10.1 Introduction
For generations, the civil rights community understandably focused its fair housing efforts largely on minority access to affordable housing. Fair Housing Act litigation resources were generally concentrated on whether minority renters and homebuyers were being denied housing opportunities in the first instance. More recently, the focus of some litigation has changed, at least in the context of homeownership, to disparities in the price at which housing is available. In the context of mortgage credit, a series of important private class actions and government investigations under both the Fair Housing Act and the Equal Credit Opportunity Act have focused on discrimination in practices that appear to have driven black and Latino families into higher-cost loans on more onerous terms than similarly situated borrowers. The cases and investigations described in this chapter are grounded largely in the disparate impact of pricing practices that appear to have resulted in hundreds or sometimes thousands of dollars in additional annual credit costs for minority homeowners.
Disparate impact has often been viewed as the poor stepchild of civil rights litigation. Even though the Supreme Court first ruled in 1971 that a discrimination claim based on disparate impact was cognizable,1 and Congress reaffirmed its status in 1991,2 the Justice Department’s Civil Rights Division didn’t bring its first disparate impact case until almost 2010.3 At the same time, however, there has been a growing recognition by academics as well as by courts that disparate treatment claims were becoming less well suited to combat a variety of civil rights problems (Kreiger and Fiske Reference Krieger and Fiske2006). Discretionary decision making tainted by unconscious racism can fly below the radar screen of traditional civil rights scrutiny (Lawrence Reference Lawrence1987). Disparate impact has offered an alternative approach to combating the detrimental effects of implicit racial bias (Hart Reference Hart2005; Primus Reference Primus2010).
This chapter argues that disparate impact proof of discrimination is especially well suited for application to loan transactions, because it can be thoroughly investigated based on the lender’s own data records. The commodification of lending has created a system of mass retail selling. While many borrowers see themselves as unique and their financial history as opaque, lenders almost always use algorithmic underwriting standards applied to a core set of underwriting variables. Assessing whether a lender’s policies allowing the final price to then be set based on other non-objective factors produced an unjustified disparate impact is straightforward because lenders’ own underwriting datasets are, by design, intended to capture information about the variables that the lending industry itself believes are germane to originating and setting the terms of loans. The key statistical evaluation is to ascertain, after controlling for the variables that the lenders themselves have gathered and evaluated, whether minority borrowers were more likely than nonminority borrowers to be charged higher credit costs.
To be sure, there is discretion in choosing the factors evaluated in algorithmic underwriting, but the most important form of discretion is ceded to the sales force who set the ultimate terms of the mortgage and who receive commissions to maximize profit. An important and widespread policy of lenders was to give brokers the discretion to price gouge consumers – if they could induce the borrower to agree to a supra-competitive interest rate or supra-competitive fees. Lenders who were not aware of the race of borrowers at the time of lending could nonetheless be liable for setting up systems that allowed salespeople (who do know the race of their customers) to exercise discretion in way that disproportionately exposed minorities to predatory terms and high-cost loans.
This chapter tracks the rise of disparate impact lending litigation and how subsequent decisions of the Supreme Court and circuit courts have limited the viability of such claims. Part 2 details the history of mortgage lending lawsuits and the kinds of information plaintiffs were able to bring to bear in such cases. Part 3 then discusses the growing judicial resistance to these kinds of claims, particularly in Wal-Mart Stores, Inc. v. Dukes, 564 U.S. 338 (2011). Part 4, in speculating on possible futures for disparate impact liability, describes the Consumer Financial Protection Bureau’s (CFPB) recent auto-lending initiatives.
10.2 History of “Reverse Redlining” Mortgage Lending Disparate Impact Litigation
Over the past two decades, a large number of academic studies have explored the relationship between borrower race and the availability or the cost of obtaining residential mortgage loans in the United States. Two literature reviews can be found in White (Reference White2009) and Courchane (Reference Courchane2007). As explained in greater detail in these reviews, early academic studies focused on the relationship between mortgage denials and the racial composition of neighborhoods (Munnell et al. Reference Munnell1996). Early studies also included audits of lenders. For example, a 1999 study by the Urban Institute found that minorities were offered mortgages at higher rates than whites in similar circumstances (Turner and Skidmore Reference Turner and Skidmore1999). The Urban Institute findings were based in part on paired audit testing conducted by the National Fair Housing Alliance that was carried out by people of different racial and ethnic backgrounds in a sample of seven cities. Each group of testers – including one white and one or more minorities – told lenders it had similar credit histories, incomes and financial histories, and the same type of mortgage needs. The testing found that minorities were less likely to receive information about loan products, and received less time from loan officers. Most important for our purposes, this audit study found that minorities “were quoted higher interest rates in most of the cities where tests were conducted” (Turner and Skidmore Reference Turner and Skidmore1999, 2).4
These earlier studies were suggestive of significant racial effects, but suffered from an absence of controls for credit risk and other underwriting considerations when examining substantial samples of actual loan originations as opposed to more limited audit tests. Over time, as government reporting requirements improved and litigation and various investigations offered more complete datasets, researchers were able to include a number of additional controls in their studies and developed more complete empirical models of the residential mortgage origination process. Some focused on the impact of race on credit spreads and found statistically significant racial disparities (Avery et al. Reference Avery2005; Bocian, Ernst, and Li Reference Bocian, Ernst and Li2006; Fishbein and Woodall Reference Fishbein and Woodall2005, Reference Fishbein and Woodall2006). Later studies expanded this analysis by controlling for loan channels, and found reduced, but still statistically significant racial effect on the APR of mortgage loans (Courchane Reference Courchane2007, LaCour-Little Reference LaCour-Little2009; White Reference White2009). Yet other studies found statistically and economically significant racial disparities in the amount of compensation mortgage brokers earned on residential mortgage originals and in FHA closing costs charged to borrowers (Jackson and Burlingame Reference Jackson and Burlingame2007; Woodward Reference Woodward2008).
