In 1992 the policy and research communities were stunned by findings from a study by analysts (Munnell et al., 1992) at the Boston Federal Reserve Bank (Boston Fed) showing that African-American applicants were 60 percent more likely to be denied mortgage credit than were white applicants.¹ The smoking gun of racial discrimination in lending had apparently been found, confirming what fair lending advocates had long asserted to be self-evident. The Boston Fed study was subsequently widely reviewed, critiqued, and reanalyzed, with a final consensus that the methodology correctly estimates a sizable racial difference in treatment by mortgage lenders in the conventional (nongovernment-insured) market.

The Office of Policy Development and Research at the U.S. Department of Housing and Urban Development (HUD) initially entered this debate with a national conference on Home Mortgage Lending and Discrimination, held in Washington, D.C., in May 1993. Revised and expanded versions of those conference papers will be published in 1996 by the Urban Institute Press. They will address, in a much more expanded manner than is possible here, the analytic, methodological, and policy issues associated with the measurement of lending discrimination. Included in that collection will be discussion and debate about the findings of lending discrimination reported in the path breaking Boston Fed study.

This volume of Cityscape presents and discusses a major research article that was first introduced at the May 1993 conference. The article involves an attempt to discern potential lending discrimination by examining home mortgage defaults. It is authored by re-searchers who worked together at the Federal Reserve Board: James Berkovec (now with the Federal Home Loan Mortgage Corporation [Freddie Mac]), Glenn Canner, Stuart Gabriel (now at the University of Southern California), and Timothy Hannon. I will refer to them collectively, as is done throughout this volume, as BCGH.

Their research captured considerable media interest after a presentation at a seminar in January 1995. The results appeared to refute the contention that mortgage-lending institutions discriminate against minorities by making it harder for them to qualify for loans. For example, the Wall Street Journal (1995) commented:

Federal Reserve Board member Lawrence Lindsey later concluded that the study shows that "blatant discrimination in the credit market is very rare today." (Lindsey, 1995.) A former chairman of the Federal Deposit Insurance Corporation also argued that:

In fewer than 3 years, perception of the evidence appeared to have shifted from finding a clear pattern of lending discrimination to challenging—or trying to challenge—the very existence of lending discrimination as a major public policy concern. Many people noticed a seesaw quality to the Boston Fed study and the BCGH research.

In other words the objective of the BCGH study was much less ambitious in scope than were the conclusions of many members of the media and other commentators. The novelty of both the database and the analytic methods makes an appropriate interpretation of their results complex and not readily understood. BCGH certainly realize this difficulty and have recommended that no changes in fair lending enforcement be made as a result of their work.

However, because media coverage of the BCGH research suggests that there has been overzealous enforcement of the Fair Housing Act (Title VIII of the Civil Rights Act of 1968, as amended) by HUD and other Federal agencies, this collection of essays was designed for two interrelated reasons: first, to promote a better understanding of the statistical techniques necessary to measure potential lending discrimination and, second, to address the fair lending policy implications of the BCGH study.

Steven Ross, George Galster, and John Yinger (RGY) participated in the January 1995 seminar and offered initial reactions. Their revised comments and criticisms of the paper presented by Berkovec and his colleagues form the second section of this colloquy.

Much of the scholarly debate surrounding the BCGH work involves the question of whether one should even attempt to measure discrimination by looking at loan default data, and this concern is addressed throughout the issue. It becomes a sometimes heated—even raucous—dialogue, because each contributor understands the policy importance of rigorous research on lending discrimination. The fabric of the discussion is woven by means of a point-counterpoint colloquy between the original authors and invited critics, followed by commentaries on the debate itself by other experts.

The BCGH Study
The BCGH study makes use of a database containing more than 200,000 Federal Housing Administration (FHA)-insured mortgages issued from 1987 to 1989 and analyzed with innovative and complex statistical techniques. Their article begins by noting the high level of attention given to the Boston Fed study and states that their intention is to "evaluate discrimination in home mortgage originations by examining performance of mortgage loan portfolios." This evaluation, unlike the Boston Fed study's approach, assumes that a prejudiced lender would approve mortgages for African-American borrowers only if they surpass minimum qualification standards. Such borrowers' credentials would make them less likely than minimally qualified white borrowers to default or foreclose on their loans. Thus the research hypothesis is that if lending discrimination exists, it is likely to be in the form of higher underwriting hurdles for marginally qualified minorities.

If this type of discrimination exists, BCGH argue, marginally qualified minority borrowers will exhibit lower foreclosure rates than will marginally qualified white borrowers. BCGH further research the question of the relative default costs of minority and white borrowers by examining whether or not defaults by individual minorities cost FHA lenders more than defaults by whites.

The authors find, after statistically controlling for many relevant factors, that African-American and white default rates differ by about 1.5 percentage points, with African-Americans more likely to default. They also report that the losses resulting from defaults by African-American borrowers are $2,000 to $3,000 higher than losses from defaults by white borrowers. These findings are then used to support the contention that African-Americans and whites are not treated unequally in the FHA loan underwriting process.

