Full text of Working Paper (Federal Reserve Bank of Cleveland) : Accounting for Racial Differences in Housing Credit Markets, Working Paper 93-10

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

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Working Paper 93 10

ACCOUNTING FOR RACIAL DIFFERENCES IN
HOUSING CREDIT MARKETS
by Robert B. Avery, Patricia E. Beeson, and Mark S. Sniderrnan

Robert B. Avery is an associate professor in the Department
of Consumer Economics and Housing at Cornell University,
Ithaca, New York, and Patricia E. Beeson is an associate
professor of economics at the University of Pittsburgh; both
are research associates at the Federal Reserve Bank of
Cleveland. Mark S. Sniderman is vice president and
associate director of research at the Federal Reserve Bank
of Cleveland. This paper was presented at the Conference
on Discrimination and Mortgage Lending: Research and
Enforcement, sponsored by the Department of Housing and
Urban Development, Washington, D.C., May 18-19, 1993.
The authors would like to thank Glenn Canner, Stuart
Gabriel, Stuart Rosenthal, John Yinger, and Peter Zorn for
helpful comments.
Working papers of the Federal Reserve Bank of Cleveland
are preliminary materials circulated to stimulate discussion
and critical comment. The views stated herein are those of
the authors and not necessarily those of the Federal Reserve
Bank of Cleveland or of the Board of Governors of the
Federal Reserve System.

December 1993

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

ABSTRACT
The release of individual housing credit application data, combined with lender and
neighborhood information required by amendments to the Home Mortgage Disclosure Act
(HMDA) in 1989, has offered new opportunities to examine the roles of both
neighborhood and individual race in credit availability. The extent to which objective
lending criteria are responsible for observed differences in home mortgage credit denial
rates, versus discriminationbased on income, race, or neighborhood (redlining), has been
the subject of considerable debate.
This paper provides a more detailed documentation of racial and neighborhood differences
in denial rates than has previously been available. Using estimates from a fixed-effects
linear probability model to decompose racial differences in application denial rates, the
authors find persistent variations between white and minority applicants, particularly
blacks. The variance is widespread and remains even after lender, neighborhood, and
applicant economic characteristics are accounted for. While the HMDA data do not
contain enough relevant information about the loan applications to draw any firm
conclusions about the reasons for these differences, some possibilities include property
location, credit or employment histories, loan-to-value ratios, or other factors considered in
the loan evaluation process that are not included in the HMDA file.

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Introduction
Despite the passage of several laws related specifically to racial differences in
housing credit availability, data constraints have limited the number of studies of this
issue.' Most existing studies use census-tract-level or lender-level data collected under
the Home Mortgage Disclosure Act (HMDA) to infer racial differences. Although
findings from such work are by necessity indirect, there is a persistent inference of
substantial differences in the availability of mortgage and other credit across racial
groups. Unfortunately, most of this work has been hampered by the inability to separate
the effects of the race of the applicant from the racial composition of the applicant's
neighb~rhood.~
Studies that use detailed applicant-level information to examine the
direct effects on mortgage denial rates of both property location and the race of the
applicant are rare.3
The release of individual application data, combined with lender and
neighborhood data as required by amendments to the HMDA in 1989, offers
unprecedented new opportunities to examine the issue of the role of both neighborhood
and individual race in credit availability. Early reports based on the 1990 HMDA data
document differences in denial rates on home mortgage credit applications by race and
income of applicants and by the average income and racial composition of
neighborhoods (see Avery, Beeson, and Sniderman [1993a] and Canner and Smith [1991,
19921). The extent to which objective lending criteria are responsible for these
differences, versus discrimination based on income, race, or neighborhood (redlining),

has been the subject of much analysis and debate.
In this paper, we provide a more detailed documentation of racial and

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

neighborhood differences in denial rates than has henceforth been available. For each
of three loan products (home purchase, refinance, and home improvement), we use
estimates from a fixed-effects linear probability model to decompose racial differences in
application denial rates into five components reflecting the portion attributable to 1)
economic characteristics of the applications reported in HMDA (income, loan amount,
loan type, etc.), 2) overall denial rates of the lenders receiving the application, 3) the
metropolitan statistical areas (MSAs), 4) census tract locations of the property, and 5) an
unexplained residual. We then compare these components across MSAs, across
neighborhood types grouped by income and racial composition, across types of lenders,
and for central city and suburban areas. We also compare racial differences in denial
rates across applications grouped by predicted denial rates based on all factors except
race.
Our objective in conducting this analysis is twofold. First, we are interested in
determining whether racial differences in credit approvals reflect activity in a small
subset of markets or whether they are endemic to most markets. Although significant
media attention has been paid to the issue of race and mortgage lending, preliminary
studies using the HMDA data have been limited in scope and restricted to either
individual cities or specific loan products. For example, in a study, that has received wide
media publicity (Mumell et al. [1992]), the Boston Federal ~ e s e r v eBank conducted an
expanded survey of loan applications in the Boston area and concluded that even when

an extensive list of individual applicant characteristics was controlled for, black and
Hispanic applicants were significantly more likely to be denied than white applicants.

This study, however, was limited to one loan product (home purchase loans) and one

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

city. Thus, it is not clear whether the authors' conclusions can be generalized or are
specific to certain areas. Second, as stated above, we are interested in determining
whether racial differences in lending stem from variations in applicant characteristics
(other than race), differences in the neighborhoods in which properties are located, or
racial differences that cannot be explained by these factors.
By way of preview, we find that denial rates for minority applicants are
consistently higher than those for white applicants with othemke identical attributes (as
reported in the HMDA data) who are applying for loans with the same lenders, and for
properties located in the same neighborhoods. We also find significant neighborhood
effects that differ across racial groups: Blacks, in particular, are more likely to apply for
loans for properties in neighborhoods with higher denial rates, ceteris paribus, than are
white applicants. On average, these neighborhood effects are less pronounced than
individual effects, although they are almost equal for home improvement loans. We find
a remarkable degree of consistency in these conclusions across geographic markets and
loan products, indicating that the observed racial differences in denial rates are
widespread and cannot be attributed to a small subset of markets. Although our analysis
reveals substantial and consistent differences in denial rates related to the race of the
applicant, even after controlling for a number of applicant characteristics, we emphasize
that the HMDA data do not contain enough relevant information about the loan
applications to draw any firm conclusions about the reasons behind these phenomena.
These residual differences may be due to credit histories, employment histories, loan-tovalue ratios, or other factors considered in the loan evaluation process that are not

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

included in the HMDA file, or may be the result of differential treatment based solely
on the race of the applicant.
The remainder of this paper is organized as follows. In the next section we
present a simple framework for analysis. In section II we provide a brief description of
the HMDA data and summary sample statistics. Section ID summarizes our results, and
concluding remarks are given in section IV.

I. Framework and Empirical Model

Consider the following simple, yet fairly general, framework in which to evaluate
the empirical findings of this study. Assume that the risk of each loan application given
all available ex ante information can be expressed as a risk score, RS. Further assume

that each lender decides to approve or deny an application based on a comparison of its
risk score and the lender's maximum acceptable risk. If the risk score is above a cutoff,
c, the loan is denied; otherwise the loan is accepted. Note that this abstracts from the
issue of price by assuming either that lenders price all loans equally or, because of
problems of moral hazard and adverse selection, that lenders have a maximum risk
acceptable at any price.
This model of lender behavior is deterministi~'butin reality error is likely to
enter the process. First, lenders may not know, or use, all available information in
computing risk scores. In this case, RS would be their estimate of the applicant's risk
given the information they use, and the loan-granting dkcision would still be made
deterministically, but based on a different set of information. To a researcher attempting

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

to quantify lender behavior, this case seems identical to the full information case
(assuming the researcher has access to all information used by the lender). A second
potential source of error is more relevant for this paper. Lenders may use risk score (or
their own estimate) and behave deterministically, but an external researcher may only
observe the lender's assessment of risk with error. That is, researchers may observe a set
of instruments for risk score for which they believe
(1)

RS

=

X/3

+ e,

where e is a stochastic error term. This implies that
(2)

Denial = 1 if

X'S + e > c,

and

Denial = 0 otherwise.
To an external researcher, who does not observe e, the evaluation process appears to be
probabilistic.
If only the lender action (acceptldeny), and not the risk score, is observed,
4

estimation of the parameters in equation (1) requires assumptions about the error term,
e. If the error in (1) is assumed to be uniform, then the probability that a loan
application will be denied, given X, is proportional to X'/3 plus a constant, and the
parameters in (1) will be estimable from a linear probability model. If e is normal, then
equation (2) gives rise to a probit probability model; and if e is double exponential, then

(2) gives rise to a logistic probability model. Although the scaling of parameters depends
critically on .the model form, the relative magnitude and signs of the parameters are
likely to be robust with respect to the model form chosen.
Df particular interest for this paper is the robustness (and interpretation) of racial

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

shift factors that may appear in X;,.Racial shifts may appear for several reasons. First,
race itself may be a predictor of future behavior and thus enter the risk score directly.
This might occur, for example, because minorities face discrimination in labor markets

and thus have more variable income. This would appear as different risk scores for
otherwise equal applications of different racial groups, or as racial shifts in estimated ps.
Note that for reasons of cost, lenders may choose to use estimates of RS rather than
fully computing it. In this case, race might be an instrument for the variables they do
not use.
Second, lenders may practice overt discrimination, and set a lower cutoe c, for
minorities. To an observer who looks only at the accept/deny process, this case would
be observationally equivalent to the first case. Overt discrimination may also take the
form of lenders (or a subset of lenders) randomly denying a fixed percentage of
minorities. This will also produce a racial shift.
Third, lenders may in fact not use race, and there may not be any racial shifts in
the true risk scores. However, race may be correlated with the omitted variables in the
error term, e, in equation (1). Minority applications could differ from others in the
expectation of e given X To the external researcher measuring RS with error, racial
shifts would show up in estimated ps, making this observationally equivalent to the first
two cases, even though race is not used by lenders and does not enter RS. Note that the
better that X is specified, the less this effect should matter.
We might also observe a combination of these effects. For example, only a subset
of lenders might have lower risk thresholds for minority applications. In this instance,

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

racial shifts would represent the average lender effect. Moreover, they would also imply
consistent residual differences across lenders in overall denial rates (we would expect
differences across lenders for other reasons, such as price or preferences for risk). We
might also observe combinations of different racial cutoffs and variations in the expected
values of the omitted variables, e. Again, the measured residual differences correlated
with race would represent a combination of effects.
The important point to emphasize here is that each of these sources of racial
shifts, with very different policy implications, is likely to produce observationally
equivalent results. Moreover, the estimated shifts will be sensitive to the econometric
model form chosen. Unfortunately, there is little other than computational convenience
to argue for a particular form (we actually employ a linear probability model for this
reason). Thus, despite the obvious value in quantifying racial shifts in denial functions,
these estimates, regardless of what they are, will be incapable of distinguishing among
competing causal models.