The notion that minority borrowers may pay more for home loans than similarly situated white borrowers is not altogether surprising. A wide body of literature has shown that individuals can be influenced (even subconsciously) by race. The theory that the racial disparities in borrowing costs are the by-product (at least in part) of racially influenced credit-pricing decisions in no way implies that loan officers and brokers must harbor animus toward minorities or that they are engaging in intentional discrimination. For example, a number of studies have found that economic decision makers are influenced by racially conscious or unconscious stereotypes (Kirschenman and Neckerman Reference Kirschenman, Neckerman, Jencks and Peterson1991). For example, the Implicit Association Tests5 suggest that many people of professed goodwill find it impossible to avoid treating African American pictures differently from white pictures when asked to perform a simple sorting exercise. These tests are part of a growing literature documenting unconscious bias against African Americans and other minorities (Chen and Bargh Reference Chen and Bargh1997; Dovidio et al. Reference Dovidio1986; Niemann et al. 1988; Vanman et al. Reference Vanman1997).
To the extent that economic decision makers often harbor unconscious, but biased racial stereotypes, it becomes more plausible that the subjective pricing process that mortgage lenders established for setting loan terms (in which a loan officer or broker can often plausibly deny that its treatment of an individual consumer was based on some attribute other than race) might mask what are in fact racially influenced decisions. In Watson v. Fort Worth Bank & Trust, supra, the Supreme Court’s recognition of the existence of subconscious stereotypes was cited as one of the reasons for approving the use of a disparate impact analysis to evaluate the subjective decision-making processes at issue in that case (ibid. at 990). (“Furthermore, even if one assumed that any such discrimination can be adequately policed through disparate treatment analysis, the problem of subconscious stereotypes and prejudices would remain.”) Similar reasoning impacted the Supreme Court’s decision in Texas Department of Housing & Community Affairs v. The Inclusive Communities Project, Inc., 135 S. Ct. 2507 (2015), where the court held that “[r]ecognition of disparate-impact liability under the FHA plays an important role in uncovering discriminatory intent: it permits plaintiffs to counteract unconscious prejudices and disguised animus that escape easy classification as disparate treatment.”
10.2.A Measuring the Effects of Discretionary Decisions on Mortgage Prices
A number of class-action cases have been brought against various lenders regarding the alleged disparate impact resulting from discretionary pricing policies.6 Plaintiffs in these cases asserted that the defendant lenders engaged in discretionary pricing policies under which the lenders’ loan officers, brokers, and correspondent lenders could impose subjective, discretionary charges and interest rate markups in the loans that they originated. These subjective charges are added to the objective, risk-based rates already established by the defendants. Plaintiffs alleged that the defendants’ policies for access to their loan products subjected minority customers to a significantly higher likelihood of exposure to discretionary points, fees, and interest rate markups.
These allegations were brought pursuant to the Fair Housing Act (FHA) and the Equal Credit Opportunity Act (ECOA). Although it has been a question of substantial dispute, both civil rights laws clearly permit use of proof of disparate impact to establish discrimination. For the FHA, the Supreme Court recently confirmed this long-standing conclusion of every court of appeals that had considered the question in Texas Department of Housing & Community Affairs v. The Inclusive Communities Project, Inc., supra.7 Both the courts and the CFPB, the agency charged with interpreting the ECOA under the Dodd-Frank Act, have found that that statute also allows for a disparate impact cause of action.8
In this section, we focus on In re Wells Fargo Mortgage Lending Discrimination Litigation as an exemplar of the kinds of evidence that plaintiffs were able to adduce in these cases.9 Wells Fargo, like many lenders, made loans both as a retail lender through its branches and mortgages offices and as a wholesale lender through ostensibly independent mortgage brokers. In either channel, Wells Fargo set its core loan prices by using an algorithm applied across a wide range of the borrower’s credit characteristics, but allowed its employees and its brokers to earn a commission, within certain limits, by marking up and adding costs to the algorithmically derived price. These markups were at the discretion of Wells Fargo’s employees and brokers, were not tethered to credit risk, and yielded a commission, based on a formula for the employee or broker that set them. Wells Fargo published price sheets that showed its core prices (subject to underwriting), the scope of the permitted markup, and the commission structure by which the sales commission for the loan would be tied to the markup. By maximizing discretionary markups, the sales force increased the loan price and maximized commissions.
The case against Wells Fargo asserted that Wells Fargo’s sales force used markups most aggressively to increase loan costs for African American and Hispanic borrowers such that Wells Fargo’s markup policy resulted in a measurable disparate impact across Wells Fargo’s mortgage lending business.10 To the extent that the markups were imposed by nonemployee brokers, Plaintiffs relied on the long-standing agency principles applicable to the discrimination laws.11
The evidence at issue was designed to show the amount by which the loan costs for African American and Hispanic borrowers exceeded those of similarly situated white borrowers. The statistical evaluation presented the actual costs of borrowers with virtually identical credit characteristics as determined in Wells Fargo’s underwriting process. In particular, the following tables are taken from the report of Professor Howell Jackson, who served as the plaintiffs’ economic expert and provided the crucial statistical tests of disparate impact. Table 10.1 summarizes both the average difference in loan costs (as measured by the Annual Percentage Rate (APR)) for Wells Fargo borrowers of different races as well as the racial differences after controlling for a host of underwriting risk factors. Professor Jackson estimated that the present value of the defendant’s overcharges had cost minority borrowers, in aggregate, approximately half a billion dollars.