Overview of the Issues
As will be observed throughout this publication, BCGH clearly and consistently reiterate their own cautious view that their study's results "do not prove that discrimination does not exist; any claim to the contrary is obviously exaggerated." They also are quite forthright in stating that their method is biased toward not finding discrimination.

Notwithstanding such disclaimers, RGY argue that the BCGH approach is a fatally flawed method for studying discrimination in mortgage lending, because it is unable to address the pervasive problem of statistical discrimination. According to Yinger, statistical discrimination occurs when:

Minority status is uniquely correlated with credit risk, such that, in Galster's words, the BCGH analysis is "deeply flawed because it cannot extricate itself from a dilemma that renders the test ambiguous." Commentators John Quigley and Jan Brueckner basically agree with the conclusions of these critics, with Brueckner noting that the BCGH findings are completely ambiguous: They may be consistent with either the conclusion that lenders are prejudiced or the conclusion that they are not, and the effect of race on defaults thus becomes indeterminable. In Quigley's view, "The finding of disparities in default rates for black and white borrowers says nothing at all about discrimination in the housing or mortgage market."

Much of this debate, then, centers on the role of unobservable credit variables. According to Yinger and Ross, BCGH have had to "assume away" what may be the major source of mortgage lending discrimination: the systematic relationship between race and credit characteristics not observed by the lender. Lenders practice statistical discrimination, they state, because it is either too costly or not possible to collect all of the information needed to fully and properly underwrite against risk at the time of loan origination. Conversely, the BCGH analysis was designed to test only for nonstatistical discrimination, which is targeted more toward individual borrowers.

Yezer argues that BCGH have made a good-faith effort to include all of the variables available in HUD's databases and that they have gathered the most inclusive information to date for a default analysis. Brent Ambrose and Charles Capone also support BCGH by noting that the critics do not give them credit for the proxy variables they use. In their own rebuttal, BCGH argue that statistical discrimination is not really discrimination. Ross and his colleagues provide the appropriate rejoinder: The BCGH study lacks crucial data on credit history, and statistical discrimination is illegal behavior.

Yinger and Ross are also concerned that the BCGH findings could be explained by discriminatory treatment of minorities in the management of the default and foreclosure process. Since the default procedure is negotiated between lender and borrower, discrimination at that point—or, at the least, differential outcomes based on unobservable financial differences—would drive the BCGH results. In a useful contribution to this debate, Ambrose and Capone conduct their own analysis of FHA defaults and foreclosures and find that African Americans are provided more time to resolve their default than whites and that the post default foreclosure experiences of the two groups are virtually the same. Thus they conclude that "lenders do not discriminate against minority borrowers in de-fault" and that the Yinger argument against the BCGH study is not supported. Yinger, however, is concerned that Ambrose and Capone have not properly disentangled all of the supply and demand factors necessary to warrant such a robust conclusion and that it is unclear whether minorities are treated less well in other aspects of foreclosure proceedings.

While they are supportive of BCGH on the appropriateness of using default data to measure lending discrimination, Yezer, Ambrose, and Capone describe a substantial number of technical and research reasons why the BCGH analysis has failed to provide a definitive, persuasive analysis. More significantly, the Ambrose-Capone article, after reanalyzing comparable FHA data, finds no additional loan loss associated with defaults by African-American households once census tract characteristics are taken into consideration, thereby challenging one of the more notable results reported by BCGH. Additionally, Ambrose and Capone caution that the loan-loss data reported by BCGH could be the result of a number of data problems. In the absence of indication of a default loss, the BCGH analysis fails to demonstrate that there is a significant behavioral component to African-American default propensities, and this failure may, in turn, reflect a host of poorly measured factors that determine when and why a borrower defaults (Quercia and Stegman, 1992).

The article by Calvin Bradford and Anne Shlay questions whether the BCGH findings provide any guidance for understanding lending behavior in the conventional, non-FHA home mortgage market. They describe how the FHA market differs from the conventional market in both design and operation, so that FHA loans to minorities and minority neighborhoods are not issued using the risk assessment practices of many conventional lenders.²

Bradford and Shlay amplify this point by noting that their data from Chicago show "poor or improper underwriting practices, representing exactly the opposite of the more restrictive underwriting hypothesized by the BCGH study." The high number of defaulted loans in areas of minority concentration and racial change in Chicago indicates that lenders have used poor underwriting practices for such communities, confirming Bradford and Shlay's assessment of the role of independent mortgage companies in fostering racial transition. Yezer worries about the possibility that minority borrowers select lenders "who have less stringent standards" and that the BCGH findings reflect some degree of self-selection bias. Bradford, Shlay, and Yezer are concerned about aggregating data from a large number of FHA lenders who use different underwriting criteria for different markets.

Conclusion
My assessment of this path breaking research is exemplified by the comments of Robert Van Order and Peter Zorn, senior economists at Freddie Mac, who conclude that the Boston Fed's approach of using rejection of loan applications is "a potentially cleaner way of testing the proposition that minorities face bigger hurdles …" than relying on the more indirect approach of using default information (Van Order and Zorn, 1995). Although it is possible that the default approach may be useful as a means to test for lending discrimination, Berkovec and his colleagues have not convincingly proven this to be true.

