Mortgage Redlining Research: A Review and Critical Analysis Discussion

Mortgage Redlining Research: A Review and
Critical Analysis Discussion

George J. Benston*
Redlining as a Subject for Research
The research on mortgage redlining is, perhaps, even a more interesting
phenomenon worthy of study than the subject of the research. Allegations of
redlining - the refusal of lenders to make mortgages or their imposition of more
onerous terms on mortgagors because of (noneconomic) bias related to the loca-
tion of the property - are of quite recent origin. Before the early 1970s, the
term, redlining, was used almost exclusively to describe an alleged practice by
insurers of drawing a red line around sections of. a city to delineate an area with-
in which they would not write policies. While mortgagees (and others) had been
accused of discriminating against borrowers solely because of their race, national
origin, sex, and age, the term "redlining" was not employed to describe these
deplorable practices. By now the term is used as a general pejorative description
of some institutions alleged refusal to serve some area or even group (e.g.,
"credit card redlining" and "Hispanic redlining).
The extension of the term "redlining" testifies to the remarkable effective-
ness with which it has been used to characterize alleged behavior. I believe that
the occurrence is due, in large measure, to the effective public presentation of
data by community group activists to support their charges. They have suc-
ceeded in getting laws of farreaching importance enacted, including the Home
Mortgage Disclosure Act of 1975, the Community Reinvestment Act of 1977
and a number of similar state laws and regulations. This may be considered the
first phase of redlining research. The second phase began with the entrance of
academic researchers. The third phase, blissful forgetfulness, may already have
begun. If so, we will have lost an opportunity to examine, scientifically, the
empirical basis on which a considerable extension of regulations rests. I consider
first some of the conditions that gave rise to the research attempts by commu-
nity groups, a remarkable phenomenon in itself. Then I review empirical studies
on redlining grouped according to the type of question considered.
Community Concerns and Allegations of Redlining
The late 1960s and early 1970s saw many older urban neighborhoods de-
faced by abandoned, boarded-up and burned-out buildings. The populations of
*George J. Benston is Professor of Accounting, Economics and Finance, in the
Graduate School of Management, University of Rochester.
144
REDLINING RESEARCH BENSTON
145
many of these neighborhoods also changed as new people purchased or rented
the houses of former residents who had moved to the suburbs. Often the new-
comers were of a different religion, ethnic background, or race. The world of
the remaining residents had changed, and not (as they perceived it) for the
better. Not surprisingly, the concerned people in these disrupted communities
did not conduct careful, well-reasoned analyses of the reasons for neighborhood
change, abandomnents, and deterioration. Had they done so, they might have
considered the role of population movements, particularly the effect of the
movement of increasingly affluent white city people to suburbs and the in-
migration of poor, often black people to the older parts of the cities. When the
numbers of families who moved out exceeded the numbers who moved in,
dwelling units necessarily became vacant and often buildings were abandoned by
owners who preferred not to accept negative cash flows.
1
They also might have
analyzed the effect of government highway programs that decreased the cost of
transportation to and from suburban areas by workers and businesses. The role
of the FHAs programs in the late 1960s that made funds available at ~low and
even zero downpayments and below market interest rates to purchasers of
houses in older, declining urban areas was particularly important.
2
In approving
loans on properties that did not meet ordinary underwriting standards and in
encouraging the purchase of homes by poor people who ordinarily would not be
considered financially qualified, the FHA was responsible for much of the
neighborhood change and for the abandonments that upset the remaining origi-
nal residents.
3
Rather than address themselves to these causes, the comlnunity activists
concentrated on the lending practices of banks and savings and loan associations.
One reason for this approach may have been reaction to the FHAs expanded,
lower standards lending program. There appeared to be a clear association be-
tween the in-migration of newcomers (such as blacks), abandoned housing and
FHA loans. In fact, the term "FHAing a neighborhood" was coined and repeated
often as a means of describing the presumably conscious destruction of an area
by a malevolent force. That force was not identified as the federal government,
HUD or the FHA, but as the chartered financial institutions and mortgage
bankers that made the FHA-guaranteed mortgages. Though a complete analysis
of the reasons for this choice of targets has not been made, to my knowledge,
the following explanations may suffice. First, public criticism of the FHA for
extending its guarantees to poorer, often black, people who previously were un-
able to utilize them is not appealing to people, though they profess sympathy
and concern for these groups. Rather, demands that financial institutions make
I See Berry [1976], who points out that, over the decade 1960-70,482,000 new units
were constructed in the Chicago SMSA, while the number of new households increased by
only 285,000.
2 Section 104 of the Housing and Urban Development Act of 1968 established a Special
Risk Fund for interest-subsidized home mortgages, mortgagors eligible for credit assistance,
and properties not meeting the requirement of economic soundness which had to be met in
the FHAs ordinary section 203 (b) mutual mortgage insurance program.
3 For a somewhate more extensive discussion, see Benston [1977], pp. 63 -73 .
146
REGULATION OF FINANCIAL INSTITUTIONS
conventional loans rather than FHA-guaranteed loans is equivalent to having
them deny loans to people who would only qualify for FHA mortgages. Second,
middle-class people who wanted to move into the older neighborhoods and who
preferred conventional mortgages (perhaps because they did not qualify for
interest subsidized FHA mortgages) may have been refused by banks, because
the banks perceived the neighborhoods to present bad risks, either because the
banks approached did not serve the areas (a not uncommon situation in the then
unit banking environment of Chicago, where the anti-bank movement was the
strongest), or because the bankers they approached were biased. Therefore, these
relatively vocal, action-motivated people saw the banks in the role of villains.
Finally, chartered financial institutions make very attractive targets. They are
regulated by the federal and state governments, and therefore are amenable to
political pressure. They appear to have control over a great amount of resources
that could be directed in alternative ways. And since they characterize them-
selves as serving the public, why should they not be required to serve a given
neighborhood?
4
Community Activists Research on Redlining
Either by chance or design, the comnrunity activists enlisted research in
their campaign against FHA lending and towards forcing banks and savings and
loan associations to make more conventional loans to buyers of property in
older urban areas. They sought to prove to the press, the public and the legis-
lators that the lenders were not serving these areas. They were remarkably suc-
cessful in this endeavor, in part as a consequence of the brilliant organizational
and publicity skills of Gail Cincotta, organizer and leader of the National Peoples
Action in Housing and the National Training and Information Center. Studies
were conducted by community groups in cities around the country, reported to
the press and presented before congressional, state and local government com-
lnittees,
s
As a consequence, laws were passed and regulations promulgated that
have resulted in a great amount of data on mortgage activity by chartered finan-
cial institutions being made available. These and other data are now being used
by community and professional researchers. An assessment of that research
follows.
Basic Requirements for a Study of Redlining
For valid conclusions to be drawn, any study of redlining requires the fol-
lowing minimal conditions, at least, to be met:
1. If the focus is on the terms of the mortgage - such as the interest rate
charged, the downpayment required or the term to maturity given - eco-
nomic factors that impose greater costs on the lenders must be accounted
See Benston [ 1978A] for additional discussion.
See Agelasto and Listokin [1977] for a smnmary of these efforts.