Table 10.1: Summary of Disparate Impact and Monetary Relief
| African Americans | Hispanics | Total | |
|---|---|---|---|
| Mean APR for Given Minority | 6.940% | 6.511% | |
| Mean APR for Whites | 6.266% | 6.266% | |
| Difference | 0.674% | 0.245% | |
| Difference after Controlling for | |||
| Relevant Risk Factors with Regressions | 0.101% | 0.064% | |
| Present Value of Relief over Five Years ($Millions) | $297.7 | $329.2 | $627.0 |
| Number of Loans | 294,983 | 452,471 | 747,454 |
| Avg. Present Value of Relief per Loan over Five Years ($) | $1,009 | $728 | $839 |
Of course, simple difference in the average APR charged to minorities and whites might be justified by difference in creditworthiness. Even though statistically significant average APR differences might be prima facie evidence of actionable disparate impacts and therefore shift the burden of justification to the defendant, plaintiffs routinely go further to establish that the disparities persist after controlling in regressions for standard underwriting variables. Because regression analysis remains opaque to many triers of fact, plaintiffs often show that average racial APR disparities persist within individual credit score ranges. Thus, Professor Jackson’s report showed (reproduced here as Table 10.2) that within most FICO score bins, the average APR charged to whites was lower – often by dozens of basis points – than the average APR charged to minority borrowers. The persistence of racial APR differences even among borrowers with similarly high credit scores particularly underscores that Professor Jackson’s finding is not driven by the possibility that minority borrowers tend to have poorer credit scores than white borrowers.
Table 10.2: Mean Annual Percentage Rate (APR) by Race and Credit Score, 2001–2007
| African American | Hispanic | White | Difference Mean between Af. Amer. APR & Mean White | Difference between Mean Hisp. APR & Mean White APR | ||||
|---|---|---|---|---|---|---|---|---|
| Loans | Mean APR | Loans | Mean APR | Loans | Mean APR | APR | Mean White APR | |
| Missing score | 24,994 | 6.370 | 33,811 | 6.336 | 190,503 | 5.986 | 0.384 | 0.350 |
| 300–539 | 10,506 | 8.847 | 5,163 | 8.609 | 25,806 | 8.875 | −0.028 | −0.266 |
| 540–559 | 8,615 | 8.395 | 5,171 | 8.149 | 26,662 | 8.279 | 0.116 | −0.131 |
| 560–579 | 13,573 | 8.286 | 8,752 | 7.906 | 45,688 | 7.954 | 0.332 | −0.048 |
| 580–599 | 18,144 | 7.984 | 13,375 | 7.648 | 70,260 | 7.618 | 0.367 | 0.031 |
| 600–619 | 22,675 | 7.609 | 20,145 | 7.251 | 107,043 | 7.181 | 0.428 | 0.070 |
| 620–639 | 29,809 | 7.333 | 32,065 | 7.014 | 165,535 | 6.882 | 0.452 | 0.133 |
| 640–659 | 30,519 | 7.086 | 37,265 | 6.807 | 218,907 | 6.630 | 0.456 | 0.177 |
| 660–679 | 31,058 | 6.776 | 46,209 | 6.567 | 294,162 | 6.395 | 0.381 | 0.172 |
| 680–699 | 29,454 | 6.562 | 52,537 | 6.416 | 365,036 | 6.246 | 0.315 | 0.170 |
| 700–719 | 26,177 | 6.424 | 52,855 | 6.335 | 412,046 | 6.169 | 0.255 | 0.166 |
| 720–739 | 22,676 | 6.355 | 49,844 | 6.268 | 450,023 | 6.126 | 0.229 | 0.143 |
| 740–759 | 21,136 | 6.263 | 50,019 | 6.194 | 525,970 | 6.071 | 0.192 | 0.123 |
| 760–779 | 18,679 | 6.171 | 46,681 | 6.111 | 617,954 | 6.019 | 0.152 | 0.092 |
| 780–799 | 14,106 | 6.124 | 33,932 | 6.053 | 563,555 | 6.014 | 0.110 | 0.039 |
| ≥ 800 | 4,990 | 6.125 | 10,610 | 10,610 | 211,130 | 6.055 | 0.070 | −0.010 |
| All Credit Scores | 327,111 | 6.940 | 498,434 | 6.511 | 4,290,280 | 6.266 | 0.674 | 0.245 |
The core evidence of unjustified disparate impacts comes, however, from regressions. Thus, for example, in the following table, Jackson reported four nested specifications testing for racial disparities:
The simplest regression (Model 1) reported in Table 10.3 only includes controls for the borrower race – and in this and the other models the reported coefficients represent the estimated APR differences measured in basis points between the indicated minority race and non-Hispanic white borrowers. Thus, Model 1 indicates that African American borrowers’ APRs averaged 67 basis points more than white borrowers. Model 1 in essence provides evidence for a disparate racial impact without considering whether it is business justified. Models 2 and 3 respectively add fixed effects controls for the month in which the interest rate lock occurred and for the FICO score bins reported in Table 10.2. These models show that African American and Hispanic borrowers continued to pay statistically higher APRs than non-Hispanic white borrowers – but that the differentials are roughly halved when one controls for borrowers’ FICO score. Finally, Model 4 adds to Model 3 controls for the comprehensive set of underwriting variables listed in the notes to Table 10.3, including loan amount, debt-to-income ratio, loan-to-value ratio, loan type, loan purpose, loan term, occupancy type, property type, borrower history of bankruptcies, foreclosures, collections, and late payments, documentation type, loan amortization type, loan product category (e.g., 30-year fixed, 5-year ARM), prepayment penalty length, and the borrower’s state and metropolitan area (MSA). Professor Jackson’s specification includes a multitude of controls that could provide plausible business justifications for charging borrowers different APRs. After controlling for all these underwriting influences, the regression tests find that African Americans and Hispanics still pay higher APRs than non-Hispanic whites who are similarly situated with regard to plausible business justifications – respectively 10.1 and 6.4 basis points higher. Moreover, the regression indicates that these disparities were highly statistically significant (p < 0.01). Model 4 thus represents the second stage of testing (and in this case showing) that the disparate racial impact persists after controlling for plausible business justifications.