Em~iricalModel
Our empirical specification follows the framework set out above. We assume that
each mortgage application's risk can be represented as a function of the economic
characteristics (such as income), neighborhood, market, lender, and race of the applicant.

As noted above, we have no basis with which to select a particulai econometric model
specification. However, the size of the data set dictates that in practice we assume a
linear probability model specification. We thus estimate a model where the probability
that a random loan application would be denied is linear in the following terms:

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

where DENIAL is one if the ia application using the La lender in the

M? MSA and Ith

census tract is denied, and zero otherwise. MSA, TRACT and LENDER are dummy
variables indicating which MSA, census tract, and lender the application relates to, and e
isaa residual. AC is a vector of application characteristics, other than race, reported in
the HMDA data. AC includes gender, marital status, occupancy, income, loan amount,
income-to-loan ratio, federal loan guarantee (Federal Housing Administration [FHA] or
Department of Veterans Affairs [VA]). RACE is a dummy variable indicating the race
of the applicant and co-applicant. The model is specified and estimated separately for
each of three types of loan applications: home purchase, refinance, and home

improvement.
To help minimize the possibility that the differences within and across
neighborhoods we identify do not reflect nonlinearities in other effects that are
correlated with location, we allow for a considerable degree of nonlinearity in the effects
of individual characteristics in estimating equation (3). Income and loan amount are
entered as linear spline functions with seven knots each, and the ratio of income to loan
amount is entered as a series of six dummy variables. Moreover, a five-knot linear spline
for income is interacted with a dummy variable indicating the presence of a co-applicant,

and with dummy variables indicating that the application is for an FHA or VA loan.
Similarly, a five-knot linear spline of loan amount, and the six dummy variables
indicating ranges of values for the ratio of income to loan amount, are also interacted

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

with a dummy variable indicating applications for FHA or VA loans. We also include
dummy variables for six applicant and two co-applicant racial categories, and the racial
dummies interacted with FHA and VA loan dummies.
To reduce the computing requirements, the actual estimation was done in two
stages. In the first stage, equation (3) was estimated with the individual application
characteristics (AC) and separate intercepts for each lender-census tract combination
included as single-component fixed effects? The MSA, lender, and tract effects are thus
intertwined in these effects. In the second stage, an iterative procedure (equivalent to
regressing the fixed-effects intercepts against MSA, census tract, and lender dummies)
was used to identify the MSA, tract, and lender effects. By construction, the MSA
effects were normalized to have overall sample means of zero, and within each MSA,
lender and tract means were normalized to zero. In cases where lender and tract effects
were not identified (a lender was the only lender in a tract and did all of its business
there), the effect was assigned to the tract.

11. Data

AU commercial banks, savings and loan associations, credit unions, and other
mortgage lending institutions (primarily mortgage bankers) that have assets of more than

$10 million, make at least one mortgage loan, and have an office in an MSA are
required to report on each mortgage loan application acted upon by the institution
during the calendar year? They must report the loan amount, the census tract of the
property (if in an MSA), whether the property is owner-occupied, the purpose of the

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

loan (home purchase, home improvement, or refinancing), loan guarantee (conventional,

FHA, VA), action taken by the lender (loan approved and originated, application
approved but withdrawn, application denied), the race and gender of the loan applicant
(and co-applicant, if any), and the income relied upon by the lending institution in

making the loan decisiort6
In total, 9,333 financial institutions made HMDA filings for 1990 on 6,595,089
loan applications. Our analysis focuses on the 3,489,235 loan applications for 1-4 family
properties in MSAs that were acted upon by the lenders? Of these loans, 1,984,688
were home purchase loans, 716,595 were applications to refinance existing mortgage
loans, and 787,952 were applications for home improvement loans (generally second or
third mortgages). These applications were received by 8,745 separate institutions
operating in 40,008 census tracts in all 340 of the MSAs in the United States defined as
of 1990. We define lenders at the MSA level: Thus, an institution reporting applications
for two different MSAs is treated as two different lenders. There are 23,248 such
lenders in our sample.8
Descriptive statistics for the applications reported in the 1990 HMDA are found

in table 1. Statistics are given separately for home purchase, refinancing, and home
improvement loan applications. Clearly, housing credit applicants are a select sample of
American households. Household mean income ($63,071) is substantially higher than
that reported for all households in the 1989 Survey of Consumer Finances ($35,700)?
The racial composition of the study sample also appears to differ from that of all U.S.
households. Blacks constituted 6.9 percent of the housing loan applicants, yet were 7.4

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

percent of the homeowners and headed 11.2 percent of the households in 1990.
Similarly, Asians, native Americans, and others accounted for 5.6 percent of the housing
loan applicants but only 2.1 percent of the homeowners and 3.0 of the households.
Hispanics were more evenly represented: 6.6 percent of the applicants, 4.1 percent of
the homeowners, and 6.4 percent of the

household^?^
<

It is also apparent that denial rates differ substantially by race for all three types
of loans (see table 2). Denial rates for black applicants are about twice as high as those
for white applicants, and for Hispanic applicants the rate is about 50 percent higher than
for whites. Other racial differences are also apparent, particularly with respect to black
applicants. Black applicants are more likely to be single and are more likely to apply for
federally guaranteed loans. In addition, a larger portion of loans originated to black
applicants are subsequently sold, and credit history is given as a reason for denial more
often. Furthermore, while the median income and loan amounts for black applicants are
considerably lower than those for white applicants, the ratio of the two is fairly similar.

In contrast, the ratio of median loan amount to median income is consistently higher for
Hispanic applicants than for the other two racial groups.

111. Results

The parameter estimates for the denial rate regressions (equation [3]) are
reported in tables 3, 4, and 5."

A positive coefficient &I be interpreted as the

expected increase in the probability that an applicant's loan would be denied resulting
from a one-unit increase in the independent variable holding all other variables constant

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

- specifically, the applicant's MSA, census tract, and lender.

Thus, the coefficients on

race, for example, represent the expected difference in the probability that a white and
black applicant with the same income, gender, FHA/VA status, loan amount, MSA,
census tract, and lender will have their loan application denied. Thus interpreted, the
estimated black/white (.103), Hispanic/white (.040) and, to a lesser extent, the native
Americanlwhite (.028) and other racelwhite (.030) differences for conventional home
purchase loans are quite significant. Differences are similar for FHA loans (.116, .030,

.028,and .040, respectively). There is little residual difference between Asian and white
denial rates on home purchase loans (.008).
Significant racial differences also exist for denial rates on refinance and home
improvement loan applications. Compared with home purchase applications, the
blacklwhite difference is somewhat smaller for conventional refinance (.070) and home
improvement (.080) loan applications. The same is true of the native American/white
differences. However, for Hispanic, Asian, and other race applicants, differences from
white denial rates for refinance and home improvement applications are larger than for
home purchase applications. Interestingly, while there is little residual difference
between Asian and white denial rates on home purchase loan applications, the disparity
is sizable for refinance (.039) and home improvement (.054) applications --comparable

to the Hispanic/white differences.
In the remainder of this section, we focus on aggregate racial differences in denial
rates. Gross denial rate differences are expressed as the sum of components
representing differences in applicant characteristics (AC), neighborhood (TRACT),

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

market (MSA), lender (LENDER), and an unexplained residual. In presenting figures
for various applicant groups, components are averaged over all group members and
expressed as percentages (by multiplying by 100) instead of fractions. By construction,
these components must add up. Thus, for example, if 30 percent of an applicant group
were denied, then the sum of the average AC, MSA, TRACT, and LENDER
components and the average unexplained residual must equal 30 percent. Similarly, the
difference in the percentage denial rates for two groups must equal the sum of the
differences in their components.
Neighborhood, MSA, and lender effects are taken directly from the estimated
components, TRACT, MSA, and LENDER. The component reflecting each applicant's
economic characteristics, AC, is computed using the coefficients from equation (3),
assuming his or her race is white. The unexplained residual is then computed for each
applicant as the difference between the lender's action (DENIAL [I] or ACCEPT [O])
and the predicted lender action based on the sum of AC, MSA, TRACT, and LENDER.
It should be remembered that MSA, TRACT, and LENDER are normalized to have

mean zero. Since the applicant characteristics, AC, are formed assuming the applicant is
white, these normalizations imply that the unexplained residual for white applicants will
be approximately, but not exactly, zero due to nonrandom distributions of white
applicants a'cross tracts, lenders, and MSAs.

Racial Differences in Denial Rates - AU Neighborhoods
The average applicant, lender, MSA, neighborhood, and residual effects for
black, Hispanic, Asian, native American, "other" race, white, and total applicants are

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

reported in column 1 of tables 6,7,and 8. Because of the normalizations, these numbers
by themselves are not particularly meaningful; it is the differences between the racial
groups that are of interest.