REDLINING RESEARCH
BENSTON 147
for. A model that describes the mortgagees decision function using only
variables that provide estimates of economic costs and benefits should be
constructed and specified. Its outcome with respect to mortgage terms
then can be compared to the actual terms experienced by mortgagors in
situations where it is claimed that the terms charged were based on non.
economic considerations. If this is not done, there will be no way of con-
cluding whether the terms charged reflect only cost differences among
mortgages, in which event income redistribution is the appropriate re-
sponse (such as direct subsidies to mortgagors and house sellers or forced
wealth redistributions from financial institutions and savers to mortgagors
and home sellers).
2. If the focus is on the supply of mortgages, either in terms of numbers or
dollars, a demand as well as a supply function must be constructed and
specified. When demand is not accounted for, there is no way to deter-
mine the reason for any given level of supply.
3 .
Furthermore, if the focus is on the effects of redlining on consumers and
on local areas, data from all mortgage lenders must be included in the
study. When this is not done, the most that can be demonstrated is that a
particular lender (or subset of lenders) serves areas differently, since other
lenders not included in the study may be a preferred or equivalent source
of funds.
It is not reasonable to expect community activists to develop and specify,
formally, the mortgage decision function model required for an analysis of
mortgage terms or the supply and demand functions required for an analysis
of the supply of mortgages to areas. However, it is reasonable to expect them
or those who rely on their presentations to consider the effect on mortgage
terms of property characteristics, such as the past variability of house prices
and prospects for increases or decreases in value, and the borrowers financial
situation, such as wealth, income, job, and credit record. Moreover, considera-
tion of the demand for mortgages is a crucial aspect of any study of the num-
ber or dollars of mortgages supplied. If these variables are disregarded or not
adequately accounted for in a study, the only conclusion that should be drawn
is that the situation may be worth a well-constructed study.
6
Types of Redlining Research
Research on redlining may be grouped into six categories: (1) analyses of
the supply of mortgages in terms of number and dollars; (2) measurement of
the relative risk of lending in areas or to borrowers in terms of defaults and
foreclosures; (3 ) estimates of the terms charged to mortgagors; (4) comparison
of the ratios of appraisal values of purchase prices of properties; (5) evaluation
61 assume (naively, perhaps) that the purpose of the laws and regulations is the removal
or mitigation of an inequity rather than the redistribution, via political power, of wealth
from banks, savers and the general public to home owners and buyers in older urban
neighborhoods.
148 REGULATION OF FINANCIAL INSTITUTIONS
of the extent and determinants of denials of applications for mortgages; and
(6) estimates of the demand for mortgages by actual and potential home buyers.
The first type of research (supply) cannot demonstrate or disprove redlining,
but may give some insights into lending patterns. The second type (defaults)
can indicate whether risk considerations are a factor in the lending patterns
observed. The third type (terms) can provide some evidence of discrimination,
given that mortgages are made. The fourth (appraisal ratios) examines an alleged
method of discrimination. The fifth (denials) seeks to determine whether mort-
gages were not granted because of discrimination. Finally the sixth (demand)
attempts to determine if applications were discouraged by lenders for unaccept-
able discriminatory reasons or whether supply patterns reflect demand. The
studies in each category are reviewed and evaluated briefly in the balance of the
paper, and in greater detail in the appendix.
Analyses of the Supply of Mortgages
Research by Community Groups
The principal procedure employed by community groups was to count
mortgages made by a subset of lenders to property buyers in a city area com-
pared to a suburban area. The period over which mortgage lending was measured
often was quite short, generally a half year to a year and a half. No attempt was
made to measure the costs of lending (in particular the expected cost of de-
faults) on property located in the areas of concern. However, since few of the
studies included measurements of mortgage terms, this omission is not too
important. Failure to measure or even consider the demand for mortgages is a
more serious shortcoming. Absent demand analysis, one cannot conclude from
the fact that more mortgages were made in, say, a city area compared to a
suburban area, that the lender discriminated against purchasers of city property.
Nor can one infer from a finding that relatively more FHA-insured mort-
gages and fewer conventional mortgages were made on city property than on
suburban property that lenders "pushed" FHAs on or denied conventionals to
buyers of city property. In addition, almost none of the studies attempted to
measure the mortgage activity of more than a subset of lenders, usually savings
and loan associations and savings banks. Mortgages made by commercial banks
and mortgage bankers almost never were included in the data. Furthermore, the
periods over which mortgage lending activity was considered generally were
quite short, with no account being taken of the prior lending practices of the
institutions studied. Hence, even were demand equal in the areas compared, one
cannot conclude that a specific institution was or was not deliberately neglecting
some area. Nor can one conclude that an area was under- or over-served (in some
sense).
Perhaps the only meaningful data presented in these studies are reports of
the experiences of individuals. These case studies may provide evidence that a
specific lender or employees of the lender denied mortgages to individuals pri-
marily on the basis of socially unacceptable criteria, such as the applicants race
or sex. However, without some evidence of independent verification, there is
no way to tell whether the applicants perception of discrimination was correct.
REDLINING RESEARCH
BENSTON
149
Nor can one determine if the practices reported were specific to an individual
lending officer, an institution, or to lenders in general. In any event, my reading
of the community activists studies revealed very few reports of individual dis-
crimination. Considering other evidence on the biases held by many white, mar-
ried, middle-class males who tend to live in suburbs (a group that dominates
mortgage lending officers), I was surprised to find so few specific complaints.
The generalizations presented above can be illustrated, in part, by reference
to a few of the more prominent studies. In addition to the shortcomings dis-
cussed above, some of these studies have presented data and conclusions that
can be characterized fairly only as fraudulent. A few of these situations are
pointed out in the brief reviews given in the appendix.
7
Research by Professionals
Since "redlining" often is defined as the non- or reduced-ava~ability of
mortgage financing for house purchases in discriminated-against areas (or to
persons), some researchers have attempted to use supply and demand analysis
(formally or implied) to determine whether a particular area or group received
fewer mortgages than would be expected were discriminating practices not
present.
8
The first (to my knowledge) of these studies, Devine [1973 ], actually only
presents an analysis of supply. Indeed, the possibility that the amount of mort-
gages supplied could be a function of the amount demanded was not even men-
tioned. This study is much more sophisticated than the community group
studies, both in language (a dispassionate discussion of presumed lending be-
havior is presented at some length) and method (multiple regression analysis).
Perhaps as a consequence, it was cited often in the hearings preceding enact-
ment of the Home Mortgage Disclosure Act of 1975. It is seriously deficient,
though, even as a supply study. The 14 geographical observations (Community
Planning Districts) are too heterogeneous with respect to race, income, housing
and other important variables. The data include less than 22 percent of the mort-
gages in the area. And the regressions indicated that the proportion of minorities
in an area was related to mortgages granted only if some observations were
dropped and independent variables redefined.
The studies following Devines effort were much more carefully modeled
and specified. Ahlbrandts [1977, pp. 474-475] paper provides a succinct illus-
tration of the type of model developed. Four equations (slightly changed here)
are presented:
(1) mortgages demanded, M
d
=
fd (T, P, i)
7See Benston, Horsky and Weingartner [1978], Chapter 1 for a detailed critique of
nine studies. Also see King [1979A] for a critical review of 14 studies and papers.
8See the appendix for more detailed reviews of the eight studies mentioned that
support the conclusions given here.
