View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

ESSAYS ON ISSUES

JULY 1995
NUMBER 95

THE FEDERAL RESERVE BANK
OF CHICAGO

Chicago Fed Letter
Discrimination in
mortgage lending
In recent years much attention has
been devoted to the issue of racial
discrimination in the home m ort­
gage lending market. This interest
has been fostered by the release of
mortgage lending data under the
Home Mortgage Disclosure Act of
1977 (HMDA). This act, which re­
quired depository institutions to
disclose mortgage originations in
m etropolitan areas by census tract,
was am ended in 1989 to require
lenders to report the disposition of
every mortgage loan application,
along with the loan am ount and the
race or national origin and annual
income of each applicant. Analysis
of the raw data contained in the an­
nual HMDA data releases shows that
there are persistent disparities in
denial rates between white and m inor­
ity applicants. The table below re­
ports HMDA denial rates by ethnicity
for home purchases involving govern­
ment-backed or guaranteed m ort­
gages and conventional mortgages.
As this table shows, denial rates for
black and Hispanic applicants were
appreciably higher than those for
whites in each year studied.
The persistence of these disparate
rejection rates has led many industry
critics to accuse banks and other mort­
gage lenders of engaging in discrimi­
natory lending practices. However,
HMDA data taken in isolation cannot
shed much light into the question of
discriminatory lending practices. This
is because the HMDA releases do not
provide other information crucial to
the credit granting decision, such as
the loan applicant’s credit history, net
worth and general financial condition,
and employment history.
In response to this weakness in the raw
HMDA data, the Federal Reserve Bank

Loan denial rates
G overnm ent-backed loans
Applicant

1990

1991

1992

1993

Conventional loans
1990

(percent)

1991

1992

1993

(percent)

A m e rica n In d ia n /
A laskan N ative

22.5

22.1

17.5

17.5

22.4

27.3

26,6

27.8

A sia n /P a cific
Isla n d er

12.8

12.5

13.5

11.7

15.0

15.3

14.6

Black

26.3

26.4

23.8

22.2

12.9
33.9

37.6

35.9

34.0

H ispanic

18.4

18.9

18.5

14.6

21.4

26.6

27.3

25.1

14.4

17.3

15.9

15.3

W h ite

12.1

16.3

12.8

O ther

18.4

16.3

16.0

17.8

19.0

19.9

21.0

23.1

J o in t (W h ite /
M in o rity )

14.1

15.9

14.8

14.7

14.9

17.5

17.6

17.3

11.8

Source: Federal Reserve System.

of Boston conducted a study of mort­
gage denial rates in the Boston metro­
politan area based on a much wider
range of data.1 Using a sample of ap­
plications from the 1990 Boston met­
ropolitan area HMDA files, Fed econo­
mists augmented the HMDA data with
detailed information on applicants’
financial condition, credit history,
personal characteristics, and the un­
employment rate in the industries in
which applicants were employed.
They then estimated the probability of
a particular mortgage loan application
being denied.
Even with the addition of this infor­
mation, race still appeared to have a
statistically significant effect on the
probability of being denied. When
other factors were held constant, the
rejection rate for black and Hispanic
applicants was about 1.6 times that for
whites. Despite the interesting in­
sights provided by the Boston Fed
analysis, some have challenged its
conclusions, questioning both the
reliability of the data and the underly­
ing empirical m odel.2 Given these
concerns, the robustness of the
study’s results remains an open ques­

tion. However, while the Boston
study has been criticized, it represents
the first rigorous study of the ques­
tion of discrimination in the loan
approval decision.
This Chicago Fed Letter reports the
results of a recent Chicago Fed study
reexam ining the role of race in the
mortgage loan approval process.3 We
began by carefully verifying and vali­
dating the data used in the Boston
study. Then, unlike the Boston study,
we analyzed the data using a model
that tapped interaction effects among
variables, examining approval proba­
bilities for various subsets of appli­
cants. While the Boston study found
that Boston-area minority applicants
were more likely to be rejected than
white applicants with similar charac­
teristics, our study indicated that this
was the case only for applicants at the
margins of creditworthiness. Margin­
al black and Hispanic applicants ap­
peared to be held to higher quantita­
tive standards on such factors as
credit history and debt ratios than
were similarly situated marginal white
applicants.