As summarized in table 1, for home purchase and re£inance

loan applications, the unexplained residual makes up most of the racial differences in
percentage denial rates. The residual accounts for two-thirds of the 16.3 percentagepoint difference between black and white percentage desal rates on home purchase loan
applications, and six-tenths of the 12.4 percent difference for refinance applications.
While the Hispanic/white percentage denial rate differential is smaller (9.0 and 9.2
percentage points on home purchase and refinances, respectively), the residual still
accounts for a significant portion of the difference (four-tenths for home purchases and
slightly over half for refinances). The same is true for the other racial groups. Census
tract locations also contribute to the racial differences in percentage denial rates on
home purchase and refinance applications, but the contribution is much less than the
residual associated with the race of the applicant.
For home improvement loan applications, the picture is somewhat different.
While the residual still accounts for over a third of the difference, disparities in applicant
characteristics (including lender and MSA) account for a sizable portion of the difference
between white percentage denial rates and those for blacks and Hispanics. Moreover,
census tract location accounts for a large share of the black/white differential.
There are some other notable differences across the three types of loans. First,
racial differences in percentage denial rates are least pronounced for refinance loan
applications. Second, for black applicants, the home purchase residual is larger than the

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

refinance and home improvement residuals, while the opposite is true for Hispanic and
Asian applicants. Finally, while the Asian percentage denial rate is virtually
indistinguishable from the white percentage denial rate on home purchase applications,
there are significant and largely unexplained differences between Asian and white
percentage denial rates for the other loan products.
Racial Differences in Denial Rates bv Neighborhood Income and Racial Compositiorl
We now examine racial differences in percentage denial rates within and across
census tracts, grouped on the basis of average applicant income: high income (mean
income of all applications for loans in the tract of more than $60,000), middle income
(mean income between $40,000 and $60,000) and low income (mean income of less than
$40,000); and racial composition: primarily white (tracts with less than 10 percent
nonwhite applicants), mixed (10 to 30 percent nonwhite applicants), and primarily
minority (more than 30 percent nonwhite). Percentage denial rates by neighborhood
income and by neighborhood racial composition for black, Hispanic, Asian, and white
applicants are given in columns 2 - 10 of tables 6,7, and 8. We report the percent of the
applications, the actual percentage denial rate, the portion attributable to applicant
characteristics, MSA,lender, census tract, and the unexplained residual, for each for
black, Hispanic, Asian, native American, white, and other race applicants, in each of the
nine types of neighborhoods.
These tables reveal a remarkable persistence in the unexplained residual. While

the size of the residual varies somewhat across loan type and across tracts that differ in
mean income and racial composihon, it is always relatively large. For black applicants,

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

the unexplained residual for home purchase loans ranges from 9 to 14 percentage points
across the nine types of neighborhoods; for refinance and home improvement, the range
is only slightly lower

- 6 to 12 percentage points.

For other minority groups, there is a

comparable persistence across neighborhoods in the unexplained residual.
The tables also reveal a remarkable persistence in the census tract effects across
racial groups. For all racial groups, applications for properties in predo&tly

minority

and low-income neighborhoods have higher percentage denial rates than for those in
predominantly white and high-income neighborhoods.
While the overall impression is one of consistency, a few systematic differences
are evident. The difference between black and white percentage denial rates is lowest in
primarily minority tracts, and in all neighborhoods the unexplained residual accounts for
almost all of the difference, though there is a tendency for it to decline with
neighborhood income. For Hispanics, on the other hand, the residual difference is
slightly higher in the minority tracts and tends to increase with neighborhood income,
though these patterns are weak. We tend to focus on minority-white comparisons, but
there are also interesting differences across the minority groups. For example, in all but
one type of neighborhood (low-income-mixed tracts), our model predicts a lower
percentage denial rate for blacks than Hispanics. This lower predicted percentage denial
rate, however, is swamped by the higher residuals for blacks, and as a result the overall
percentage denial rates within each type of neighborhood are 5 to 10 percentage points
higher for black applicants.
To examine the robustness of these results, a number of other comparisons were

16

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

made. The sample was restricted to center city areas (table 9) and non-center-city areas
(table 10). The sample was restricted by lender type (tables 11, 12, and 13).
Neighborhoods were defined by the percentage of applicants that were black (table 14)
and Hispanic (table 15). Data were also disaggregated by MSA, with results presented
for the top 25 MSAs and grouped for smaller ones (tables 16, 17, and 18). In all cases,
the results support the basic findings of tables 6, 7, and 8.
Despite the apparent thoroughness of these robustness tests, there remains a
concern that the validity of each of these findings rests upon the appropriateness of the
same basic denial model, and our assumption that the form of this model is linear. To
examine this assumption, one final robustness test was employed. Observations were
grouped according to their predicted probability of denial based on AC, MSA, and

LENDER. This could be considered a nonparametric rank-ordering of observations by
risk (except for race and neighborhood). Average differences in the blacklwhite and
Hispanic/white unexplained residual and tract effects were then computed for each
predicted denial probability group and are presented in tables 19 and 20. By
construction, within each group the sum of the other predicted characteristics is the same
for blacks and whites (or Hispanics and whites), so the sum of the residual and tract
racial differences must equal the differences in racial percentage denial rates.
The linear probability model assumption implies that the differences in racial
denial rates (and the residual and neighborhood subcomponents) should be consta.nt
across risk groups. If the underlying model form were logistic or probit, then the
differences would be increasing as the denial probability rose from zero to 50 percent.

17

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

The results presented in tables 19 and 20 suggest that whereas the residual and
neighborhood group differences do rise when the denial probability increases from zero
to 10 percent, they are fairly constant above that level. This suggests that the linear
probability model specification is no less appropriate than the logistic or probit model
form.

N. Conclusions
We find a persistent difference in the denial rates of white and minority
applicants, particularly blacks. These differences remain even after lender,
neighborhood, and applicant economic characteristics (as best we can measure them with
the HMDA data) are accounted for. Moreover, we find a remarkable degree of
consistency in these conclusions across geographic markets and loan products, indicating
that the observed racial differences in denial rates are widespread and cannot be
attributed to a subset of markets or type of lender.
It is by now well known that the HMDA data do not contain enough relevant
information about the loan applications to draw any firm conclusions regarding the
reasons for these differences. We cannot determine whether these findings are
generated by a process of lender discrimination against minorities, because our residual
differences may be due to credit histories, employment histories, loan-to-value ratios,
wealth, or other factors that lenders consider in the loan evaluation process but that are
not included in the HMDA file. Because our analysis excludes these variables, we
cannot conclude that the unexplained residual unambiguously stems from differential

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

treatment based solely on the race of the applicant. There is some evidence in the

HMDA data that these variables may be correlated with race, as witnessed by the more
prevalent citation of credit history as a reason for denial for minorities (table 2). Such a
correlation could confound the estimation of the pure racial effect.
Despite this weakness of the HMDA data, our analysis does shed some light on
the reasons for observed differences in denial rates across racial groups and
neighborhoods. It has been argued that property location is an important source of
racial differences in denial rates. Because house value appreciation tends to be lower in
low-income and minority neighborhoods, these areas are considered to be more risky
from the lenders' point of view. Moreover, some lenders argue that appraisals are
harder to conduct and interpret in low-income and minority neighborhoods, because the
housing stock is generally older and more heterogeneous, and because appraisers are less
familiar with these neighborhoods.12 Our analysis indicates that property location does
contribute to racial differences in denial rates, but on average neighborhood effects are
smaller than those stemming from applicant characteristics. Moreover, when comparing
similar applicants, racial differences in denial rates still exist and are roughly the same
size within neighborhoods, regardless of the type of neighborhood.
Since there are a number of potential explanations for the racial differences we
find in our residual denial rates, further study will be necessary to pinpoint the causes.
For example, one explanation could be that factors observed by the lenders but not
contained in our data are driving the results. If so, one would expect larger residual
differences for home purchase loan denials than for refinance and home improvement

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

loans, because the latter applicants are a select group that has already received at least
one loan

- the original home purchase loan.

We find some evidence that this is the

case: for black applicants, the residual denial rate is higher for home purchase loans
than for refinances. Interestingly, this pattern does not hold for Asian and Hispanic
applicants; their residual denial rates are greater for refinances than for home purchase
loans. Moreover, for all minority groups there are sizable unexplained residuals for
>

refinance and home improvement loan applications as well as for home purchase
applications, suggesting that having once qualified for a new home loan brings little
useful information to the regressions. Exactly what kind of process could generate these
outcomes for different credit products requires more thought.
One possibly fruitful approach would be to pay more attention to the individual
lenders and their characteristics. In several previous studies (Avery, Beeson, and
Sniderman [1992, 1993b]), we demonstrate that lenders are quite heterogeneous in terms
of the propensities to attract and approve minority applicants, and that there appears to
be little consistency either within or between lenders in their actions toward minorities.
Theories regarding the operation of housing credit markets should exploit these £indings

as part of a general explanation of the process generating the data
Future studies of the relationship between race and risk outcomes would also
appear to be particularly important in order to shed light on the reasons for observed
racial differences in our residuals. If the patterns we observe are due to discrimination
by lenders, and such discrimination takes the form of a higher risk threshold for
minorities, then we would expect loans granted to black applicants to perform better

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

than those granted to whites, ceteris paribus. Given the findings of this study, such
examinations would seem very important. At the same time, we are cautious about the
power of such hypothesis tests. Several different explanations for significant racial
intercepts can be observationally equivalent, making it very difficult to claim persuasively
that any one process adequately accounts for the variations in the data. Accordingly,
careful attention to distinguishing among competing hypotheses through choice of data
and modeling strategies seems especially important.