150
(2) mortgages supplied, M
s
=
fs (i, C, N, P)
(3 ) potentia! real estate transactions, T
=
ft (i, Ms, P, C, N)
(4) in equilibrium, M
d
=
M
s
=
M
where P
=
price of land and housing
i
=
loan terms (a vector interest rate, downpayment, and term to maturity)
C
=
creditworthiness of loan applicant
N
=
neighborhood characteristics (current and expected change).
The model then is silnplified by substitutions to obtain the following reduced
form:
(5) M = fm
(P
C, N).
The reduced form then is specified with mortgage flow data (M) from a subset of
lenders, aggregated by census tracts, with borrower and neighborhood character-
istic variables (C and N) obtained from census data (usually from the 1970
census).
Ahlbrandt [ 1977] used an expanded version of equation (5) in his study of
mortgage lending in Pittsburgh. Mortgages made, by census tracts, was regressed
on percent black and other variables. His study is a considerable improvement
over earlier supply studies in several important regards. He included mortgages
made by all financial institutional lenders (though other sources of funds were
excluded). Variables other than race (or area) are included (though the values
are four years out of date).
However, the reduced form of equation cannot yield estimates of mortgage
over- or under-supply, since demand for mortgages is not accounted for. If cen-
sus tracts occupied largely by whites are characterized by greater mobility and
hence greater demands for mortgages, "percent black" will be negatively asso-
ciated with the number of mortgages demanded and supplied. Or if blacks are
discriminated against by mortgage lenders, they may obtain funds at more
onerous terms but these terms will not be revealed by the analysis. Or blacks
may prefer to borrow from noninstitutional lenders because the terms offered
(such as low downpayments) are less onerous or the service supplied is better.
Furthermore, other independent variables (such as family income and the crime
rate) are likely to be associated with percent black because of the economic
condition of blacks due to past (and present) discrimination in employment
and education.
9
In this event, tlie relationship between race, demand for mort-
gages, and risk cannot be separated.
Hutchinson, Ostas and Reed [1977] used data provided by Toledo, Ohio
savings and loan associations. Though their study is more elaborate than Ahl-
brandts, it suffers from the same serious shortcomings. At most, they find that
See Hauser and Elkhanialy [1978] and Dingemans [1979].
REDLINING RESEARCH
BENSTON
151
the mortgages made by these lenders in 1975 may have been related to the
average percent black in census tracts. Based on an analysis of the data, the re-
searchers conclude: "These ~esults are consistent with the hypothesis that red-
lining takes place on the basis of risk aversion... [rather than] on a taste for
discrimination." [p. 469]. In another updated and better paper, (Ostas, Reed
and Hutchinson [1979] ), they used an improved variable to measure demand -
the number of one-to-four family units shown in real estate brokers multiple
listing guides. The results are similar to those found in their earlier paper. But
the reduced-form equations cannot provide information as to why or whether
more or fewer loans were made in census tracts.
Hauser and Elkhanialy [1978]present data on lending by savings and loan
associations in Chicago. They show that relatively fewer loans were made in
census tracts that are heavily black, and that "black" is positively correlated
with low family income, nonowner occupied, and low home values, variables
which explain much of the differences in lending.
Dinglemans [1979] also finds that factors other than minority status
explain differences in lending by savings and loan associations. His careful
study of Sacramento describes the activities of real estate agents and mort-
gage bankers in providing mortgages and shows that if their supply is excluded,
areas not said to be redlined will appear to be so.
Whalens [1976] study of Flint, Michigan attempts to separate the effect
of variables such as income and quality of housing from race by regressing
mortgages made first on these "prudent lending" variables (as he calls them)
and then on race and other "discrimination" variables. With respect to race,
he finds only that the average price of houses in a census tract (which he mis-
takenly identifies as the average mortgage amount) is negatively related to the
percent black and racial change.
Schafers [1978, Chapter 5] study of New York City represents the only
example of which I am aware of measuring the possible shortfall of mortgage
funds rather than simply the relationship between the characteristics of census
tracts and the amount of mortgages made (and perhaps demanded). The three
counties in New York City studied, Bronx, Kings (Brooklyn) and Queens, were
dichotomized into neighborhoods that are alleged to be redlined and neighbor-
hoods that are not. The identifications were based on interviews with people
from community organizations and from published reports.
~
A model relating
mortgage lending to explanatory variables was estimated for each area separate-
ly, and the coefficients compared. The coefficients from the presumably uncon-
strained (by redlining) data then were multiplied by the values of the inde-
pendent variables from the alleged redlined observations and then aggregated to
predict the expected amount of mortgage funds, were there no redlining. The
predicted compared to the actual provides an indication of the extent to which
mortgage funds are undersupplied to the alleged redlined neighborhoods.
11
The
10 This procedure previously was employed by Benston, Horsky and Weingattner [ 1978]
in their study of Rochester, New York.
11A similar procedure was followed for mortgage terms by Benston, Horsky and
Weingartner [1978], as is discussed below.
152
data are mortgages made by a subset of savings banks in 1975. Schafer finds no
evidence of redlining with respect to conventional one-to-four family mortgages.
Indeed, he finds that two of the seven alleged redlined neighborhoods would
receive more funds if they were redlined than if they were not. The exercise
was also undertaken for FHA and VA one-to-four family mortgages and the
multifamily mortgages, with similar results.
Finally, Koebel [1978] presents a very detailed description of all sources
of mortgage funds in Louisville, Kentucky, which reveals the considerable
extent to which chartered financial institutions are not the sole source of hous-
ing finance.
In summary, the reports of differences in the quantity of mortgages made
to city compared to suburban areas clearly demonstrate very little and certainly
prove nothing about the existence of redlining or are inconsistent with the claim
that redlining is practiced. Though professional researchers have attempted to
structure theoretically based supply and demand models, with a few exceptions
they employed supply models, at best. For the most part, they examined data
that are the result of the interaction of supply and demand as reflected in the
records of a subset of lenders. A few studies employed better instrumental
variables for demand (in particular, the real estate listings used by Ostas, Reed
and Hutchinson [1979]), but at best all that can be demonstrated is that some
lenders did not serve all markets equally. For this purpose, Schafers [1978,
Chapter 5] procedure of applying a supply function derived from assumed non-
redlined data to estimate supply to an allegedly redlined area might be prefer-
able. However, as Dingemans [1979] and Koebels [1978] studies show,
lenders tend to specialize in serving different areas, which need not disadvantage
borrowers. Therefore, other subjects of research must be considered before any
knowledge about redlining can be obtained.
The Relative Risk of Lending - Defaults and Foreclosures
The possibility that borrowers may default on their obligations, thereby
imposing costs on lenders, is a factor that obviously must be considered in
studies of lender performance. Default costs include additional administrative
expenses of processing notices when loans become delinquent less late fees col-
lected from delinquent borrowers. When defaults are repaired, the net cost usual-
ly is small or there even may be a net benefit. However, when a mortgage must
be foreclosed, the costs usually are positive, since if the net market value of the
property exceeded the loan balance, the mortgagor could have benefited by
selling the property, thereby protecting his or her credit rating as well as gaining
directly. The costs to the lender of foreclosed mortgages include legal and other
expenses related to processing documents to obtain possession of the property,
expenses related to the maintenance of the property until it is sold, legal and
other selling expenses, the loan balance owed and interest foregone on funds
invested. These costs are reduced by the amount received for the foreclosed
REDLINING RESEARCH BENSTON 15 3
property, either from sale
x
2 or from an insurance agency, such as the FHA, VA,
or private mortgage insurance company.