Table 10.3: Effect of Race on APR (Basis Points) Using Regressions Estimated on All Loans
| Race | Model (1) | Model (2) | Model (3) | Model (4) |
|---|---|---|---|---|
| African American | 67.39*** | 62.53*** | 26.24*** | 10.10*** |
| (0.29) | (0.26) | (0.22) | (0.16) | |
| Hispanic | 24.53*** | 24.69*** | 13.41*** | 6.39*** |
| (0.19) | (0.16) | (0.14) | (0.11) | |
| Observations | 5,654,985 | 5,654,985 | 5,654,985 | 5,654,985 |
| R-Squared | 2.6% | 30.7% | 46.4% | 70.5% |
| Adjusted R-Squared | 2.6% | 30.7% | 46.4% | 70.5% |
Note: Standard errors in parentheses.
*** Statistically significant at 1%, ** Statistically significant at 5%, * Statistically significant at 10%.
Coefficients and standard errors for other explanatory variables are shown in Appendix 5 of Professor Jackson’s expert report.
Explanatory variables for each model consist of:
Model (1): Race dummy variables only.
Model (2): Race dummy variables and interest rate lock month dummy variables.
Model (3): Same as Model (2), but add FICO score bin dummy variables.
Model (4): Same as Model (3), but add loan amount bin dummy variables, total debt-to-income ratio bin dummy variables, housing debt-to-income ratio dummy variables, loan-to-value (LTV) bin dummy variables, combined loan-to-value (CLTV) bin dummy variables, loan type (conventional, FHA, VA, or RHS) dummy variables, self-employed borrower/co-borrower dummy variable, loan purpose dummy variables, loan term dummy variables (e.g., 15-year, 20-year, 30-year), dummy variables for occupancy type interacted with property type, property subclass dummy variables, dummy variables for credit report items (such as the presence of bankruptcies, foreclosures, collections, and late payments), documentation type dummy variables, loan amortization type dummy variables, loan product category dummy variables (e.g., 30-year fixed, 5-year ARM), escrow waiver dummy variables, length of rate lock dummy variables, rate float-down option dummy variables, lender-paid mortgage insurance dummy variable, combination loan dummy variable, prepayment penalty length dummy variables, state dummy variables, and metropolitan area (MSA) dummy variables.
Professor Jackson used these two racial APR differentials estimated in Model 4 to estimate the monetary relief due to the plaintiff class. Portions of his calculations for monetary relief are reprinted here as Table 10.4.
Table 10.4: Present Value of Monetary Relief to Wells Fargo Minority Borrowers Using the APRs Predicted by Model (4)
| African Americans | Hispanics | Total | |
|---|---|---|---|
| Present Value of Relief over Entire Loan Term ($Millions) | $923.0 | $996.7 | $1,919.7 |
| Present Value of Relief over 10 Years ($Millions) | $539.8 | $592.9 | $1,132.7 |
| Present Value of Relief over Five Years ($Millions) | $297.7 | $329.2 | $627.0 |
| Number of Loans* | 294,983 | 452,471 | 747,454 |
| Avg. Present Value of Relief per Loan over 5 Years ($) | $1,009 | $728 | $839 |
Note: *Monetary relief calculations are restricted to those loans in Wells Fargo’s loan database with APR data.
Professor Jackson calculated how much less the monthly payment for minority borrowers would have been if these borrowers had been charged the expected APR for similarly situated white borrowers. He then calculated the present value of this monthly differential (discounting at the Treasury rate) under different assumptions of about how long the minority borrowers were subjected to the higher monthly payments. Thus, Table 10.4 shows that if the average minority borrower pays for just five years of inflated fees (before paying off or refinancing their loans), the present value of the expected additional payments is more than $600 million.12
10.2.B Predatory Terms
While we have focused on litigation challenging disparate racial impact with regard to the cost of borrowing, a number of lawsuits have alleged that minority borrowers were disproportionately subjected to potentially predatory mortgage terms that artificially increased the risk of default. For example, loan characteristics described as potentially predatory in these lawsuits include higher interest rates reportable under the rate spread thresholds established by the Home Mortgage Disclosure Act (HMDA) regulations,13 subprime status, high LTVs, high debt-to-income ratios, interest-only payment periods, balloon payments, prepayment penalties, negative amortization, “stated” or no documentation requirement during loan underwriting, and teaser rates (in which the loan’s initial interest rate was substantially lower than the interest rate that could be imposed later during the life of the loan).14 Moreover, some banks used distinct marketing tactics and product development strategies in communities of color that some have argued lead to more expensive loans in those communities. An example is a case that resulted in a $3.5 million jury verdict: Jones et al. v. Wells Fargo Bank NA, et al., Case No. BC337821 (Ca. Super. Court, LA Cty., 2011). Certainly it would make sense to study whether loan terms are, on average, more favorable at suburban institutions where loan officers are more common, for example, than in urban branches of large national banks where mortgages are more often made through loan brokers. Similarly, examination of advertisements and other marketing materials available in different communities and possibly a renewed focus on paired testing may be useful.