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

ENDNOTES
1. See, for example, the Fair Housing Act of 1968 and the Equal Credit Opportunity Act

of 1975,which prohibit lenders from discriminating against individual loan applicants on the
basis of race or ethnic origin, gender, and other factors. The latter law also prohibits the
explicit use of such variables in credit screening, even if cost-related. Also, the Community
Reinvestment Act of 1977 requires that depository institutions help meet the credit needs
of their communities, including low-income and minority areas, in a manner consistent .with
safe and sound banking.
2. Canner (1981), Avery and Buynak (1981), Avery and Canner (1983), and .Bradbury, Case,
and Dunham (1989) contrast the differences in mortgage credit originations between
predominantly white and predominantly minority neighborhoods in various MSAs. These
studies use either pre-1990 HMDA data or lien title data to infer from the neighborhoods'
characteristics whether mortgage lenders treat neighborhoods differently depending on their
racial composition Calem (1992) contrasts the experiences of individual lenders
participating in a Philadelphia area mortgage-lending plan with those who did not
participate. His paper does document the existence of lender differences in the penetration
of minority communities, but the primary focus is on the characteristics of the voluntary
mortgage plan operated by a group of lenders. Avery (1989) notes the differences between
studies based on lending in a neighborhood and the lending procedures adopted by
individual lenders.
3. Two exceptions are King (1980) and Schafer and Ladd (1981), which find little evidence
of neighborhood redlining but some evidence of higher denial rates for black and Hispanic
applicants, after controlling for all available information on other factors, such as income
and credit history, relevant to the lending decision. While quite informative, these studies
are limited in their geographic coverage and in the number and types of lenders surveyed.
In addition, there have been several studies that use household-level data without
neighborhood effects. Canner, Gabriel, and Wooley (1991), Gabriel and Rosenthal(1991),
and Duca and Rosenthal (1992) study racial aspects of credit rationing and market
performance by using data from the Survey of Consumer Finances, which comprises
information collected from a sample of households. These studies attempt to infer from the
households' experiences and demographic characteristics whether lenders treat people
differently as a result of their racial status. Canner and,Luckett (1991) do not consider race,
but do discuss factors associated with consumer and mortgage debt payment problems.
4. The model was actually estimated using deviations about the means, which is
computationally equivalent to adding intercepts. For the new purchase sample, the
1,984,688 observations were located in 607,631 unique combinations of the 40,008 tracts and
20,695 lenders in the sample spread across 340 MSAs; #us, the average tract had about 15
lenders, each of whom served about 30 tracts per MSA. For the refinancing sample, the
716,595 observations were located in 326,535 unique combinations of tracts and lenders.
For the home improvement loan sample, the 787,951 observations were located in 267,158
unique combinations of tract and lender.

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

5. Mortgage banks are considered to have an office in an MSA if they take five or more
mortgage applications there. There is some evidence that a significant portion of
applications to mortgage bankers, perhaps as high as 30 percent, may not have been
reported in HMDA for 1990 and 1991 because firms fell below the $10 million asset
requirement. This may be particularly true for firms serving primarily as originators, selling
loans in the secondary market. In November 1991, the Federal Reserve Board tightened
the reporting requirements for mortgage banks, which should increase coverage.
6. Institutions with assets of less than $30 million were not required to report race, income,
and gender for loan applicants. In addition, the HMDA f i g s contained many errors and
inconsistencies even after extensive editing by the receiving agencies. We dealt with missing
and implausible data using a "hot deck" imputation procedure similar to that used by the
U.S. Census Bureau. Applications with missing or implausible data were statistically
matched to applications for the same type of loan in the same census tract that came closest
to them in reported characteristics (race, loan action, income, and loan amount). Missing
values were filled in using the variable value of the matched observation. Overall, income
was imputed for 4.9 percent, loan amount for 1.5 percent, gender for 4.0 percent, and race
for 5.6 percent of the study sample applications.
7. Applications were omitted from our sample for the following reasons: loans purchased
from other institutions (1,137,741) because they did not require an action by the reporting
lender; applications for properties outside the MSAs in which the lender had an office
(1,523,429 loans) because of inconsistent reporting requirements; applications for
multifamily homes and those that never reached the stage of lender action because they
were either withdrawn by the applicant or closed for incompleteness (444,684).
8. The 8,745 financial institutions filing 1990 HMDA reports that had at least one loan in
the study sample operated in an average of 2.7 MSAs. This translated into 23,248 study
lenders when lenders were defined at the MSA level.
9. Household income of sample applicants may be higher than this figure, since the
applicant's income used for mortgage qualification may not reflect all of the income received
by the household.
10. The percent Hispanic in the HMDA sample is slightly higher than the overall U.S.
population, due in part to the inclusion of Puerto Rico, and the percent black is slightly
lower. U.S. figures are taken from the whole 1990 Census, which may differ somewhat from
the coverage of the study sample, in that rural areas we included.
11. The reported standard errors in tables 3, 4, and 5 are those from a standard regression
program. These may be biased due to heteroskedasticity stemming from the fact that the
underlying model is a linear probability model.

12. See Lang and Nakamura (1993) for more discussion on this point.

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

REFERENCES

Avery, Robert B. 1989. "Making Judgments about Mortgage Lending Patterns."
hkonomic Commenfary, Federal Reserve Bank of Cleveland (December 15).
Avery, Robert B., Patricia E. Beeson, and Mark S. Sniderman. 1992. "Cross-Lender
Variation in Home Mortgage Lending." Working Paper 9219, Federal Reserve Bank of
Cleveland (December).
1993a. "Home M~rtgageLending by the Numbers." Economic
Cornmenfly, Federal Reserve Bank of Cleveland (February 15).
1993b. "Lender Consistency in Housing Credit Markets." Proceedings,
Conference on Bank Smture, Federal Reserve Bank of Chicago, forthcoming.
Avery, Robert B, and Thomas M. Buynak. 1981. "Mortgage Redlining: Some New
Evidence." Economic Review, Federal Reserve Bank of Cleveland (Summer), pp. 18-32.
Avery, Robert B., and Glenn B. Canner. 1983. "Mortgage Redlining: A Multicity CrossSection Analysis." Unpublished Working Paper, Board of Governors of the Federal
Reserve System, Washington, D.C.
Bradbury, Katharine, Karl E. Case, and Constance R. Dunham. 1989. "Geographic
Patterns of Mortgage Lending in Boston, 1982-1987." New England Economic Review
(September/October), pp. 3-30.
Calem, Paul S. 1992. 'The Delaware Valley Mortgage Plan: An Analysis Using HMDA
Data." Working Paper 92-3, Federal Reserve Bank of Philadelphia (February).
Canner, Glenn B. 1981. "Redlining and Mortgage Lending Patterns." In Research in
Urban Economics, edited by J. Vernon Henderson, Greenwich, C P JAI Press, pp. 67101.
Canner, Glenn B., Stuart A. Gabriel, and J. Michael Wooley. 1991. "Race, Default Risk,
and Mortgage Lending: A Study of the FHA and Conventional Loan Markets." S o u t h
Economic Jozvnalj vol. 58 (no. I), pp. 249-262.
Canner, Glenn B., and Charles A. Luckett. 1991. "Payment of Household Debts."
Federal &?Se?VeBulletin, vol. 77 (April), pp. 218-229.
Canner, Glenn B., and Delores S. Smith. 1991. "Home Mortgage Disclosure Act:
Expanded Data on Residential Lending." Federal Reserve Bulletin, vol. 77 (November),
pp. 859-881.

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

1992. "Expanded HMDA Data on Residential Lending: One Year Later."
Federal Resene Bulletin, vol. 78 (November), pp. 801-824.

Duca, John V., and Stuart S. Rosenthal. 1992. "Borrowing Constraints, Household Debt,
and Racial Didmination in Loan Markets." Unpublished manuscript.
Gabriel, Stuart A,,and Stuart S. Rosenthal. 1991. "Credit Rationing, Race, and the
Mortgage Market." J o d of Urban Economics, vol. 29 (May), pp. 371-379.

ICF, Incorporated. 1991. 'The Secondary Market and Community Lending Through
Lenders' Eyes." Paper prepared for the Federal Home Loan Mortgage Corporation
(February).

King,A. Thomas. 1980. "Discrimination in Mortgage Lending: A Study of Three Cities."
Working Paper No. 91, Office of Policy and Economic Research, Federal Home Loan
Bank Board (February).

Lang, William W., and Leonard I. Nakamura. 1993. "A Model of Redlining." J o d of
Urban Economics, vol. 33 (March), pp. 223-234.
Munnell, Alicia H., Lynne E. Browne, James McEnearney, and Geoffrey M. B. Tootell.
1992. "Mortgage Lending in Boston: Interpreting HMDA Data" Working Paper Series
92-7, Federal Reserve Bank of Boston (October).
Shafer, Robert, and Helen F. Ladd. 1981. Discrimination in Mortgage Lending.
Cambridge, MA: MIT Press.

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 1:

Characteristicsof Mortgage Applications. National Sample, 1990 HMDA

Home

Percent Percent Denial
Sample Loan$ Rate

Purchase-

Pacent Percent Denial
Sample Loans Rate

Percent Percent Denial
Sample Loan$ Rate

Race ofApp1icanr

Native American
Asian (or Pacifrc Islander)
Black
w
c
white
Other

0.6% 0.6%
4.6
6.8
6.2
4.8
6.6
6.4
81.4
80.5
0.7
1.0

19.3%
14.4
29.4
22.1
13.1
19.8

0.6%
4.9
5.1
7.7
80.9
0.7

0.6%
7.2
3.9
73
79.9
1.0

21.2%
21.3
28.8
25.6
16.4
26.8

0.9%
2.5
10.3
5.7
79.9
0.8

1.0%
5.4
5.9
5.4
81.3
1.0

22.7%
27.7
43.4
35.4
20.3
35.4

28.4
69.4
2.2

24.1
73.4
2.5

17.3
13.8
15.6

24.8
73.2
2.0

23.8
73.9
23

21.0
17.1
19.4

33.5
64.9
1.6

26.3
71.6
2.1

29.8
20.8
21.1

64.0
4.3
2.0
1.2
16.9
115

682
4.2
2.3
1.2
15.6
8.5

13.4
18.6
16.4
18.1
17.9
16.5

67.7
4.9
1.6
0.9
14.7
10.1

692
42
20
0.8
15.7
8.1

16.8
21.4
19.6
20.2
22.0
19.6

58.0
6.9
0.8
0.8
195
14.0

65.8
6.1
1.0
0.8
16.3
9.9

19.7
28.6
27.8
28.1
29.5
30.1

93.6

945

14.9.