Lenders may adjust to the expected loss from foreclosures in several ways.
The most direct method would be to charge sufficient interest yields to offset
the net costs (adjusted for the time of occurrence). But this price adjustment is
constrained where state usury ceilings are effective or where the cost of adverse
publicity and criticism by regulators makes raising interest rates infeasible. The
lender then can reduce the risk of default by requiring higher downpayments
and shorter terms to maturity. Borrowers who present a higher probability of
default, such as those with lower levels of present and future income and wealth,
and those for whom the cost of default in the form of a poor credit rating is
low, can be screened out. Similarly, properties which present a relatively high
probability of a market value decline below the mortgage balance can be screened
out. (Indeed, where neighborhood factors are believed by lenders to offer an
efficient means of identifying such properties, the areas may be "redlined,"
though not in the pejorative discrimination sense of the word.) Therefore, since
lenders usually cannot be compensated with higher yields, they will tend to
accept only mortgages on which the expected cost of default is equal, given the
other terms, etc.
13
Defaults nevertheless occur because of errors, regulation, and possibly,
lenders biases. The errors could be caused by unexpectedly changed events,
such as an unexpected economic decline that reduces borrowers wealth and
the market value of houses. Or the default experience could be caused by regu.
latory or political agencies that force lenders to make mortgages on properties
in particular areas. Alternatively, risk-averse regulators (such as bank examiners
before the enactment of the Community Reinvestment Act of 1977) could
cause a less than desired level of defaults if they imposed costs on lenders who
wanted to make mortgages on properties in certain areas. Or the lenders could be
biased against or for persons or neighborhoods despite the expected economic
consequences of their lending decisions. Only the last possibility should be
considered to be redlining or favoritism in its pejorative sense. But it does not
seem possible to distinguish between the bias motive, prediction errors, and
regulatory influences by examining data on mortgage defaults.
Therefore, a default study should use data that were not prescreened by
lenders. Conventional mortgages made by chartered financial institutions gener.
ally do not meet this criterion, particularly in a state (such as New York) that
imposes a restrictive usury ceiling on interest rates. FHA insured and VA guaran-
teed loans would meet the criterion, except that the lenders and the government
agencies have incentives to prescreen applicants. The lenders bear costs when
mortgagors with FHA and VA loans default because these agencies do not com.
~ If the property is kept by the lender, this is the equivalent of a sale.
l~This equality holds across mortgages and is equal, on the margin (on a present value
basis), to the revenue from risk premiums charged on mortgages (relative to risk-free alterna-
tive investments and net of administrative and transactions costs).
pletely reimburse the lenders for the economic costs incurred.
14
The agencies
bear most of the costs of default. They have an even greater incentive to deny
risky loans, particularly because the insurance premium charged is the same for
all loans. However, they are supposed to disregard general neighborhood factors
and such borrower characteristics as race and sex. A particular exception to the
usual lending criteria is the FHAs Special Risk Fund loans (e.g., section 23 5 and
221 (d)(2)) and the post-1968 policy imposed on the agency of accepting loans
on properties in declining urban areas that previously would not have been
accepted. Data from these loans would be very useful for a study of default
experience related to neighborhood characteristics.
Only one of the six studies reviewed very briefly below (Marcis and Hull)
meets the full sampling criterion and another (Barth, Cordes and Yezer) meets it
partially. Therefore, the relevance of the others to the redlining question is
limited. Nevertheless, they provide some information on the actual experience
of some lenders.
~s
Characteristics that usually describe older, poorer neighborhoods, with
other factors held constant, are found to be positively associated with higher
foreclosures in most of the studies. In Williams, Barenek and Kenkels [1973 ]
study of Pittsburgh, this variable is the unemployment rate in a census tract.
Percent black was excluded as an independent variable because it was correlated
at .75 with the unemployment rate. (Per capita crimes against persons and
median family income also were omitted for that reason.) The unemployment
rate is significantly, positively associated with defaults, though it is not as statis-
tically significant as some of the other explanatory variables. In Mortons [1975]
study of Connecticut, the neighborhood variable is three-family houses. These
generally were converted from older houses in declining neighborhoods. His find-
ings indicate that neighborhood quality, as measured by this variable, affects
defaults positively.
Schafers [1978, Chapter 3 ] study of foreclosures in alleged redlined and
other New York City neighborhoods shows some evidence of greater foreclosures
in the former. A multiple regression analysis, which did not include alleged red-
lined and other neighborhoods as variables, shows statistically significant co-
efficients indicating higher foreclosures in census tracts with higher percentages
of poor condition buildings, multiple-wage earner households, and single-family
houses. His studies of four upstate New York cities [Schafer, 1978, Chapter 8]
also found some (weak) evidence of relatively greater foreclosure rates in alleged
redlined compared to other neighborhoods. The borrowers race was not in-
cluded in any of his analyses.
laThe FHA does not fully reimburse the lender for interest lost on the first 60 days
after default, for one thkd of direct foreclosure and acquisitions costs, or for the lenders
own (internal) foreclosure expenses. Furthermore, interest on the lenders investment (after
60 days) is allowed at the FHAs debenture rate. The VA does not reimburse the lender for
attorneys fees greater than $250 or for the lenders own expenses. However, the principal
balance due is credited at face value, gross of unamortized points. The expected cost is the
present value of the increased probability of default times the unreimbursed cost.
lSSee the appendix for detailed reviews of the methodologies and findings of the six
studies.
REDLINING RESEARCH
BENSTON
155
Marcis and Hull [1975] is the only study to use largely unscreened data on
FHA section 221(d)(2) and 23 5 insured mortgages. For the former, they found
the zip code area characteristics of higher percentage of female family headship,
lower family income, greater instability of occupancy, and lower level of educa-
tion to be associated with higher foreclosure rates. The percentage differences
found are considerable (e.g., a 10 percent increase in the percentage of female
heads of households is associated with a 1.5 percentage point increase in fore-
closure rates, which averaged about 5.5 percent). This study also did not specify
the borrowers race or the racial characteristics of the area.
Finally, Barth, Cordes and Yezer [1979] used nationwide FHA data for
regular, section 203 (b) insured mortgages. As compared to conventional mort-
gages, these mortgages are less likely to have been prescreened to reduce defaults.
The study also benefits from a large set of carefully sampled data. The authors
found that defaults are statistically significantly higher on mortgages made to
blacks (1.2 percent), lower to female heads of household (0.3 percent), higher
for wood constructed houses (0.2 percent), higher on buildings located in the
central city (0.1 percent) and in blighted areas (0.3 percent), compared to an
overall average default rate of 1.5 percent.
Thus all the studies report data that are consistent with the conclusion that
foreclosures are higher in areas often alleged to be redlined or to some persons
alleged to be discriminated against, even though all but two used data that were
screened to reduce (if not eliminate) foreclosures, and the least screened of the
two used overly aggregated data. (I should note, again, that black" is likely to
be a surrogate for other variables.)