Municipalities, including the cities of Atlanta, Baltimore, Cleveland, Memphis, Los Angeles, Miami, Miami Gardens, and Oakland, have pursued lawsuits against some or all of the four largest lenders (Bank of America, Wells Fargo, JPMorgan Chase, and Citibank), alleging that these lenders disproportionately originated loans with predatory terms to minority borrowers, which increased their likelihood of default, resulted in more foreclosures, and caused the municipalities to suffer damages through losses in property taxes (through decreased property values) and increased municipal services.15 The defendant lenders argued that the FHA does not cover municipalities seeking monetary recovery for these types of claims. The Supreme Court recently ruled that the municipalities have standing under the FHA and that the cases may go forward, albeit with some admonitions to the underlying courts to consider the question of whether the violations proximately caused the injuries complained of. Bank of America Corp., v. City of Miami, Slip Op., 581 U.S. ___ (May 1, 2017) (Stern 2017).
10.2.C DOJ Settlements
The Justice Department’s Civil Rights Division during the Obama administration in a series of enforcement actions aggressively pursued disparate impact theories against major mortgage lenders.
10.2.C.1 Countrywide (2011)
In December 2011, the U.S. Department of Justice settled an investigation against Countrywide alleging FHA and ECOA violations between 2004 and 2008. The U.S. Department of Justice alleged that “more than 200,000 Hispanic and African-American borrowers paid Countrywide higher loan fees and costs for their home mortgages than non-Hispanic White borrowers, not based on their creditworthiness or other objective criteria related to borrower risk, but because of their race or national origin” (Complaint, U.S. v. Countrywide, 2).16 The U.S. Department of Justice also alleged that, between 2004 and 2007, “more than 10,000 Hispanic and African-American wholesale borrowers received subprime loans, with adverse terms and conditions such as high interest rates, excessive fees, prepayment penalties, and unavoidable future payment hikes, rather than prime loans from Countrywide, not based on their creditworthiness or other objective criteria related to borrower risk, but because of their race or national origin” (3).
The Justice Department’s core evidence was quite similar to the kinds of evidence used in the previous class-action suits (exemplified by Professor Jackson’s analysis discussed earlier). The Department found that Hispanic and African American borrowers paid between 13 and 28 basis points more in interest than similarly situated non-Hispanic white borrowers in Countrywide’s retail Consumer Markets Division channel from 2004 to 2008, and these disparities were statistically significant (39–40). The Department also found that Hispanic and African American borrowers paid between 12 and 67 basis points more in broker fees than similarly situated non-Hispanic white borrowers in Countrywide’s wholesale channel from 2004 to 2008 (65–68). With respect to allegations of steering, the Department concluded:
Statistical analyses of loan data kept by Countrywide on wholesale 30-year term prime and subprime loans originated by Countrywide between January 2004 and August 2007 demonstrate that on a nationwide basis Hispanics who qualified for a Countrywide home mortgage loan and who obtained wholesale loans from Countrywide had odds between approximately 2.6 and 3.5 times higher than similarly-situated non-Hispanic White borrowers of receiving a subprime loan instead of a prime loan, after accounting for objective credit qualifications. Those odds ratios demonstrate a pattern of statistically significant differences between Hispanic and non-Hispanic White borrowers with respect to their placement by Countrywide in one of these two loan product categories even after controlling for objective credit qualifications such as credit score, loan amount, debt-to-income ratio, loan-to-value ratio, and others.
Moreover, the Department’s causal explanation for these disparities emulated the discretionary-pricing theories of the plaintiff class litigation.
The disparate placement of both Hispanic and African-American wholesale borrowers whom Countrywide determined had the credit characteristics to qualify for a home mortgage loan into subprime loan products, when compared to similarly-situated non-Hispanic White borrowers … resulted from the implementation and interaction of Countrywide’s policies and practices that: (a) permitted mortgage brokers and Countrywide’s own employees to place an applicant in a subprime loan product even if the applicant could qualify for a prime loan product; (b) did not require mortgage brokers or its employees to justify or document the reasons for placing an applicant in a subprime loan product even if the applicant could qualify for a prime loan product; (c) did not require mortgage brokers to notify subprime loan applicants that they could qualify for a prime loan product; (d) created a financial incentive for brokers to place loan applicants in subprime loan products; (e) allowed brokers and Countrywide loan officers and underwriters to request and to grant underwriting exceptions in a subjective, unguided manner; and (f) failed to monitor these discretionary practices to ensure that borrowers were being placed in loan products on a nondiscriminatory basis.