90.9

91.5

18.1

97.2

96.1

23.8

Race of Co-applicant

No Co-applicant
Same Race as Applicant
Different Race than Applicant
Gender

Male Applicant. Female Co-applicant
Female Applican~Male Co-applicant
Male Applicant and Co-applicant
Female Applicant and &applicant
Single Male Applicant
Single Female Applicant
Owner-Occupied
Loan Type

Conventional
FHA
VA
FmHA
Lcnder Action
Loan Denied

Loan Accepted and Withdrawn
Loan Originated
Loan Kept by Originator (% of originations)
Loan Sold to FNMA (% of originations)
Loan Sold to GNMA (% of originations)
Loan Sold to FHLMC (% of originations)
Loan Sold Elsewhere (8of originations)
Reasonsfor Denial (of b a n s ~ e n i e d ) '

No Reason Given
Debt-to-Income Ratio
Employment History
Credit Hist~ry
Collateral
Insufficient Cash
Unverifiable Information
Application Incomplete
Mongage Insurance Denied
Other
Memo I f e m :

Median Income (S 1.000s)
Median Loan Request ($1.000~)
Number of Loans

Up to three reasons for denial could be given, and k w e r s were voluntary. Each category gives the percent of all denials that
gave that reason as one of the three.
SOURCZFOR ALL TABLES: Authol'~.

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 2:

Characteristicsof ~ o r t ~ Applications
a ~ e
by Race. National Sample. 1990 HMDA

Home
Black Hispanic White

Refinance
Black Hispanic White

Home lmorovement
Black Hispanic White

Gender

Two Applicants
Single Male Applicant
Single Female Applicant
Owner-Occupied

94.5

93.6

93.7

88.0

90.4

91.2

96.6

$47
$50
$71 $100

$56
$79

$27
$5

96.5 97.3

Loan Type

Conventional

FHA
VA
Lender Acrion

-

h Denied
h Accepted and Withdrawn
horiginatkl
Loan Kept by Originator (% of originations)
Loan Sold to FNMA (% of originations)
Loan Sold to GNMA (% of originations)
Loan Sold to FHLMC (96 of originations)
h Sold Elsewhere (% of originations)
Reasonsfor Denial (of Loans Denied)'

No Reason Given
Debt-to-Income Ratio
Employment History
Credit History
Collateral
Insufficient Carh
Unverifiable Information
Application Incomplete
Mortgage Insurance Denied
Other
Memo Iremr:

Median Income ($1,000s)
Median Loan Request ($1,000~)

$36
$61

$44
$85

$48
$76

$35
$11

$40
$10

Up to three reasons for denial could be giuen. and answers were voluntary. Each category gives the percent of all denials that
gave that reason as one of the three.

http://www.clevelandfed.org/Research/Workpaper/Index.cfm
Table 3: Linear Probability Model of Loan Denial (I) or Acepawe (0). Home Purchase

Parameter Estimate
Race (Dummies, 'WhiteWlsBase Group)
Black Applicant
Hispanic Applicant
Native American Applicant
A s i i Applicant
Other Race Applicant
Mixed Race, Minority Co-applicant (Dummy)
Mixed Race. Nonminority Co-applicant (Dummy)
Owner-occupied (Dummy)

.024 10"'
-0.02690"'

.00630'"

Income ($1,000'~)
Income
Income Spline at 1620.000
Income Spline at $40.000
Income Spline at $60.000
Income Spline at $80.000
Income Spline at $100,000
Income Spline at $150.000
Income Spline at $200,000
Loan Amounr ( $ I , W s )
Loan Amount
Loan Amount Spline at $20.000
Loan Amount Spline at $40,000
Loan Amount Spline at $60.000
Loan Amount Spline at $80.000
Loan Amount Spline at $100.000
Loan Amount Spline at $125,000
Loan Amount Spline at $200,000
Loon-to-Income Ratio (Dummies, LESSfhan I 5 Is Base Group)
-0.01016'"
Ratio of 1.5 to 2.0
-0.0 1168"'
Ratio of 2.0 to 2.25
Ratio of 2.25 to 2.5
-0.01 195"'
-0.00737"'
Ratio of 2.5 to 2.75
Ratio of 2.75 to 3.0
.00323
Ratio over 3.0
,05062"'
Applicant Gender (Dummies. Female Applicant, No Co-applicant Is Base Croup)
Male Applicant, Female Co-applicant
-0.01886Female Applicant. Male Co-applicant
-0.00766
Male Applicant and Co-applicant
-0.00390
Female Applicant and Co-applicant
-0.01021
Male Applicanl. No Co-applicant
.02834"'
Income. Interacted WithNo Co-applicant
Income
Income Spline at $20.000
lncome Spline at $40.000
lncome Spline at $60,000
lncome Spline at $80,000

Standard Error

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 3: (continued)

Parameter Estimate
Race and Marital Status. Interacted With VA Loan

Black Applicant
Hispanic Applicant
Native American Applicant
Asian Applicant
White Applicant
Other Race Applicant
No Co-applicant

-0.00667
-0.00866
.04929'
.01699
-0.02033
.02562
-0.00619'

Race and Marital Stafus. Interacted With F H A Loan

Black Applicant
Hispanic Applicant
Native American Applicant
Asian Applicant
White Applicant
Other Race Applicant
No Co-applicant

-0.01967
-0.04312"
.W29
-0.03294'
-0.03329'
-0.02377
-0.01230"'

Income, Interacted With VA or F H A Loan

Income
Income Spline at $20,000
Income Spline at $40,000
Income Spline at $60.000
Income Spline at $80,000
Income Spline at $100,000
Loan Amount, Interacted With VA or F H A Loan

Loan Amount
Loan Amount Spline at $20,000
Loan Amount Spline at $40.000
Loan Amount Spline at $60.000
Loan Amount Spline at $80.000
Loan Amount Spline at $100,000

.00366"'
-0.00256'"
-0.0023 '"1
.OOO66'
-0.00038
.00052

Loan-to-Income Ratio. Interacted With VA or F H A Loan

Ratio of 1.5 to 2.0
Ratio of 2.0 to 2.25
Ratio of 2.25 to 2.5
Ratio of 2.5 to 2.75
Ratio of 2.75 to 3.0
Ratio over 3.0

-0.00333
-0.0051 1
-0.00612
.OOO29
-0.00449
-0.00681

Memo Item:

Number of Observations
Mean Denial Rate in Regression Sample
Number of TracVInstitution Dummies
R squared (Including Tractllnstitution Dummies)
R squared (Variation around Tracthtitution Means)
Significantat the 5 percent level.
"Significant at the 1 percent level.
"'Signific,ant at the 0.1 percent level.

1984.688
.I48
607.631
A56 '
.022

Standard Error

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 4: Linear Probability Model of Loan Denial (1) a A c q m n c e (0). R e f m c e

Parameter Estimate

Race (Dummies. 'While" Is Base Group)

Black Applicant
Hispanic Applicant
Native American Applicant
A s i i Applicant
Other Race Applicant
Mixed Race. Minority Co-applicant (Dummy)
Mixed Raix. Nonminority Co-applicant (Dummy)

.

.00576

-0.02336"

Ownex-oczupied (Dummy)
VA Loan (Dummy)
Income ($1 IXX)'s)

Income
Income Spline at $20,000
Income Spline at S40.000
Income Spline at W.000
Income Spline at $80,000
Income Spline at $100,000
Income Spline at $150,000
Income Spline at WO.000
Loan Amounr ($1 W's)

Loan Amount
Loan Amount Spline at $20,000
Loan Amount Spline at $40,000
Loan Amount Spline at W.000
Loan Amount Spline at $80,000
Loan Amount Spline at $100.000
Loan Amount Spline at $125.000
Loan Amount Spline at O2Ml.000
Loan-&-Income Ratio (Dummies. Less than 1 5 Is Base Group)
-0.00218
Ratio of 1.5 to 2.0
.0045 1
Ratio of 2.0 to 2.25

Ratio of 2.25 to 2.5
Ratio of 2.5 to 2.75
Ratio of 2.75 to 3.0
Ratio over 3.0

.00700'
.01506"'
.02567"'
.08614"'

Applicant Gender (Dumtdes, Female Applicant. No Co-applicanfIs Base Group)

Male Applicant, Female Co-applicant
Female Applicant, Male Co-applicant
Male Applicant and Co-applicant
Female Applicant and Cbapplicant
Male Applicant, No Co-applicant

-0.09269"'
-0.08497"'
-0.06650"'
4.08148"'

.024n"'

Standard Error

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 4: (continued)

Parameter Esrimate
Income. Interacted With N o Co-applicant

Income
Income Spline at $20.000
Income Spline at $40.000
Income Spline at %60,000
Income Spline at $80,000
Income Spline at $100.000
Interacted With VA or FHA Loan

Black Applicant
Hispanic Applicant
Native American Applicant
Mi Applicant
White Applicant
Other Race Applicant
No Co-applicant
Income
Loan Amount
Memo I t e m :

Number of Observations
Mean Denial Rate in Regression Sample
Number of TractlInstitution Dummies
R squared (Including Tracthstitution Dummies)
R squared (Variation around Tracthstitution Means)

' Significantat the 5 percent level.
Significant at the 1 percent level.
"'Significant at the 0.1 percent level.
"

Standard Error

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 5: Linear Robability Model of Loan Denial (I) or Acceptance (0). Home Impmvement

Parametex Estimate
Race (Dummies. 'White" Is Base Group)

Black Applicant
Hispanic Applicant
Native American Applicant
A s i i Applicant
Other Race Applicant
Mixed Race. Minority Co-applicant (Dummy)
Mixed Race. Nonminority Co-applicant (Dummy)

.00107
-0.04042"'

Owner-occupied (Dummy)
VA Loan W ~ Y )
Income ($1,000'~)

Income
Income Spline at $20,000
Income Spline at $40,000
Income Spline at $60,000
Income Spline at $80,000
Income Spline at $ 100,000
Income Spline at $150.000
Income Spline at $200,000
Loan Amount ($1,0005)

Loan Amount
Loan Amount Spline at $20.000
Loan Amount Spline at $40.000
Loan Amount Spline at IF60.000
Loan Amount Spline at $80,000
Loan Amount Spline at $100.000
Loan Amount Spline at $125,000
Loan Amount Spline at $200.000
Loan-to-IncomeRatio (Dummies. Less than 15 Is Base Group)