Terms Charged to Mortgagors
The most direct measure of bias against mortgagors is the price paid for the
funds they receive compared to the price paid by others, cet. par. Two problems
must be solved for such an analysis to yield meaningful findings: price must be
defined and measured and all the other factors that result in price differences
must be accounted for. Price for mortgages essentially is the interest rate paid
(including points). However, it also includes other mortgage terms, particularly
the downpayment required and the term-to-maturity offered. The others consist
primarily of those tha~ affect the expected cost of nonrepayment and non-
economic factors, including possible bias against or for certain mortgagors. The
four studies reviewed below represent different attempts at quantifying these
variables and measuring their relationship,
x6
The studies reviewed are inconsistent with the hypothesis that buyers of
houses in allegedly redlined areas, defined by community groups or as census
tracts with a high percentage of blacks, were discriminated against. Schafers
[1978, Chapter 6] study of New York City is the most elaborate. He computed
simultaneous equations that included the average mortgage terms, in census
tracts, of the interest rate, loan-to-value ratio, and term to maturity. The regres-
1~ See the appendix for more detailed reviews on which the conclusions given here are
based.
156
sions have 45 independent variables, including dummy variables for six alleged
redlined neighborhoods, and the percentage nonwhite and change in the per-
centage nonwhite. The results for conventional one-to-four family mortgages
generally are inconsistent with the hypothesis that terms are more onerous for
alleged redlined neighborhoods or groups. However, the results are mixed,
indeed a bit confusing. Few of the coefficients have statistically significant
coefficients and some have the "wrong" sign. These results also are reported for
data from a special study and for multifamily houses. However, the quality of
the data are so questionable that they limit acceptance of his results severely.
Muths [1979] study of terms on 1903 conventional one-to-four family
mortgages made by state-chartered savings and loan associations in Oakland,
California in 1976 and 1977 is very well designed. As he points out:
For redlining if it exists would imply a reduction in the supply of mortgage
loans. Fewer loans might also result from a smaller demand for conventional
mortgage loans, either because potential purchasers of homes prefer other
methods of finance or because the demand for purchase itself is lower in
areas where fewer mortgage loans were made. A reduction in the supply
of mortgage loans, given their demand, would lead to higher prices or
mortgage yields on such loans. Alternatively, if fewer loans are made be-
canse the demand for them is smaller, the yields on loans made would tend
to be the same or lower than those on loans made in other areas. In seeking
to determine why fewer conventional mortgage loans are made in certain
parts of cities, then, the variation of mortgage yields is of critical impor-
tance. [pp. 1-2]
Muth calculated the theoretically correct rate of interest (which is not con-
strained by a usury law) for each mortgage and regressed it on variables that
measured borrower and property characteristics. The coefficients calculated
indicate very slightly higher interest rates (which averaged 9.3 percent) associ-
ated with black male borrowers (on average, .02 higher), other minority males
(.03 higher), but lower rates with females (.03 less). Loan.to-value ratios are
slightly higher for black than for white male mortgagors (.84 compared to .79)
and maturities averaged 5 percent less in all black compared to all white census
tracts. However, when variables for the quality of housing and family income
were included in the regression, the coefficients for "black" became small and
insignificant. Muth concludes: "This suggests to me that the higher riskiness
lenders may attach to loans in black neighborhoods is not due to race per se but
rather to the poorer quality of dwellings and lower borrower income on loans
in such areas." [p. 15]
Benston, Horsky and Weingartners [1978] study of mortgage terms in
Rochester, New York is unique in that they obtained data on 712 one-to-four
family house mortgages made over a three-year period from all the lenders
in an area, including mortgage bankers. They also employed the technique of
applying a lending function from a control to an alleged redlined area (as defined
by anti-redlining group leaders), and vice versa. Analyzing mortgages by type
(conventional, FHA and VA) separately, they found almost no difference in
157
interest rates between the areas. Loan-to-value ratios, the number of months to
maturity and a variable that combined the terms were regressed separately on
variables that characterized the borrowers and the properties. Separate regres-
sions were computed for each area and for conventional, FHA, and VA mort-
gages. The analysis showed that virtually all of the differences between the areas
in loan-to-value ratios and most of the differences in months to maturity were
due to factors other than area. The combination variable showed almost no dif-
ference between the areas, even though account was not taken of area-related
risk factors.
Finally, King [1979B] analyzed the terms onmortgages made by savings
and loan associations in Miami, San Antonio and Toledo on 1960, 953 and 559
owner-occupied, residential properties made during three months in 1978. For
mortgages made in each city he regressed interest rates, term to maturity, loan-
to-value ratios, and fees on variables that specified the borrowers characteristics
(particularly minority and marital status, sex and age), the age of the structure
mortgaged, neighborhood characteristics (particularly the percentage of minority
population in census tracts in 1970), the neighborhood in which the pro15erty is
located (for Toledo only), and the savings and loan association that made the
loan. With OLS and simultaneous equations, he found almost no evidence of
more onerous terms related to possible discriminatory factors (such as race or
sex), particularly with respect to the difference between the requested and
granted term to maturity and loan-to-value ratios.
Appraisal Values and Purchase Prices of Properties
Critics of bank lending practices claim that appraisers systematically under-
appraise properties in some areas and/or on which minorities or others make
offers. Consequently, they claim, the potential purchasers either must make
larger downpayments or are prevented from purchasing the houses. Alternatively
(and perhaps partially) the sellers can reduce the price of the houses.
17
Four
studies analyze this situation in various ways. None of them find evidence that
supports the biased appraisal hypothesis.
18
Schafers [1978, Chapter 4] study of the New York City area used data
from a long period (at least 27 years) gathered from savings banks. He regressed
the ratios of appraisal to selling price on variables measuring the age of the build-
ing, the type of mortgage, the neighborhood in which the building is located
(alleged redlined or other in New York City), and the year in which the mortgage
was made. The regressions also were computed separately for alleged redlined,
other New York City neighborhoods, and three suburban counties. The analyses
reveal no evidence that the banks systematically underappraised older houses or
properties in the allegedly redlined neighborhoods, either for one-to-four family
17It is worth noting that if the selling price of a house were depressed because of under-
appraisals, the purchasers would benefit at the expense of the sellers, but there is no other
necessary effect. See Benston [1978] for an analysis.
aaSee the appendix for details.
or multifamily houses. Schafer [1978, Chapter 9] repeated the analysis for four
upstate New York cities and their suburbs for one-to-four family properties, with
similar results.
Benston, Horsky and Weingartners [1978] study of Rochester, New York
mortgages compared the ratios to appraisal to selling price in an allegedly red-
lined area with a presumably favored suburb. Conventional, FHA and VA mort-
gages were scheduled separately. Virtually no differences were found between
the areas for each type of mortgage.
Kings [1979B] study of savings and loan associations in Miami, San Antonio
and Toledo is unique in that he included mortgages applied for and denied as
well as mortgages made. He first reviewed the reasons given for denials and
found very few to be "inadequate appraised value" and these few were not related
to the borrowers race. Second, he regressed the appraisal ratio on variables
representing the type of mortgage, age of property, the borrowers race, sex and
age, and neighborhood characteristics that might be associated with redlining.
This analysis also found no evidence of discriminatory behavior by the associa-
tions; to the contrary, significantly higher ratios were found for blacks in Miami,
and Spanish-American and some city neighborhoods in Toledo.