The Department settled the case for $335 million.17
10.2.C.2 Wells Fargo (2012)
In July 2012, using some of the same evidence described earlier, the Justice Department resolved allegations that Wells Fargo Bank engaged in a pattern or practice of discrimination against qualified African American and Hispanic borrowers in its mortgage lending from 2004 through 2009 (Complaint, U.S. v. Wells Fargo, 15–16).18 The Department’s investigation showed that the odds that an African American borrower of a Wells Fargo wholesale channel loan would receive a subprime loan rather than a prime loan were approximately 2.9 times as high as the odds for a similarly situated non-Hispanic white borrower from 2004 to 2008. Over the same time period, the same odds for an African American borrower of a Wells Fargo retail channel loan were 2.0 times the odds for a similarly situated non-Hispanic white borrower. The odds that a Hispanic borrower of a Wells Fargo wholesale channel loan would receive a subprime loan rather than a prime loan were approximately 1.8 times as high as the odds for a similarly situated non-Hispanic white borrower from 2004 to 2008. Over the same time period, the same odds for a Hispanic borrower of a Wells Fargo retail channel loan were 1.3 times the odds for a similarly situated non-Hispanic white borrower. All of these disparities were statistically significant (15–16). The Department also found that Wells Fargo charged minority borrowers in its wholesale channel up to 78 basis points more in broker fees than similar white borrowers (26).
The settlement provided $125 million in compensation to wholesale borrowers who were steered into subprime mortgages or who paid higher fees and rates than white borrowers because of their race or national origin (Consent Order, U.S. v. Wells Fargo, 13).19 In addition, Wells Fargo agreed to internally review its retail mortgage lending policies and to compensate African American and Hispanic retail borrowers who were placed into subprime loans when similarly qualified white retail borrowers received prime loans (21–22). Wells Fargo also agreed to provide $50 million in down payment assistance for new loans to borrowers in communities around the country that were especially hard hit by the housing crisis (18–19).
10.2.C.3 Sage Bank (2015)
In 2015, the Justice Department reached a smaller settlement on similar theories with Massachusetts-based Sage Bank. The United States alleged that Sage had set a target price for each mortgage loan and allowed loan officers to mark up loans above that target (Complaint, U.S. v. Sage Bank).20 It further alleged that the discretion was exercised in a manner that resulted in higher prices for African American and Hispanic borrowers. Sage agreed to practice changes and to create a fund of just over $1 million in compensation for affected borrowers (Consent Order, U.S. v. Sage Bank, 4–10).21
10.3 Rejection of Statistical Analysis as a Basis for Certification of a Disparate Impact Class
In Dukes v. Wal-Mart Stores, Inc., plaintiffs brought an ambitious broad-based challenge to Wal-Mart’s treatment of its female employees. Although the plaintiffs successfully sought class certification in the district court in a decision that was ultimately affirmed both by a panel of the Ninth Circuit and by the Ninth Circuit sitting en banc (603 F. 3d 571 (9th Cir. 2010)), the Supreme Court reversed in a far reaching decision on what it means to have a “common question” under the class-action rule and on the use of statistical analysis to establish commonality in a disparate impact case (Wal-Mart Stores, Inc. v. Dukes, 564 U.S. 338 (2011)).
From its inception, the Wal-Mart class action involved claims of both disparate treatment and disparate impact regarding the hiring and promotion of more than a million female employees. The plaintiffs alleged that the company delegated employment decisions to local managers who intentionally discriminated against women. The Supreme Court held that if employment discrimination is alleged to occur because local managers are exercising discretion in a discriminatory manner, no common issue exists for purposes of class certification. The Court explained that the company essentially had a policy against having uniform employment practices (355). Accordingly, managers “were left to their own devices” to determine criteria for making hiring and promotion decisions for millions of employees (355). The Court concluded (in a 5–4 decision) that granting employees discretion was the antithesis of having a policy:
The only corporate policy that the plaintiffs’ evidence convincingly establishes is Wal-Mart’s “policy” of allowing discretion by local supervisors over employment matters. On its face, of course, that is just the opposite of a uniform employment practice that would provide the commonality needed for a class action; it is a policy against having uniform employment practices.
The Court thus found that where there was no challenge to a uniform policy or practice, a court would need to look at millions of individual decisions by the local managers (352). The Court explained there needs to be “some glue holding the alleged reasons for all those decisions together” to meet the commonality requirement (352). Class certification was therefore not possible.22
In reaching this conclusion, the Court rejected the plaintiffs’ view that adequate statistical analysis could function as “glue” by establishing that Wal-Mart’s grant of discretion had a statistically significant overall discriminatory impact on female employees. Notably, this rejection appears to be inconsistent with the driving impetus behind a “disparate impact” claim itself and is therefore an implicit rejection of Watson and perhaps even Griggs.
The “impact” of any policy is represented by its aggregate effects. Where those effects tend to fall negatively on a protected class, a conclusion of discrimination is appropriate even if not every class member is affected. In Griggs, for example, some African American applicants apparently did have high school diplomas; nevertheless, the Supreme Court correctly recognized that the overall effect of the diploma requirement fell more heavily on African American applicants. Similarly, some applicants, with or without diplomas, would properly be denied employment irrespective of their educational background.23 A disparate impact claim arises from the negative impact of being subjected to the policy in the first instance, particularly if the impact is demonstrated by a measurable factor such as loan cost. A policy that results in an average increase in the amount charged to members of a protected class affects borrowers both above and below the mean loan payment. That is, a disparate impact claimant paying below the mean might have a payment even further below the mean absent the impact of the policy.