The cultural affinity and thicker
file hypotheses

Despite the existence of legislation
outlawing the use of irrelevant factors
such as race and sex in mortgage
lending decisions, lenders might still
use these noneconomic facts as sig­
nals of whether to obtain additional
information on marginal borrowers.
Such information could ultimately
influence the credit approval deci­
sion. One recent hypothesis suggests
that a lack of “cultural affinity” be­
tween white loan officers and m inori­
ty applicants reduces the reliability
and accuracy of the loan officers’
subjective evaluations, as a result of
which white banks have more de­
m anding approval standards for black
and Hispanic borrowers.4 According
to this hypothesis, if the marginal cost
to white lenders of obtaining addi­
tional credit information on white
borrowers is lower than that associat­
ed with minority borrowers, then
white loan officers will rely more
heavily on (inexpensive) objective
loan application information for their
minority applicants than for their
white applicants. As a result, white
applicants with marginal objective
indicators of creditworthiness will be
more likely to be approved than mi­
nority applicants with identical mar­
ginal credit records, because the
whites will have additional subjective
information supporting their case.
A somewhat similar hypothesis ad­
dresses the so-called “thicker file”
phenom enon. Accepted marginal
white mortgage applicants often have
thicker loan application files than
rejected marginal minorities. The
common presum ption is that white
applicants receive special counseling
or extra coaching by sympathetic
white loan officers, while minority
applicants do not. The cultural affini­
ty and thicker file hypotheses both
predict similar outcomes; they differ
in that the cultural affinity hypothesis
does not necessarily imply that whites
receive special counseling and coach­
ing or even have thicker application
files than minorities.
The Chicago Fed study tested the
cultural affinity hypothesis by examin­
ing whether loan officers perceive

probability was 90%, blacks and His­
panics’ was 82%. These results suggest
that for high-quality applicants, race
was unim portant to lenders. It was the
marginal minority applicant who was
impacted by race.
It appears, then, that lenders did
treat the objective loan application
inform ation of marginal black and
Hispanic applicants differently from
that of m arginal white applicants. In
Our data and statistical model
particular, credit history had a sub­
stantially greater im pact on the prob­
To correct for data errors and missing ability of loan approval for marginal
values in the initial sample used in
black and Hispanic applicants than
the Boston study, we adopted an ex­
for marginal whites. While the ap­
tended version of the criteria devel­
by about
oped by Carr and Megbolugbe.5 Our proval probability dropped minority
13 percentage points for a
final sample contained 1,991 observa­ applicant undergoing an adverse
tions: 1,726 approvals and 265 rejec­
change in credit history, the proba­
tions. The overall characteristics of
bility dropped by only 5 percentage
this subsample m irrored that of the
points for a white applicant undergo­
original data set.
ing a similar change.
Our statistical analysis used a standard These results provide strong support
logistic or logit regression model in
for the cultural affinity hypothesis.
which the lender’s loan approval
On the other hand, they provide no
decision is determ ined by a com pre­
support for the thicker file hypothe­
hensive set of borrower characteris­
sis. The estimated direct effect of
tics. The model predicts the proba­
factors proxying for file thickness
bility that a mortgage applicant with a on the probability of approval was
given set of characteristics will receive
loan approval. The variables entered insignificant.
into the model were very similar to
The study shows that race alone did
those in the Boston Fed study, includ­ not determ ine the differences in
ing standard financial ratios such as
denial rates between applicants. To
housing expenses relative to monthly analyze the determ inants more fully,
income, total debt obligations relative however, one must consider the inter­
to monthly income, and other miscel­ action between race and several other
laneous inform ation such as marital
variables. In particular, the relation­
status, employment status, and wheth­ ship between race and creditworthi­
er the loan applicant had a cosigner.
ness is key. The impact of race on the
creditworthiness of a loan applicant
will depend on the level of the appli­
Race effects
cant’s ratio of total monthly obliga­
The results of our study indicated that tions to total monthly income. We
when the credit history of an applicant com puted the effect of race on creditwas classified as good (i.e., the appli­
worthiness within two levels of the
cant had no accounts 60 or more days debt obligation ratio—the lowest
past due), the approval probability for quartile and the third quartile. For
white applicants versus black and His­ applicants with bad credit histories,
panic applicants was not appreciably
the effect of race on denial rates was
different—96% for whites and 95% for small and statistically insignificant as
blacks and Hispanics. On the other
long as the obligation-to-income ratio
hand, when credit history was classi­
was low or favorable. However, when
fied as bad (i.e., the applicant had one the ratio increased to an unfavorable
or more accounts 60 or more days past level, the racial effect became larger
due), the difference in approval prob­ and significantly significant, with
ability was 8 percentage points in favor black and Hispanic applicants being
of white applicants: Whites’ approval
objective inform ation such as credit
history and leverage ratios differently
for minority applicants than for
whites. In addition, using a measure
of applicants’ prior credit m arket
experience, the study examined the
thicker file hypothesis. O ur results
provided support for the cultural
affinity hypothesis but not the thicker
file hypothesis.