Denials of Mortgage Applications
The studies reviewed to this point present findings that either are not mean-
ingful or that are inconsistent with the hypothesis that mortgage lenders dis-
criminate against areas or minority persons. However, except for Kings study
of appraisals, they use data that results from a mortgage having been made. It
may be that lenders discriminate by denying mortgages applied for by minorities
or for houses located in allegedly redlined areas. They also might accept such
mortgage applications, but only with modifications that impose harsher terms
on those against whom they are biased. But if this practice had occurred, mort-
gages with harsher terms would have been written and should have been dis-
covered by the studies of mortgage terms reviewed above. Since these studies
reach contrary conclusions, the modification method of discriminating does not
seem likely as a practice. Two other possibilities remain. Lenders could be so
negative that they cause potential borrowers to withdraw their applications.
But considering the cost of filing an application (the borrowers time and effort
plus application and appraisal fees charged by lenders), it seems much more
probable that completed applications were withdrawn because borrowers found
superior alternatives. The other possibility, that potential borrowers are dis-
couraged from applying or are turned down before filing an application, seems a
much more likely possibility. Research on this possible redlining method is
reviewed in the next section.
The studies of mortgage applications seek to determine whether applications
are denied for reasons not associated with ordinary economic considerations, but
rather with the applicants race or the location of the property. The possible
limitations of this type of analysis are well stated by Schafer [1978, p. 7-4], as
follows: "If the discrimination variables [e.g., applicants race] are not statisti-
REDLINING RESEARCH
BENSTON
159
cally significant, however, it can be inferred that discrimination is not a factor,
provided that the race variable is not correlated to any relevant variable that has
been excluded because of inadequate information. If the discrimination variables
are statistically significant, it can only be stated that these results are consistent
with the existence of discrimination. The extent to which a result is taken to
support some form of discrimination depends on how complete the rest of the
model is; that is, on how well the model controls for non-discriminatory factors
that enter into the decision to lend."
The six studies reviewed apply somewhat different approaches to different
sets of data and report somewhat different results.
19
The first study suffers from
a paucity of denied mortgages. As is elaborated below, I believe that the Warner
and Ingram study is superior to Schafers two studies. Kings study is more com-
prehensive than the other studies in that it covers three cities. The last study is
included for completeness.
Black, Schweitzer and Mandel [1978] used data from a survey of applica-
tions filed at 176 banks nationwide (out of 3 00 asked to participate). Problems
with the data reduced the number of.observations to less than half the original
alnount, which limits the conclusions that can be drawn from the study. The
data show a denial rate of o~fly 2.7 percent. A probit regression model was em-
ployed, from which they found a positive relationship between black race and
the probability of denials that is statistically significant, but at the .10 level.
Inclusion of the applicants race in the probit regression, though, improves the
fit of the model but slightly. Unfortunately, the authors do not give (nor can the
reader compute) the amount of the greater probability of denial associated with
the applicants race.
Schafer [1978, Chapter 7] used data derived from Equal Housing Oppor-
tunity Lender forms filed at 27 mutual savings banks in New York City. His
study also suffers from missing data, which reduced the number of usable obser-
vations considerably. Nine percent of the applications analyzed are denials.
From logit and OLS analyses of denials and acceptances, he found statistically
significant relationships between the applicants race and denials which imply
that compared to a wlfite applicant, black and other minority applicants have a
greater probability of denial of from 9 to 21 percent (depending on the pro-
cedure employed). He also found that the location of the property on which a
mortgage is desired was not a factor in the denial or acceptance of the appli-
cation.
Schafer [1978, Chapters 11 and 12] also studied applications at savings
banks in the Albany-Schenectady-Troy SMSA, Buffalo, Rochester and Syracuse,
New York. The denial rates in these areas averaged 8, 8, 3 and 6 percent. The data
base also was greatly reduced because of missing information in the forms. His
logit regressions lead him to reject the hypothesis that the institutions denied
relatively more applications from buyers of properties in alleged redlined areas.
1~ See the appendix for more detailed reviews and analyses on from which these brief
synopses are derived.
160
(The exception is two neighborhoods in Albany, from which five applications
were denied.) But he reports strong estimates that blacks faced statistically sig-
nificantly greater probabilities of having their mortgage applications denied
that are about twice those of similar whites in all areas except the Albany SMSA.
But these findings are based on only 45 denied applications from blacks in
Buffalo, 6 in Rochester and 5 in Syracuse.
King [1979B] analyzed applications recorded at savings and loan associa-
tions in Miami, San Antonio and Toledo from September through November
1978. He found evidence of no adverse selection of borrowers or properties,
with the exception of Hispanic borrowers in Miami and San Antonio; denial
rates for them average about .03 higher than the mean rates of .12 for Miami
and .05 for San Antonio.
Warner and Ingram [1979] studied the behavior of almost all the chartered
financial institutions in Columbia, South Carolina. These institutions denied
about 5 percent of the conventional mortgage applications made. The authors
use multiple discriminant analysis to test the hypothesis that the addition of
variables that might be used for noneconomic discrimination affected the insti-
tutions denials. They first used a function that includes only economic variables
(e.g., applicants credit rating, loan-to-value ratio requested, and applicants
tenure in current occupation) to distinguish between accepted and rejected
applications. Then they added "discrimination" variables, such as the applicants
age, marital status, age of dwelling, sex, and race. They found that the additional
variables did not significantly increase the predictive power of the function.
Their results were confirmed with a validation sample. Tyree and Yeager [1979,
Chapter 10] study of applications for high-risk loans in St. Louis also found no
evidence of different denial rates associated with the applicants race or the
location of the property. .
Thus all the studies found no evidence of differential denial rates related to
the area in which a house is located. These findings, therefore, are inconsistent
with the hypothesis that lenders redline by denying prospective mortgagors
applications. However, the studies conducted by Schafer (and possibly by Black,
Schweitzer, and Mandel) report evidence of higher denial rates related to the
applicants black race. King finds some evidence of higher denial rates for His-
panics but not for blacks. None of the studies determine whether the results
are due to risk-related or other nonobjectionable discriminatory factors. Warner-
Ingram and Tyree-Yeager find evidence of no discrimination. The evidence,
therefore, is not consistent, except in finding that denials apparently were not
a form of area redlining.
The Demand for Mortgages by Actual and Potential Home Buyers
The applications data analyzed may not reveal discrimination by lenders
against minorities or people who want to buy property in certain areas if the
lenders are able to discourage these people flom filing an application. This form
of redlining is relatively easy to do. As Dingemans [1979] describes, real estate
agents are a major source of information about the availability of mortgage
REDLINING RESEARCH
BENSTON
161
funds. The agents have a considerable economic incentive (the commission) to
"close the deal," for which a mortgage usually is required. Therefore, they are
likely to avoid lenders who have let it be known, by word or past actions, that
they are unlikely to approve applications from certain types of people on cer-
tain properties. Lenders also require an appraisal for which they usually charge
a fee. Thus they can discourage a potential mortgagor from filing an application
by indicating that the appraisal will be delayed or that the application is likely
to be turned down. Since there are no reported data on these discouraged appli-
cants, special studies are required.