Ratio of 1.5 to 2.0
Ratio of 2.0 to 2.25
Ratio of 2.25 to 2.5
Ratio of 2.5 to 2.75
Ratio of 2.75 to 3.0
Ratio over 3.0

-02051"'
.00433"'
.02663'
.05256"'
.08344"'
.04087"'

Applicant Gender (Dummies,Fema!e Applicant, No Co-applicanl Is Base Group)

Male Applicant, Female Co-applicant
Female Applicant. Male Co-applicant
Male Applicant and Co-applicant
Female Applicant and Co-applicant
Male Applicant, No Co-applicant

-0.10888"'
-0.07293"'
-0.04480"'
-0.07792"'
.03575"'

Standard Error

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 5: (continued)

Parameter Eslimate
Income. Interacted With N o Co-applicant

Income
Income Spline at $20.000
Income Spline at $40,000
Income Spline at $60,000
Income Spline at $80.000
Income Spline at $100.000
Interacted With VA or FHA Loan

Black Applicant
Hispanic Applicant
Native American Applicant
A s i i Applicant
White Applicant
Other Race Applicant
No Co-applicant
Income
Loan Amount
Memo I t e m :

Number of Observations
Mean Denial Rate in Regression Sample
Number of TracUInstitutionDummies
R squared (Including TracUInstitution Dwnmies)
R squared (Variation around TractlInstitution Means)
Signir~cantat the 5 percent level.
Significant at the 1 percent level.
"'Significant at the 0.1 percent level.

"

A73

.on

Standard F m r

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 6: Difkwce in Average Percentage Denial Rates Attributable to Various Sources. Home Purchase Loans, by
Neighborhood and Race, 1990 HMDA

Total

-7

w-

-MiddleIncome'

White4Mixedsndty6

WhisMix%in3

Black Applicants

Percent of Black
Actual Denial Rate
Applicant Economic
Characteristics
MSA Effect
'
O v d Lender Effect
Ccnsus Tract Effect
Residual'
Hispanic Applicants

100.0
22.1

3.4
17.4

13.9
19.1

26.5
20.0

3.3
19.6

10.1
205

22.0
22.1

1.3
24.6

3.7
27.8

15.8
28.9

14.8
2.0
0.1
1.4
3.7

12.6
0.8
1.6
-1.5
3.8

13.0
1.9
0.6
-0.6
4.2

13.1
3.0
-0.9
1.4
3.4

14.0
-0.4
0.8
-0.4
5.6

14.5
1.3
0.6
0.6
3.4

14.4
23
-0.8
28
3.4

17.1
-0.3
1.0
0.8
6.0

18.4
0.6
1.5
1.6
5.8

19.6
1.7
1.4
3.0
3.2

100.0
14.4

6.3
10.3

25.8
13.5

36.6
14.5

4.7
11.9

8.5
13.9

12.4
17.7

1.4
13.1

2.1
17.3

22
20.9

13.0
0.6
-0.2
-0.0
1.O

121
-0.4
-0.2
-1.9
0.7

12.6
0.9
0.2
-1.3
1.2

12.7
1.0
-0.4
0.4
0.8

12.8
-1.3
-0.6
-0.5
1.6

13.1
-0.1
-0.7
0.0
1.6

13.6
0.9
-0.0
2.2
1.0

16.3
-1.3
-1.4
0.0
-0.5

16.4
-0.2
-0.4
1.2
0.3

16.7
0.2
-0.3
2.1
2.2

Percent of Native Americans 100.0
Actual Denial Rate
19.3
Applicant Economic
Characteristics
14.4
USA E t k t
0.3
Overall Lender Effect
1.1
0.1
Census Tract Effect
3.4
Residual7

13.1
14.9

20.6
14.6

10.0
19.3

16.2
17.3

123
19.3

7.2
24.6

8.8
22.9

7.1
26.6

4.8
33.3

12.9
-0.4
0.3
-1.7
3.8

12.5
1.1
1.0
-1.2

12.8
1.7
1.3
0.5
3.1

14.6
-1.1
0.6
-0.6
3.8

14.4
0.1
0.7
-0.0
4.1

14.1
1.4
-0.2
2.8
6.5

17.8
-0.8
1.3
1.0
3.6

18.0
0.6
2.8
2.6
2.6

18.9
0.8
4.3
4.2
4.9

Percent of Hispanics
Actual Denial Rate
Applicant Economic
Charactaistics
MSA Effect
O v d Lender Effect
Census Tract Effect
Residual7
Asian Applicants

Percent of Asians
Actual Denial Rate
Applicant Economic
Chamctaistics
MSA Effect
O v d Lender Effect
Census Tract Effect
Residual'
Native American Applicants

1.1

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 6:(continued)

Total

Other Race Applicants

Percent of Other Race
Actual Denial Rate
Applicant Economic
Characteristics
M A Effect
Overall Lender Effect
Census Tract Effect
Residual7

100.0%
19.8

10.9% 26.8%
16.3 18.1

15.8%
22.4

9.2% 11.9% 14.8%
14.9
18.1 24.3

3.0% 3.1% 4.5%
21.3 24.1 25.1

14.0
1.1
0.7
0.2
3.8

12.5
-0.1
1.2
-1.6
4.2

12.9
1.3
1.1
-1.3
4.0

13.4
2.2
0.8
1.0
4.9

13.3
-1.0
0.0
-0.6
3.2

13.8
0.3
0.6
0.2
3.1

15.7
2.6
0.3
2.5
3.1

16.3
-0.7
0.3
0.8
4.6

17.3
-0.1
0.6
2.2
4.1

18.6
1.4
-0.4
3.3
2.2

100.0
13.1

18.9
9.5

16.2
12.2

3.9
15.6

26.3
11.0

11.1
13.4

27
18.0

13.5
17.0

5.2
20.0

2.1
23.7

13.6
-0.2
-0.0
-0.3
-0.0

12.0
-0.4
-0.4
-1.7
-0.0

12.4
-0.1
-1.2
-0.0

12.6
1.9
0.0
0.6
0.5

13.3
-1.3
-0.3
-0.6
-0.1

13.5
0.2
-0.4
0.1
-0.1

13.5
1.0
0.5
2.2
0.7

16.6 17.1
-0.9
0.3
0.6
12
0.8
1.7
-0.01 -0.4

17.3
0.7
1.3
3.9
0.5

100.0
14.8

16.2
9.9

16.1
13.1

7.4
17.4

22.3
11.3

10.9
14.8

5.7
21.2

11.4
17.2

5.2
22.0

4.8
28.7

13.8
0.0
0.0
0.0
1.0

12.1
-0.4
-0.3
-1.7
02

12.5
1.1
0.0
-1.1
0.7

12.8
1.9
-0.3
0.9
2.0

13.3
-1.3
-0.3
-0.6
0.2

13.7
0.3
-0.3
0.2
1.0

14.0
1.1
-0.1
25
3.7

16.6
-0.9
0.5
0.8
0.2

17.3
0.4
1.3
1.7
1.4

18.1
0.5
0.9
4.1
5.2

White Applicants

Percent of Whites
Actual Denial Rate
Applicant Economic
Charactexistics
M A Effect
Overall Lender Effect
Census Tract Effect
Residual'

1

Total Applicants

Percent of Applicants
Actual Denial Rate
Applicant Economic
Characteristics
M A Effect
Overall Lender Effect
Census Tract Effect
Residual7

Census tracts with mean applicant income of more. than $60.000.
'Census tracts with mean applicant income greater than $40.000 and less than or equal to $60,000.
Census tracts with mean applicant income of $40.000or less.
Census aacts with less than 10percent minority applicants (native Americans. A s i i . Blacks. Hispanics. or other).
Census tracts with 10 percent or more. and 30 percent or less applications from minority applicants.
Census tram with more lhan 30 percent of all loan applications from minority applicants.
'The residual is d e f d as the average difference between the actual denial rate and the sum of the economic. MSA, tract. and lender
effects.

'
'

'

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 7: Diffeenee in Average Percenrage Denial Rates Auributable to Various Sources. R e f m c c Loans.by Neighbohood
and Race. 1990 HMDA

REFINANCE
Black Applicants

Percent of Blacks
Actual Denial Rate
Applicant Economic
Characteristics
MSA Effect
Overall Lender Effect
Census Tract Effect
~esidual~

100.0%
28.8
18.0
0.1
-0.4
3.4
7.6

Hispanic Applicants

P e m t of Hispanics
Actual Denial Rate
Applicant Economic
Characteristics
MSA Effect
Overall Lender Effect
Census Tract Effect
Residual7

100.0
25.6
17.9
1.4
-0.3
1.6
4.9

Asian Applicants

Percent of Asians
Actual Denial Rate
Applicant Economic
Charactexistics
MSA Effect
Overall Lender Effect
Census Tract Effect
Residual7

100.0
21.3
18.3
-1.0
-0.0
0.2
39

Native American Applicants

Percent of Native American. 100.0
Actual Denial Rate
21.2
Applicant Economic
17.8
Characteristics
MSA Effect
0.4
Overall Lender Effect
0.0
Census Tract Effect
0.3
Residual'
2.7

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 7: (continued)

Other Race Applicants
Percent of Other Race
Actual Denial Rate
Applicant Economic
Characteristics
MSA Effect
Overall Lender Effect
Census Tract Effect
Residual7
White Applicants
Percent of Whites
Actual Denial Rate
Applicant Economic
Charclcteristics
MSA Effect
Overall Lender Effect
Census Tract Effect
Residual7
Total Applicants
Percent of Applicants
Actual Denial Rate
Applicant Economic
Characteristics
MSA Effect
Overall Lender Effect
Census Tract Effect
Residual7

' ' '' 'See notes for table 6.