adversely affected. For example,
when the obligation-to-income ratio
changed from a favorable 30% to an
unfavorable 60%, the approval proba­
bility for marginal black and Hispanic
applicants decreased by 72.1 percent­
age points, com pared to only a 24.5
percentage point reduction for mar­
ginal white applicants. Thus, as the
ratio became less favorable, marginal
white applicants appeared to be
judged more creditworthy than were
similarly situated marginal black and
Hispanic applicants. These findings
lend additional support to the cultur­
al affinity hypothesis.

Other results

The other findings reported in the
study are consistent with traditional
expectations. Specifically, increases
in the ratio of total debt obligations
to monthly income, evidence of prior
public record defaults, self-employ­
ment, and having a less than favor­
able debt repayment history all low­
ered the probability of approval. So
did being unm arried, being em­
ployed in an industry with a high
probability of unemployment, and
purchasing multifamily property. On
the other hand, gender, the appli­
cant’s num ber of dependents, and
the racial or economic status of the
neighborhood in which the property
was located had no significant impact
on the probability of approval. The
finding that neighborhood racial or
economic status had no impact on
approval probability implies that the
loan officers in the sample did not
engage in the practice of redlining,
i.e., denying loans on the basis of the
racial or economic status of the
neighborhood in which the property
was located.

Implications

Overall, the study suggests that the
marginal minority applicants in the
sample were held to higher quantita­
tive credit standards than were m ar­
ginal white applicants. This is a
m ore precise finding than that of the
Boston Fed’s study. O ur conclusion
is im plied by the behavior of the race
variable under differing sets of con­

ditioning inform ation. Race was not
a significant factor in the accept/
reject decision for applicants with
good credit profiles. However, it was
very significant for those with bad
credit histories, and a bad credit
history, in turn, lowered the proba­
bility of approval for minorities
m uch more than it did for whites.
Similarly, race became an im portant
factor only for those with high ratios
of monthly debt to income. Minority
applicants with high debt ratios were
very much less likely to receive loan
approval than similarly situated white
applicants.
These findings are consistent with the
existence of a cultural affinity be­
tween white lending officers and
white applicants, and a cultural gap
between white loan officers and mar­
ginal minority applicants. Bridging
or closing this gap will require more
research to identify the factors that
determ ine the repayment patterns of
marginal minority and non-traditional borrowers.
The search for evidence of racial
discrimination in economic life is
exceedingly difficult but can be made
more manageable with the aid of
statistical models. W hen used appro­
priately, these models can help regu­
lators and researchers uncover dis­
criminatory practices in mortgage
lending. However, they should not
be treated as replacements for sound
judgm ent and inquiry. If one incor­
porates additional information into
these models such as that included in
bank underwriting policies and guide­
lines or other “omitted variables,” the
models sometimes yield different
results.6 Thus, in using statistical
models in regulatory processes, poli­
cymakers face a classical problem:
They must decide whether to expend
additional resources to acquire more
comprehensive information to build
more customized models, when those
models may add little value to the
policymaking process.
—William C. H unter
Senior Vice President and
Director of Research