Direct analysis of the demand for mortgages also is necessary to determine
whether the lower amount supplied to certain areas (as shown by many studies)
is due to lack of demand or to lender discrimination. Instrumental variables,
such as the stock of existing houses and other census tract data, are unlikely to
be effective for this purpose. (Additionally, the census data are now 10 years
out of date.) While the number of houses listed for sale with real estate agents
(the multiple listing data used by Ostas, Reed and Hutchinson [1979]) are
perhaps as good a proxy for demand of this type as one can obtain, it cannot
distinguish between demands for conventional as compared to FHA, VA or
private mortgages. Nor can its use reveal whether potential home sellers did not
list their homes because they believed that buyers could not get mortgages.
Hence it is necessary to obtain information directly. This is what the two studies
reviewed next attempted.
~
Benston and Horsky [1979] employed survey methods to determine
whether the demand for mortgages and home improvement loans by owners and
potential buyers of homes in the allegedly redlined area of the city had been met
and whether home buyers received the type of mortgages and service they
wanted. The allegedly redlined area was defined by community group leaders
and other experts; essentially it is the central city of Rochester, New York. A
suburb, suggested by these experts, was chosen as a control area. Because poten-
tial home buyers cannot be identified and located, home owners were inter-
viewed, since a potential home buyer necessarily would have to contact the
owner before attempting to get a mortgage. The home owners in both areas re-
ported that none of the problems they encountered in selling their homes over
the past five years were due to difficulties encountered by prospective buyers
in getting mortgages because of the location of the house. However, three times
as many central city as suburban home owners (3 6 percent vs. 13 percent)
said that no potential buyer had inspected their homes that were offered for
sale. Of the home buyers interviewed, all but 0.4 percent (one person) who
bought homes in the central city and wanted a mortgage, got one. Mortgages
were obtained predominantly at the first lending institution contacted (84 per-
cent of the central city and 89 percent of the suburban mortgagors). The data
also indicate that very few people got FHA insured or conventional mortgages
for reasons other than their choice. Nor did home owners in either area experi-
ence difficulty in obtaining home improvement loans.
See the appendix for more detailed summaries.
162
Tyree and Yeager [ 1979] interviewed all the cormnunity leaders in St. Louis,
Missouri who could be contacted. Many indicated dissatisfactions with real
estate lending institutions, which they considered to be biased against their areas.
The leaders also spoke of peoples experiences with redlining. When pressed to
identify specific people, repeated requests and consultations yielded but seven
names. Of these two could not be found, four denied they had the problem,
and the seventh appeared to be a case of lack of creditworthiness.
The two studies reviewed used different methods for determining the extent
to which the demand for mortgages from allegedly redlined areas was not met.
Both provide evidence that this demand was not present and that allegations of
redlining were based on misperceptions. Their principal limitation is their Use of
stated experiences and opinions rather than the hard data that economists
prefer (which often are obtained by someone else from interviews). This limita-
tion is particularly important for the Benston-Horsky method of asking home
owners about the problems encountered by potential buyers of their homes.
Unfortunately, to my knowledge, no one has designed a better method.
Summary and Implications for Further Research
It seems clear that the studies of the supply of mortgages to specific areas
and groups by subsets of lenders are of very little value. They cannot be used to
determine whether potential borrowers are well served or discriminated against.
At the least, data also are required on mortgages made by other lenders. The
role played by such usually omitted sources as mortgage bankers and private
funds is revealed by Dingemans [1979] and Koebel [1979]. The demand by
home buyers for mortgages is an even more important factor that must be
considered. Only the Ostas, Reed and Hutchinson [1979] study employed a
possibly valid measure of demand - houses for sale as reported in multiple
real estate listings. But this approach would be useful only if the activities of all
lenders in the area were included. And even then, one could not know whether
private or any specific type of financing was used because it was preferred or
the only alternative available. Schafers [1978, Chapter 5] procedure of esti-
mating a supply function for areas acknowledged as nonredlined and applying
the coefficients estimated to determine the expected supply to allegedly red-
lined areas offers some promise. But unless all lenders are included, it cannot
begin to speak to the hypothesis that potential borrowers are disadvantaged.
Nor can the quality of the activities of a particular lender or group of lenders
be determined without knowledge of demand and of the past lending behavior
of the institution(s) in question. Therefore, I can see very little, if any, research
value in studies that use data such as those made available by the Home Mort-
gage Disclosure Act.
Studies of the relative risk (actually cost) of lending on properties in alleged-
ly redlined areas or to allegedly discriminated-against persons could be useful for
determining whether the terms charged are based on economic or noneconomic
factors. But data on loans made in the ordinary course of business are unlikely
to provide these estirr~ates because lenders can be expected to avoid loans that
offer more than a minimal expectation of default. Rather, studies should use
163
essentially unscreened or differently screened data, such as the experience from
the FHAs programs to make funds available to borrowers and on properties that
do not meet ordinary underwriting standards. High-risk lending programs, such
as the St. Louis Savings Service Corporation whose denial decisions are analyzed
by Tyree and Yeager [1979, Chapter 9], are another source of such data. Only
the Marcis and Hull [1975] study used such data; they found significantly higher
foreclosures in zip code areas that are characterized by female heads of house-
holds, low income and education, and greater instability of occupancy. Barth,
Cordes and Yezer [1979] used regular program (203 b) FHA insured mortgage
data, which are not screened as completely as are conventional mortgages. They
found that mortgages made to blacks had almost twice the default rate of the
average mortgage, cet. par. Considering that the other studies reviewed used
data that had been screened by lending officers, it is interesting to note, never-
theless, that they found significantly higher foreclosure rates on mortgages made
on properties located in possible or allegedly redlined areas and to allegedly
discriminated-against persons.
Comparison of the terms of mortgages between areas and persons offers a
better opportunity to determine whether or not redlining is practiced. If it is,
the interest rate, downpayment requirement and/or term to maturity should be
harsher for the discriminated-against mortgagors than for others, cet. par.
Schafers [1978, Chapter 6] study used simultaneous equations; Muths [1979]
study used OLS; Benston, Horsky and Weingartners [1978] study used a loan
offer function estimated for a presumably favored area and then applied to an
allegedly redlined area; and King [1979B] compared requested with granted
terms and included an analysis of fees. The first study used data from a subset
of New York City state-chartered banks, the second Oakland, California state-
chartered savings and loan associations, the third all the major institutional lend-
ers (including mortgage bankers) in Rochester, New York, and the fourth savings
and loan associations in Miami, San Antonio, and Toledo. The four studies
reviewed report that mortgage terms do not differ much (if at all) between
mortgages made on properties in allegedly redlined and other areas and by
blacks, Hispanics, women, older people or unmarried people. Because higher
prices (terms) charged to borrowers provides a meaningful measurement of
the presence of discrimination, given that the effect of economic factors is
accounted for, the extension of such studies to other areas would be useful.
The allegation that lenders systematically underappraise property located in
certain areas or purchased by minority buyers also was studied. Schafer [1978,
Chapters 4 and 9] studied appraisals on mortgages granted in New York City
and in four upstate New York cities and Benston, Horsky and Weingartner
[1978] analyzed appraisals in Rochester, New York. King [1979B] included
denied applications in his study of Miami, San Antonio, and Toledo. All the
studies find evidence contrary to the underappraisal allegation.