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 8: Difference in Average Percentage Denial Rates Atmbutabk to Variw Sources. Home Improvement Lorn. by Neighborhood
and Race. 1990 HMDA

Total

Black Applicants

Percent of B k k s
Actual Denial Rate
Applicant Economic
Chalacreristics
MSA Effect
Overall Lender Effect
Census Tract Effect
Residual'
Hispanic Appliconts

Percent of Hispanics
Actual Denial Rate
Applicant Economic
characteristics
MSA Effect
Overall Lender Effect
Census Tract Effect
~esidual'
Asion Appliconts

Percent of Asians
Actual Denial Rate
Applicant Economic
Characteristics
MSA Effect
Overall Lender Effect
Census Tract Effect
~esidual'
Notive American Applicants

Percent of Native Americans
Actual Denial Rate
Applicant Economic
characteristics
MSA Effect
Overall Lender Effect
Census Tract Effect
~esidual'

Middle-

Low-

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 8: (continued)

Other Race Applicants

Percent of Other Race
Actual Denial Rate
Applicant Economic
Characteristics
MSA Effect.
Overall Lender Effect
Census Tract Effect
~esidual'
White Applicants
Percent of Whites

Actual Denial Rate
Applicant Economic
Characteristics
MSA Effect
Overall Lender Effect
Census Tract Effect
Residual7
Total Applicants

Percent of Appicants
Actual Denial Rate
Applicant Economic
Characteristics
MSA Effect
Overall Lender Effect
Census Tract Effect
Residual7

''

See notes for table 6.

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 9: Differencein Average Percentage Denial Rates Attributable to Various Sourn, Center City, by Neighborhood
and Race. 1990 HMDA

Total

Middle-

%
7
-

White4h4ixedJMinoritf

Black Applicants

m n t of Blacks
Actual Denial Rate
Census Tract Effect
~esidual'

3.3% 8.7% 20.7%
21.0
28.1
29.5
3.3
-1.2
0.9
10.1
14.0
12.3

100.0%
31.2
3.1
13.4

2.1% 4.7% 7.8%
26.2
28.3 28.8
3.3
1.9
-0.2
12.5
9.7
11.0

100.0
23.8
2.2
3.8

2.3
17.6
-1.8
3.9

10.1
21.0
-0.5
5.4

215
21.8
28
3.4

23
202
-0.8
6.1

8.9
20.7
0.6
3.8

24.7
23.4
3.1
3.6

100.0
13.9
-0.1
-0.0

15.3
9.8
-1.9
-0.1

15.6
13.1
-1.0
-0.2

4.7
16.6
1.5
0.3

212
10.6
1.1
-0.1

125
135
02
-0.1

4.2
18.2
2.7
0.5

100.0
29.6
4.2
7.5

1.5
28.7
-1.8
11.6

6.9
29.7
-0.9
11.4

255
25.6
3.3
6.2

1.7
24.6
-0.3
8.5

4.1
35.8
2.4
13.6

29.7
26.9
4.4
6.0

100.0
26.1
2.6
4.8

2.0
25.8
-2.6
6.2

15.4
22.9
-1.1
4.0

41.0
25.2
3.5
4.4

1.0
31.3
-0.3
8.6

4.3
29.9
1.4
6.2

24.5
26.8
2.9
5.4

100.0
17.5
-0.1
-0.1

16.3
14.8
-2.9
-0.1

25.4
18.6
-1.4
-0.1

10.1
20.3
1.9
0.1

16.5
12.9
-1.2
-0.0

9.1
18.1
0.9
-0.4

5.1
21.7
3.5
0.3

100.0%
45.1
7.5
7.6

0.9% 2.4% 5.0%
34.5
38.6
31.5
1.1
4.4
-3.2
6.8
9.4
12.1

100.0
38.6
2.5
6.5

1.6
27.8
-4.3
7.7

8.7
31.6
-1.9
7.5

15.5
35.8
3.4
5.9

1.8
30.5
-1.6
6.7

7.3
34.7
' -0.4
, 7.1

20.9
39.5
27
7.4

100.0
22.7
0.5
0.0

. 11.3

127
21.2
-1.9
-0.0

4.4
27.3
2.5
0.6

19.4
15.8
-1.9
-0.1

10.5
23.2
1.0
-0.3

4.5
32.0
4.3
0.2

Hispanic Applicants

Percent of Hispanics
Actual Denial Rate
Census Tract Effect
~esidual~
White Applicants

Percent of Whites
Actual Denial Rate
Census Tract Effect
Residual7

Black Applicants

Percent of Blacks
Actual Denial Rate
Census Tract Effect
Residual7
Hispanic Applicants

Percent of Hispanics
Actual Denial Rate
Census Tract Effect
~esidud
White Applicants

Percent of Whites
Actual Denial Rate
Census Tract Effect
Residual7

Black Applicants

Percent of Blacks
Actual Denial Rate
Census Tract Effect
Residual7

1.5% 3.8% 14.0%
31.4 35.3 42.4
5.8
-1.2 , 1.0
6.9
11.4 10.5

Hispanic Applicanrs

Percent of Hispanics
Actual Denial Rate
Census Tract Effect
Residual7
While Applicants

Percent of Whites
Actual Denial Rate
Census Tract Effect
Residud

15.6
-3.5
-0.1

'

LowIncome'

White4MixedJMinority''

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 10: Difference in Avenge Percentage Denid Rates Attributable to Various Sources, Non-Center City, by Neighborhood
and Rqce. 1990 HMDA

Black Applicants
Percent of Blacks
Actual Denial Rate
Census Tract Effect
Residual7
Hispanic Applicants
Percent of Hispanics
Actual Denial Rate
Census Tract Effect
~esidual'
White Applicants
Percent of Whites
Actual Denial Rate
Census Tract Effect
Residual7

REFINANCE
Black Applicants
Percent of Blacks
Actual Denial Rate
Census Tract Effect
Residual7
Hispanic Applicants
Percent of Hispanics
Actual Denial Rate
Census Tract Effect
Residual7
White Applicants
Percent of Whites
Actual Denial Rate
Census Tract Effect
~esidual~

Black Applicants
Percent of Blacks
Actual Denial Rate
Census Tract Effect
Residual7
Hispanic Applicants
Percent of Hispanics
Actual Denial Rate.
Census Tract Effect
~esidual~
White Applicants
Percent of Whites
Achlal I k ~ Rate
d
Census Tract Effect
~esidual~

' '''

See notes for table 6.

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 11:

Difference in Average Percentage Denial Rates Attributable to Various Sources, Commercial Banks,
by Neighborhood and Race, 1990 HMDA

Low-

Middle-

Total

White'Mu~d'Minority~

WhidMixedsMinority6

5.2% 10.9% 21.8%
28.0
28.7
29.1
0.6
-0.6
2.6
126
12.4
10.0

3.1% 9.2% 33.9%
35.0
35.3
35.5
4.7
0.8
2.1
14.6
12.5
11.5

Black Applicanu

Percent of Blacks
Actual Denial Rate
Census Tract Effect
Residual7
Hispanic Applicanls

,

Percent of Hispanics
Actual Denial Rate
Census Tract Effect
Residual7
White Applicants

Percent of Whites
Actual Denial Rate
Census Tract Effect
Residual7

REFINANCE
Black Applicants

Percent of Blacks
Actual Denial Rate
Census Tract Effect
Residual7
Hispanic Applicants

Percent of Hispanics
Actual Denial Rate
Census Tract Effect
Residual7
White Applicants

Percent of Whites
Actual Denial Rate
Census Tract Effect
Residual7

Black Applicants

Percent of Blacks
Actual Denial Rate
Census Tract Effect
Residual'
Hispanic Applicants

Percent of Hispanics
Actual Denial Rate
Census Tract Effect
Residual7
White Applicants
Percent of Whites

Actual Denial Rate
Census Tract Effect
Residual7

,

100.0%
31.8
2.6
11.4

3.2% 6.1% 6.6%
33.3
23.6
26.3
3.1
-0.5
-1.7
10.9
10.9
10.9

.

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 12: Difference in Average Percentage Denial Rates Attributable to Various Sources. M
by Neighborhood and Race, 1990 HMDA

Tolal

Black Applicants

Percent of Blacks
Actual Denial Rate
Census Tract Effect
Residual7
Hispanic Applicants

Percent of Hispanics
Actual Denial Rate
Census Tract Effect
Residual7
White Applicants

Percent of Whites
Actual Denial Rate
Census Tract Effect
Residual7

Black Applicants

Percent of Blacks
Actual Denial Rate
Census Tract Effect
Residual'
Hispanic Applicants

Percent of Hispanics
Actual Denial Rate
Census Tract Effect
~esidual'
White Applicants

Percent of Whites
Actual Denial Rate
Census Tract Effect
Residual7

Black Applicants

Percent of Blacks
Actual Denial Rate
Census Tract Effect
Residual7
Hispanic Applicants

Percent of Hispanics
Actual Denial Rate
Census Tract Effect
Residual'
White Applicants

Percent of Whites
Actual M a l Rate
Census Tract Effect
Residual7

' ''

See notes for table 6.

Hieh I n c o m L
Middlfdmmd~ h i t e ~ M i x e d ~ ~ i n & t ~White4Mixed5Minority6
~

t Institutions,

LowWhite4Mixed5@ority6

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 13: Diffmnce in Average Percentage Denial Rates Ataibutable to Various Sources. Mongage Banks.
by Neighborhood and Race, 1990 HMDA

Total

Black Applicants

~ e k e nof
t Blacks
Acmal Denial Rate
Gms Tract Effect
~esidual~
Hispanic Applicants

Percent of Hispanics
Actual Denial Rate
Census Tract Effect
Residual7
White Applicants
Percent of Whites

Actual Denial Rate
Census Tract Effect
Residual7

REFINANCE
Black Applicants

Percent of Blacks
Actual Denial Rate
Census Tract Effect
Residual'
Hispanic Applicants

Percent of Hispanics
Actual Denial Rate
Census Tract Effect
Residual7
White Applicants

Percent of Whites
Actual Denial Rate
Census Tract Effect
Residual7

Black Applicants

Percent of Blacks
Actual Denial Rate
Census Tract Effect
Residual7
Hispanic Applicants

Percent of Hispanics
Actual Denial Rate
Census Tract Effect
Residual7
White Applicanfs

Percent of Whites
Actual Denial Rate
Census Tract Effect
~esidual~

-

' '

-

See notes for table 6.