For policy purposes, the small size of the racial difference that BCGH report makes it difficult to justify any major policy shift on fair lending enforcement. This view is stated by Acting Assistant Secretary for Fair Housing and Equal Opportunity Elizabeth Julian in the policy commentary that appears at the end of the volume. BCGH report a 1- to 2- percentage point difference that, while statistically significant, is not substantial enough to justify changes in HUD's fair lending policy, because there are many unexamined reasons why a difference this small may have occurred. Such a small difference, in the absence of any loan losses and accompanied by the host of technical limitations noted by the commentators, is clearly insufficient to justify a shift in fair lending policy priorities or to challenge the clearly documented pattern of lending discrimination shown in studies using direct or front-end measures of discriminatory treatment.

Several of the presenters in the colloquy argue that studies of the performance of minority loans should be part of an overall agenda of measuring and tracking the forms, virulence, and negative effects of lending disparities in both the insured and conventional markets. A program of research to assess the methodological contribution of the Boston Fed study's approach to understanding both racial discrimination in loan approvals and the role of race in loan performance would be most useful. Such research efforts could use expanded conventional and FHA loan files to address more definitively such questions as: What forms of racial discrimination exist? In which kinds of housing markets do they exist? What are the consequences?

Debate about the Boston Fed study and the BCGH research has greatly aided our understanding of the methodological and data requirements for a rigorous, full-scale examination of lending discrimination. Why, for example, is persistent evidence of differences in the treatment of minorities revealed in annual reports on the Home Mortgage Disclosure Act and in the investigations of allegations of lending discrimination by agencies such as HUD? Recent testing data gathered for HUD have shown numerous examples of the differential treatment of comparable white and African-American loan applicants by lenders in some of America's largest cities. These cases are currently under investigation.

As Assistant Secretary Stegman concludes in "From the Editor":

It is easy to see why the press might misunderstand the message from this complex, esoteric colloquy. However, there is no need to go much beyond the clear comments of BCGH that they never intended their article to be misconstrued as a test for the existence of any and all forms of lending discrimination.

As guest editor for this issue of Cityscape, I would like to express appreciation to all of the contributors for their willingness to participate in this exceptional form of discussion. The opportunity for a number of critics to comment on the BCGH article and the opportunity for them to respond is, I hope, a hallmark of the fairness that will make this volume useful to others.

John Yinger, Charles Capone, and Ira Goldstein (Office of Fair Housing and Equal Opportunity, HUD) have been particularly helpful to me throughout the preparation of this issue and have offered useful comments and suggestions. I am certain that they will not agree with all that I have written above. May I also express my appreciation to Ann Weeks and Carolee Lanier for their editorial assistance. My thanks to all.

Author
John Goering, after receiving his Ph.D. in sociology and demography from Brown University, taught at the University of Leicester in England, Washington University in St. Louis, and the Graduate Center of the City University of New York. Subsequently, he joined HUD's Office of Policy Development and Research, where he directs civil rights research and evaluation activities.

Notes
A note regarding terminology used in this issue is in order. There is a lack of consistency in the way authors use the terms blacks and African Americans. The National Academy of Sciences and the Office of Management and Budget have expended considerable effort over the past year in an attempt to clarify appropriate usage but have not issued a final recommendation. We have, therefore, elected to allow the authors to use the term of their choice and have not standardized such references.

2. There are critical differences between the FHA and portfolio lenders; see Benjamin, Heuson, and Sirmans, 1994.

References
Benjamin, John, A. Heuson, and C. Sirmans. 1994. "Are Depository Institutions and Mortgage Bankers Different?" Journal of Housing Research 5(1):139-170.

Isaac, William M. 1995. "Epilogue," in Anthony Yezer, ed., Fair Lending Analysis: A Compendium of Essays on the Use of Statistics, pp. 163-166. Washington, D.C.: American Bankers Association.

Lindsey, Lawrence. 1995. "Foreword," in Anthony Yezer, ed., Fair Lending Analysis: A Compendium of Essays on the Use of Statistics, pp. ix-xiv. Washington, D.C.: American Bankers Association.

Munnell, Alicia H., Lynn E. Browne, James McEneaney, and Geoffrey M.B. Tootell. October 1992. "Mortgage Lending in Boston: Interpreting HMDA Data." Federal Reserve Bank of Boston Working Paper No. 92-7.

Quercia, Roberto, and Michael A. Stegman. 1992. "Residential Mortgage Default: A Review of the Literature," Journal of Housing Research 3(2): 341-379.

Van Order, Robert, and Peter Zorn. 1995. "Testing for Discrimination: Combining De-fault and Rejection Data," in Anthony Yezer, ed., Fair Lending Analysis: A Compendium of Essays on the Use of Statistics, pp. 105-112.

Yinger, John. 1995. Closed Doors, Opportunities Lost: The Continuing Costs of Housing Discrimination. New York: Russell Sage Foundation, p. 183.

Wall Street Journal. January 26, 1995. "Fed Study Challenges Notion of Bias Against Minorities in Mortgage Lending," p. A16.