1Alicia H. Munnell, Lynne E. Browne,
James McEneaney, and Geoffrey Tootell,
“Mortgage lending in Boston: Interpret­
ing the data,” Federal Reserve Bank of
Boston, working paper, 1992.
2David K. Horne, “Evaluating the role of
race in mortgage lending,” FDICBanking
Review, Vol. 7, Spring/Summer 1994, pp.
1-15, and Stan Liebowitz, “A study that
deserves no credit,” The Wall Street Jour­
nal, September 1, 1993, p. A14.
William C. Hunter and Mary Beth Walk­
er, “The cultural affinity hypothesis and
mortgage lending decisions,” Federal
Reserve Bank of Chicago, working paper,
1995, forthcoming.
4Charles W. Calomiris, Charles M. Kahn,
and Stanley D. Longhofer, “Housingfinance intervention and private incen­
tives: Helping minorities and the poor,”
Journal of Money, Credit and Banking, Vol.
26, August 1994, pp. 634-674.
5
James H. Carr and Isaac F. Megbolugbe,
“The Federal Reserve Bank study on
mortgage lending revisited,” Federal
National Mortgage Association, Office of
Housing Research, working paper, 1993.
6See, for example, Mitchell Stengel and
Dennis Glennon, “Evaluating statistical
models of mortgage lending discrimina­
tion: A bank-specific analysis,” Office of
the Comptroller of the Currency, Eco­
nomic and Policy Analysis, working pa­
per no. 95-3, May 1995.
Michael H. Moskow, President, William C.
Hunter, Senior Vice President and Director of
Research; David R. Allardice, Vice President,
regional programs; Douglas Evanoff, Assistant
Vice President, financial studies; Charles Evans
and Kenneth Kuttner, Assistant Vice Presidents,
macroeconomic policy research; Daniel Sullivan,
Assistant Vice President, microeconomic policy
research; Anne Weaver, Manager, administration;
Janice Weiss, Editor.
Chicago Fed Letter is published monthly by the
Research Department of the Federal Reserve
Bank of Chicago. The views expressed are the
authors’ and are not necessarily those of the
Federal Reserve Bank of Chicago or the
Federal Reserve System. Articles may be
reprinted if the source is credited and the
Research Department is provided with copies
of the reprints.
Chicago Fed Letter is available without charge
from the Public Information Center, Federal
Reserve Bank of Chicago, P.O. Box 834,
Chicago, Illinois, 60690-0834, (312) 322-5111.
ISSN 0895-0164

Tracking Midwest manufacturing activity
Purchasing managers’ surveys (production index)
80

Manufacturing output indexes
(1987=100)
Apr.
MMI
IP

Month ago

Year ago

138.6
123.3

141.6

132.2
118.4

124.0

70

Motor vehicle production
(millions, seasonally adj. annual rate)
Apr.

Month ago

Year ago

Cars

6.5

7.1

6.7

Light trucks

4.9

5.6

4.2

Purchasing managers’ surveys:
net % reporting production growth
May

Month ago

60

50

Year ago

MW

54.2

60.2

71.7

U.S.

47.9

55.3

62.2
40

1992

1993

Inventory paring prom pted retrenchm ent in Midwest industrial output
growth in recent months. The composite index for purchasing m anagers’
surveys in Chicago, Detroit, and Milwaukee fell back significantly in March,
April, and May. This indicator continued to depict positive growth in the
m anufacturing sector, but just barely. Even after its recent decline, the Mid­
west index remains in line with or stronger than its level during previous epi­
sodes of slowing growth in late 1991, mid-1992, and mid-1993.

1994

1995

Sources: The Midwest Manufacturing Index (MMI)
is a composite index of 15 industries, based on
monthly hours worked and kilowatt hours. IP rep­
resents the Federal Reserve Board industrial pro­
duction index for the U.S. manufacturing sector.
Autos and light trucks are measured in annualized
units, using seasonal adjustments developed by the
Board. The purchasing managers’ survey data
for the Midwest are weighted averages of the sea­
sonally adjusted production components from the
Chicago, Detroit, and Milwaukee Purchasing Man­
agers’ Association surveys, with assistance from
Bishop Associates, Comerica, and the University of
Wisconsin-Milwaukee.

IU9-ZZZ (2I£)

PF80-06909 siouipi ‘oStesuo

P£8 xoH O d

joiuoL) uopninjojuj oiyqnj

OOVOIH3 TO MNW TAdTSTd lYHTQTT

T > n o r] p a j o b t o i i p )


Federal Reserve Bank of St. Louis, One Federal Reserve Bank Plaza, St. Louis, MO 63102