-L4!dudIncome'

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 14: Difference in Average PercentageDenial Rates. Neighborhoods Soned by Percentage Black. 1990 HMDA

Total

Black Applicants
Percent of Blacks
Actual Denial Rate
Census Tract Effect
Residual7
Hispanic Applicants
Percent of Hispanics
Actual Denial Rate
Census Tract Effect
Residual7
White Applicants
Percent of Whites
Actual Denial Rate
Census Tract Effect
~esidual~

REFINANCE
Black Applicants
Percent of Blacks
Actual Denial Rate
Census Tract Effect
Residual7
Hispanic Applicants
Percent of Hispanics
Actual Denial Rate
Census Tract Effect
Residual7
White Applicants
Percent of Whites
Actual Denial Rate
Census Tract Effect
Residual7

m

Low-

W h i z %ix?Bl$~

White' Mixeds Black6

I

le

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 14: (continued)

Lowlncomd

Total

White' Mixed' Black6

Black Applicants

Percent of Blacks
Actual Denial Rate
Census Tract Effect
Residual7

100.0%
43.4
6.3
8.0

3.6% 4.8% 3.8%
35.2
38.5
29.1
-0.2
4.0
-2.5
7.9
8.9
6.4

3.2% 8.4% 13.8%
32.4
36.7
44.1
-1.5
1.5
6.8
10.1
10.2
7.1

2.1% 8.2% 52.1%
35.2
38.5 48.1
0.3
3.6
9.5
9.4
9.8
7.4

Hispanic Applicanis

Percent of Hispanics
Actual Denial Rate
Census Tract Effect
~esidu&
White Applicanfs

Percent of Whites
Actual Denial Rate
Census Tract Effect
~esiduap

' Census sacs with mean applicant income of more than $60,000.
'Census aacts with mean a~ulicantincome greater than $40,000 and less than or equal to $60,000.
)Census rracts with mean applicant income i f $40.000 or less.
Census hacts with less Ihan 5 percent black applicants.
'Census mas with 5 percent or more and 25 percent or less applications from bhck applicants.
Census hacts with more than 25 percent of all loan applications from black applicants.
'The residual is defmed as the average difference between the actual denial rate and the sum of the economic. USA, hact, and lender
effects.

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 15: Difference in Average Percentage Denial Rates. Neighborhoods Soned by Percentage Hispanic. 1990 HMDA

Total
WhidMixed'Hi@c6

Block Appliconts

Percent of Blacks
Actual Denial Rate
Census Tract Effect
Residual7
Hispanic Applicants

Percent of Hispanics
Actual Denial Rate
Census Tract Effect
Residual7
White Appliconts

Percent of Whites
Actual Denial Rate
Census Tract Effect
Residual7

REFINANCE
Block Applicants

Percent of Blacks
Actual Denial Rate
Census Tract Effect
Residual7
H:spanic Applicants

Percent of Hispanics
Actual Denial Rate
Census Tract Effect
Residual7
White Appliconts

Percent of Whites
Actual Denial Rate
Census Tract Effect
Residual7

M i d d l e White4Mixed'Hispanic6

LowWhite4Mixed'Hispanic6

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table IS:(continued)

Lowv

Tod
White'%x:e$

W h i z ~ E t s

White4Mired'Minority6

Black Applicants

Percent of Blacks
A c ~ aDenial
l
Rate
Census Tmct Effect
Residual7

17.6% 6.0% 1.7%
39.3 43.0 39.5
3.4
5.5
5.3
9.2
7.8
3.9

54.8% 6.4% 1.3%
46.0 48.7 51.0
8.2
9;s
9.0
7.8
7.5
7.2

13.4
33.1
1.0
5.3

3.7
27.6
-1.3
5.1

11.9
37.3
1.7
6.9

14.6
37.8
2.0
7.3

2.6
32.6
2.5
5.2

6.2
42.0
5.8
5.8

25.3
38.4
2.1
6.0

1.0
25.9
0.3
0.8

32.4
17.4
-1.8
-0.1

5.9
27.4
0.9
-0.1

0.9
29.8
2.2
-0.2

24.7
21.6
0.7
-0.0

2.8
32.6
4.5
-0.0

0.8
36.8
4.2
0.3

100.0%
43.4
6.3
8.0

1.2%
6.7% 4.3%
33.4 35.2 37.8
-0.9
1.6
4.2
8.8
6.9
5.4

100.0
35.4
1.4
6.2

5.0
27.6
-2.1
5.5

17.5
31.4
-0.3
6.3

100.0
Zll.3
-0.9
-0.0

21.8
17.3
-3.1
0.0

9.7
22.4
-0.8
0.1

Hispanic Applicants

Percent of Hispanics
A c ~ aDenial
l
Rate
Census Tract Effect
Residual7
White Applicants

Percent of Whites
A c ~ aDenial
l
Rate
Census Tract Effect
Residual7

' Census uacts with mean applicant income of more than $60.000.
Census tmcts with mean applicant income greater than $40,000and less than or equal to $60.000.
'Census m t s with mean applicant income of $40,000or less.
Census m t s with less than 5 percent Hispanic applicants.
'Census m t s with 5 percent or more and 25 percent or less applications from Hispanic applicants.
Census traas with more than 25 percent of all loan applications from Hispanic applicants.
The residual is defmd as the average difference between the acNal denial rate and the sum of the economic. MSA, tracL and lender
effects.

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 16: Neighbohood and Unexplained Denial Rate Residuals, Blacks, by MSA, 1990 HMDA

Home Purchase
Percent Denial Tract Residual
Black Rate Effect Effect

All MSAs c 1 Million
All MSAs 1 - 2 Million
Anaheim
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dallas
Detroit
Houston
Los Angela
Miami
Minneapolis
Nassau/SuffoUt NY
New York
Oakland
Philadelphia
Phoenix
Pittsburgh
Riverside CA
S t Louis
San Diego
San Francisco
Seattle
Tampa
Washington
Total

Refinance
Percent Denial Tract Residual
Black Rate Effect Effect

Home lmorovement
Percent Denial Tract Residual
Black Rate Effect Effect

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

-

Table 17: Neighborhood and Unexplained Denial Rate Residuals. Hispanics. by MSA. 1990 HMDA

Purchase
Percent Denial Tract Residual
Hispanic Rate Effect Effect
All MSAs c 1 Million

AU MSAs 1 - 2 Million
Anaheim
Atlanta
Ealtimore
Boston
Chicago
Cleveland
Dallas
Detroit
Houston
Los Angels
Miami

Minneapolis
Nassau/Suffolk NY

New York
Oakland
Philadelphia
PhoeniT
Pittsburgh
Riverside CA
S t Louis
San Diego
San Francisco
Seattle
Tampa
Washington
Total

Refinance

Percent Denial Tract Residual
Hispanic Rate Effect Effect

Percent Denial Tract Residual
Hispanic Rate Effect Lffect

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

-

Table 18: Neighborhood and Unexplained Denial Rate Residuals, Whites, by MSA, 1990 HMDA

Home Purchase
Percent Denial Tract Residual
White Rate Effect Effect
All MSAs < 1 Million
All MSAs 1 - 2 Million
Anaheim
Atlanta
Baltimore
Boston
Chicago
Cleveland
Dalh
Detroit
Houston
Los Angeles
Miami
Minneapolis
Nassau/Suffolk NY
New York
Oakland
Philadelphia
Phoenix
Pittsburgh
Riverside CA
St. Louis
San Diego
San Francisco
Seattle
Tampa
Wahington
Total

Refinance
Percent Denial Tract Residual
White Rate Effect Effect

Percent Denial Tract Residual
White Rate Effect Effect

-

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 19: Black-White Residuals by Denial Probability. 1990 HMDA

Denial Probability
Want)
Less than 0
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19

2.0
21
22
23
24

25
26

n

28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50

More than 50

Cumulative Residual Tract
Distribution Difference Difference

Refinance

Cumulative Residual Tract
Distribution Difference Difference

Homelmnrovement

Cumulative Residual Tmct
Distribution Difference Difference

http://www.clevelandfed.org/Research/Workpaper/Index.cfm

Table 20: Hispanic-White Residuals by Denial Probability. 1990 HMDA

Denial Probability
(percent)
Less than0
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26

n

28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
More than 50

Home Purchase
Cumulative Residual Tract
Dishbution Difference Difference

Refmance
Cumulative Residual Tract
Dishbution Difference Diference
7.8%
8.6
10.4
12.4
14.6
16.9
19.4
22.1
25.0
28.0
31.2
34.5
37.9
41.3
44.7
48.0
51.3
54.3
57.3
60.1
62.7
65.2
67.6
69.9
72.2
74.3
76.3
78.2
80.0
81.6
83.2
84.7
86.1
87.3
88.4
89.5
90.5
91.4
92.2
93.0
93.7
94.3
94.9
95.4
95.9
96.3
%.7
97.0
97.3
97.5
97.7
97.9
100.0

2.4%
-0.4
2.0
2.2
3.1
2.4
05
1.5
1.3
2.0
4.1
2.6
4.2
5.4
6.0
5.0
5.9
7.0
6.4
4.8
6.7
6.7
7.0
3.4
6.3
6.6
2.6
4.3
4.4
8.1
7.1
9.1
5.3
8.1
6.3
8.0
8.8
2.7
5.0
4.6
6.4
5.7
8.3
7.6
6.8
9.4
9.1
9.5
9.1
14.8
11.9
9.4
4.3

.

0.0%
1.3
1.3
1A
1A
1.9
2.5
1.9
2.1
2.5
2.6
1.9
2.5
2.4
2.4
2.4
2.4
2.7
2.1
2.9
2.1
2.7
2.2
2.3
2.1
22
2.3
1.7
2.2
2.4
2.2
2.1
2.3
1.8
2.2
2.0
2.4
2.1
2.0
3.3
1.7
12
2.3
15
1.4
2.4
1.1
1.7
1.7
4.6
4.5
1.3
1.6

Home Imorovement
Cumulative Residual Tract
Dishbution Difference Difference