It would be well-nigh impossible for the individual evidence of the impact of any corporate policy in employment or lending, particularly one granting discretionary autonomy to those making subjective decisions, to point in a single direction across a large group of individuals. Wal-Mart’s class certification rubric, taken at face value, may thus render any group private remedy for disparate impact unachievable.24 Despite this, the Supreme Court explicitly declined to overrule Watson v. Fort Worth Bank and Trust, 487 U.S. 977 (1988) in which the court concluded:
We are also persuaded that disparate impact analysis is in principle no less applicable to subjective employment criteria than to objective or standardized tests. In either case, a facially neutral practice, adopted without discriminatory intent, may have effects that are indistinguishable from intentionally discriminatory practices. … If an employer’s undisciplined system of subjective decision-making has precisely the same effects as a system pervaded by impermissible intentional discrimination, it is difficult to see why Title VII’s proscription against discriminatory actions should not apply. … We conclude, accordingly, that subjective or discretionary employment practices may be analyzed under the disparate impact approach in appropriate cases.
As one judge noted in Miller v. Countrywide, 571 F.Supp.2d 251, 258 (D. Mass. 2008), a mortgage lending discrimination case against Countrywide:
Where the allocation of subjective decision-making authority is at issue, the “practice” amounts to the absence of a policy, that allows racial bias to seep into the process. Allowing this “practice” to escape scrutiny would enable companies responsible for complying with anti-discrimination laws to “insulate” themselves by “refrain[ing] from making standardized criteria absolutely determinative.” Watson, 487 U.S. at 990. This is especially the case in this context. Unlike in the employment context, subjective criteria, unrelated to creditworthiness, should play no part in determining a potential borrower’s eligibility for credit.
By neglecting to recognize that a policy permitting discretionary decision making can let bias enter the system and that the overall effect of that bias can present a common question, the Supreme Court’s analysis of class certification of a disparate impact claim in Wal-Mart undermines, or perhaps eviscerates, Watson. To reconcile Wal-Mart and Watson, if it’s possible, one needs to look carefully, on a case-by-case basis, at the nature of the available proof.
If Wal-Mart makes sense as a rubric for disparate impact, it is perhaps only in connection with evaluating which individuals are entitled to damages. Absent analysis of each individual outcome, it is perhaps difficult to assess the monetary impact of the discriminatory effect in order to provide appropriate compensation. Traditionally, courts dealt with this by awarding injunctive relief and disgorgement or other forms of equitable penalties to be split among those exposed to the policy.25 More recently, however, cases like Coleman v. GMAC made clear that any relief for the individual effects of discrimination was unavailable to be awarded in conjunction with class certification for injunctive relief (Cubita, Willis, and Selkowitz Reference Cubita, Willis and Selkowitz2015).
Wal-Mart put a final nail in this coffin. Not only was certification for injunctive relief rejected, but by rejecting statistical evidence of the disparate effect of discretion as a valid basis for evaluating commonality under the class-action rule, one never gets to the question of whether injunctive relief, let alone whether monetary relief consistent with the injunction, is available. This is because finding commonality under Rule 23(a)(2) is a prerequisite to evaluating whether injunctive relief under 23(b)(2) is available at all.26 Absent the injunction, monetary relief incidental to the injunction never comes into play.
After Wal-Mart, almost no class remedies based on the impact of discretionary decision making remain.27 Remarkably, in Rodriguez v. National City Bank,28 the Court concluded that a bank could not even choose to settle a disparate impact mortgage lending claim against it for a class, because commonality under the class-action rule was necessary to approve the settlement. Seven million dollars that the bank was willing to pay to African American and Hispanic mortgage borrowers to settle claims was therefore returned to the bank and the class members were left with no remedy.
For private plaintiffs, Texas Department of Housing & Community Affairs v. The Inclusive Communities Project, Inc., supra, provides little comfort.29 Although Inclusive Communities does reaffirm the availability of disparate impact to establish discrimination under the FHA, it imposes restrictions on disparate impact claims that would doom any but the least ambitious disparate impact cases. Inclusive Communities emphasizes the importance of adequate safeguards at the prima facie stage to make sure that the prospect of disparate impact liability does not “almost inexorably lead” to the imposition of quotas and thus raise “serious constitutional questions.” In particular, Inclusive Communities exhorts judges to apply a “robust causality requirement” under which “a statistical disparity must fail if the plaintiff cannot point to a defendant’s policy or policies causing that disparity” (Hancock and Glass Reference Hancock and Glass2015). Moreover, even when plaintiffs can establish a prima facie case of disparity, the Inclusive Communities decision arguably expanded the scope of the defendant’s business necessity defense by finding that “policies are not contrary to the disparate-impact requirement unless they are ‘artificial, arbitrary, and unnecessary barriers.’” It is hard to see how this restriction can apply in the context of subjective decision-making processes that tend to result in biased choices. Again, Watson and its progeny may be nothing but dead letters.
Perhaps, after Wal-Mart, the Court is starting to move back toward the science of statistics as a tool for evaluating class cases. In Tyson Foods, Inc. v. Bouaphakeo, 136 S. Ct. 1036 (2016), the Court concluded that average time to don and doff equipment could be a basis fairly to award damages to class members with Fair Labor Standards Act claims for uncompensated time that they spent preparing for work.30 The Court concluded that statistical evidence may be used to certify and provide relief in a class action if the same sampling techniques could be used to establish liability in an individual action. Perhaps this points to an approach to measuring impact. If individuals can use representative statistics to show that their loan price exceeds what they might have paid if they were white, that same evidence should be equally available to the group.
