minutes left. I'm going to turn this back to Jennifer to just wrap up with us. I want to thank both of you. Had to keep you moving there for a little while to save time. Thanks for all the information. Jennifer, back to you.
All right. Thanks, Tim. And thank you so much, Isabelle and Austin. It's always a pleasure to have your firm represented on our webinars and at the colloquium. Speaking of which, we've been very busy preparing for the 26th Annual CRA and Fair Lending Colloquium. It'll be held live in person this November 13th through 16th at Caesars . I just want to warn everyone on the phone, last time we were at Caesars , apparently the venue or the location is so popular in addition to the conference content that we sold out and we had to turn people away. If you do know that you're planning to attend the colloquium, I highly recommend that you register as soon as feasible and get your hotel room reserved. We do have a limited number of hotel rooms blocked at the conference venue hotel.
I look forward to seeing you again in person. It's been two long years, and it's going to be a great pleasure to see all of you in person and to be back together. Thank you for joining today's webinar. Please watch for an email in the coming days with a recording to today's session. That concludes our webinar for today. Thank you, and we'll see you in November.