10.4 Possible Futures
The foregoing impediments to private class-action litigation have coincided with the emergence of the CFPB as an active enforcer of ECOA disparate impact claims. The CFPB has been aggressive in “reminding” lenders that ECOA prohibits policies that result in a disparate racial impact unless those policies “meet a legitimate business need that cannot reasonably be achieved as well by means that are less disparate in their impact” (CFPB 2012). The Bureau has been aggressive in interpreting ECOA to apply to so-called “indirect lenders” – who, for example, may have arrangements to purchase loans from car dealerships at pre-established “buy rates” (CFPB 2013). A CFPB Bulletin explains:
Some indirect auto lenders may be operating under the incorrect assumption that they are not liable under the ECOA for pricing disparities caused by markup and compensation policies because Regulation B provides that “[a] person is not a creditor regarding any violation of the [ECOA] or [Regulation B] committed by another creditor unless the person knew or had reasonable notice of the act, policy, or practice that constituted the violation before becoming involved in the credit transaction.” This provision limits a creditor’s liability for another creditor’s ECOA violations under certain circumstances. But it does not limit a creditor’s liability for its own violations – including, for example, disparities on a prohibited basis that result from the creditor’s own markup and compensation policies.
Notwithstanding the Wal-Mart finding that granting discretion is “opposite of a uniform employment practice,” the CFPB has notified indirect lenders that discretion-granting policies that “permit dealers to increase consumer interest rates and that compensate dealers with a share of the increased interest revenues” may be actionable (CFPB 2013).
The Bureau’s aggressive stance has not been limited to just its interpretation of ECOA’s scope, but also in calling for “institutions subject to CFPB jurisdiction, including indirect auto lenders” to develop “a robust fair lending compliance management program” that includes regular assessment of lending policies “for potential fair lending violations, including potential disparate impact.” To avoid liability, indirect and direct lenders “should take steps to ensure that they are operating in compliance with the ECOA and Regulation [B],” including possibly “imposing controls on dealer markup” or “eliminating dealer discretion to mark up buy rates and fairly compensating dealers using another mechanism, such as a flat fee per transaction” (CFPB 2013). Thus, the CFPB has felt empowered to call on indirect lenders such as GMAC or Ford Motor Credit to exert their influence to substantially restructure dealership compensation or to engage in an ongoing manner in the same kinds of number-crunching undertaken by plaintiffs in the previous section.
The Bureau has translated these regulatory positions into a series of enforcement actions that have resulted in a series of multimillion-dollar settlements that have attracted the lending industry’s attention and ire. For example, in December 2013, the CFPB and the Justice Department ordered Ally Bank to pay $80 million in damages to consumers harmed by Ally’s auto loan pricing policies. The agencies found that “Ally’s markup policy resulted in African-American, Hispanic, Asian and Pacific Islander borrowers paying more for auto loans than similarly situated non-Hispanic white borrowers” (Ficklin Reference Ficklin2016).
Other actors, including cities and counties as well as national and local groups, that can assert standing under the civil rights laws may continue to pursue disparate impact claims that do not require class certification. Unfortunately, it is less clear that these actions can provide specific and targeted remedies for the economic harm to individuals that is associated with disparate pricing.
Finally, the principles described in this chapter may not apply to class-action cases designed to test the discriminatory impact of discrete practices, unrelated to discretion available to decision makers, that may lead to either disparate treatment or disparate impact claims. For example, if a bank assigns mortgage officers to its branches in white communities31 while making loans through a network of high-cost brokers in minority communities, class certification and class remedies may remain viable. Some of these practices may emerge most clearly as explanatory in communities of color where rates of foreclosure remain persistently high.
Some may argue that litigation remedies, whether initiated by private actors or governmental entities, are among the least efficient methods for establishing discipline and fairness in the housing market. Whether one accepts this premise turns on one’s views about voluntary compliance with new regulation, including the changes associated with Dodd-Frank, as well as one’s beliefs about the effectiveness of competition to regulate markets, and about whether new tools can achieve more complete consumer understanding of complex transactions. Others in this volume address those issues directly or indirectly (see, e.g., Bostic and Orlando, Chapter 13, this volume). Our view is that absent effective enforcement mechanisms, including meaningful opportunities for aggregation of claims, new mechanisms will be found to discriminate by manipulating the cost of housing credit for those least able to afford high credit prices.
10.5 Conclusion
The motivating force behind applying disparate impact theories to mortgage lending has been the happenstance that the defendants collect and retain all of the borrower characteristics that are relevant to the defendants’ underwriting decisions. Defendants are, in an important sense, estopped from criticizing plaintiffs’ regressions for not controlling relevant variables when the plaintiffs have controlled for all the variables that defendants relied on in their own underwriting.
However, given the increased hostility to class actions and private disparate impact claims, it is uncertain whether private plaintiffs can feasibly pursue such claims. At the moment, it seems most likely that disparate impact discipline of lenders will come from government enforcers, especially the CFPB.
Authors’ Note
The authors were involved as lawyers or expert consultants on a number of cases raising claims of disparate racial impacts in mortgage lending. Mr. Klein worked on many of the cases discussed in this chapter as a partner at Klein Kavanagh Costello, LLP before joining the Massachusetts attorney general’s office. The views expressed in this chapter are not necessarily the views of the Massachusetts attorney general’s office.
Open access