However, analyses of mortgage terms can reflect only mortgages made.
Therefore, studies of denials of mortgage applications can provide insight into
the possible existence of redlining. Four of the studies reviewed report some evi-
dence of racial discrimination. Black, Schweitzer and Mandel [1978], using
nationwide data in a multiple variable probit analysis, found some weak evidence
that blacks experienced higher rates of denials, cet. par. Schafer [1978, Chapter
164
7] found strong evidence of this situation in New York City mutual savings
banks. His studies of mutual savings banks in four upstate New York cities
[Schafer, 1978, Chapter 12] also reports similarly strong evidence, but three of
these studies are based on very few observations. All of his studies found little
or no evidence of higher denials related to applications for mortgages on prop-
erties located in allegedly redlined neighborhoods. However, his analyses suffer
from possible collinearity and other serious problems. King [1979B] found
evidence of higher denial rates to Hispanics in Miami and San Antonio but no
evidence of adverse selection of blacks or other persons or applicants for !oans
on properties located in possibly redlined areas. Warner and Ingrams [1979]
multiple discriminant analysis of most of Columbia, South Carolina banks and
savings and loan associations found no evidence of discrimination. Nor did
Tyree [1979]. Hence, I conclude that the weight of the evidence is contrary to
the hypothesis that lenders discriminate against minority borrowers or areas by
denying mortgage applications, but there is also some reason to believe the con-
trary with respect to some minority borrowers. Clearly, more research on mort-
gage denials is desirable.
Since lenders might redline by not permitting mortgage seekers even to
file applications, studies of the demand for mortgages and of the experiences of
possibly discriminated-against home buyers in obtaining financing are necessary.
Only one study, by Benston and Horsky [1979], addressed this question. Their
interview survey of home owners and home buyers in Rochester, New York pro-
vides evidence that mortgage seekers in an allegedly redlined area experienced
almost no difficulty in obtaining funds because of the properties location. (The
sample size, however, was small and the information "second hand.") Home
buyers in that area and in a presumably favored (control) area equally had
little trouble obtaining the type of financing they wanted. Tyree [1978] polled
community organization leaders for their beliefs about bank lending practices
in their areas. Despite their often negative attitudes and assertions about the
prevalence of redlining, they were able to identify only seven instances, none
of which was found to have occurred as claimed. More studies like these are
needed, however, before one can generalize.
On the whole, then, only two studies of the research reviewed (all that I
could find) reveals evidence of possible redlining, and this is only in a higher
rate of denial of mortgage applications by blacks or Hispanics. But the second
studys findings are not consistent. Another, very well constructed study finds
no evidence of bias. Furthermore, foreclosures appear to be more probable in
areas characterized as redlined and on mortgages made to blacks. But "black,"
or "Hispanic" also might be proxy variables for omitted economic factors in
the analyses of mortgage denials and foreclosures. Virtually all of the other
studies either do not speak meaningfully to the redlining question or provide
evidence that is contrary to the redlining hypothesis. The question is still open,
however, since this conclusion is based on data from only a few cities. More
research on the demand for mortgages, in particular, would be helpful. Studies
of why such strong legislation as the Mortgage Disclosure Act of 1975, the
Community Reinvestment Act of 1977, state laws and state and federal regula-
tions was enacted on the basis of such poor evidence as is reviewed above also
would be interesting.
REDLINING RESEARCH
BENSTON
165
APPENDIX
Review of Studies Referred to in the Text
Community Group Studies of Mortgage Supply
The following three studies, two of which were presented to the U.S. Senate Commit-
tee on Banking, Housing and Urban Affairs hearings in May, 1975 are representative of the
evidence presented by community activists that apparently was instrumental in inflnencing
the Congress to enact the Home Mortgage Disclosure Act of 1975.
21
Redlining and Disinvestment in Buffalo [ 1975 ] is typical of many community activist
studies. It was based solely on the number of dollars of mortgages made by commercial
banks, savings banks and savings and loan associations in Erie County over an 18-month
period (June 1973 through December 1974). Because fewer mortgages were made in the
city of Buffalo compared to its suburbs (26 percent in nmnber, 18 percent in dollars), the
authors concinde that the city was redlined. They further assert, without any reasoning or
evidence, that "The evidence has been amassed, pointing the way powerfully and clearly
to a new perspective on the root cause of blight: redlining and disinvestment. For the first
time in the City of Buffalo, the bineprint of these bank policies are hereby laid open for
public scrutiny." [p. 7 ]
Where the Money Is: Mortgage Lending, Los Angeles County [1975 ] presents a more
extensive analysis, since it presents data on downpayments and interest rates as well as the
number and dollar amount of mortgages, aggregated by census tract. Mortgages on other
than single-family houses were included. However, only mortgages made over a five-month
period (January through May 1974) by state-chartered savings and loan associations were
included. Charts were used to compare mortgage lending terms and activity in the first five
months of 1974 with the 1970 values of percent minority population, median family
income and median home value in census tracts. No other variables were included to ac-
count for differences in mortgage terms, numbers or amounts. Nor was the possible relation-
ship between income, price of house and minority considered. Despite these shortcomings
and the limited coverage of the data, the report concludes: "patterns of discrimination are
shown [and] ... over a million people live in neighborhoods where there is no prospect of
lending at all." [p. 5 ]
Home Ownership and the Baltimore Mortgage Market [ 1975 ] employed by far the best
data base of any of the studies presented at the hearings. Included were all Baltimore real
estate transactions in 1971, and thus mortgage loans made by all lenders. The study finds
that different types of lenders tend to specialize in different types of neighborhoods. For
example, savings and loan associations and savings banks made relatively more loans in high
income, high house price, high home ownership, white areas, while mortgage companies that
made FHA insured loaus and cash sales dominated the low income, low house price, black
areas. The study suggests that FHA insured mortgages are undesirable, primarily because
points are charged. However, neither the effective interest rate paid (the annual interest
equivalent of points plus interest payments) nor the relatively low downpayment required
were considered. In any event, the study essentially is limited to description: redlining and
racial discrimination are not charged.
A nmnber of other studies were presented at the 1977 Hearings on Community Credit
Needs that preceded enactment of the Community Reinvestment Act of 1977. The follow-
ing two studies are representative of the papers presented there and elsewhere by commu-
nity activists.
Take the Money and Run! Redlining in Brooklyn [1976] was prepared by the New
York Public Interest Research Group. They analyzed mortgages made in 1975 by seven
savings banks according to zip code and census tract areas. The number of mortgages made
in one year were related to the number of owner-occupied houses in the areas and the dollar
amounts of mortgages made were compared to the total assets of the savings banks. From
21 See U.S. Senate Committee on Banking, Housing and Urban Affairs. Hearings on the
Home Mortgage Disclosure Act, 1975.
these relationships, the authors concluded that all except one of the banks studied redlined
Brooklyn. The distribution of mortgages made by zip codes is used to suggest that "black
neighborhoods fare worse than white neighborhoods." [p. 5] This conclusion is illustrated
by reference to four census tracts. Two tracts that are mostly black received zero and three
mortgages while two tracts that are mostly white received the highest number and relatively
many (65) mortgages. From these data, the study concludes: "All seven of the banks sur-
veyed have discriminated against the black population of Brooklyn by channeling mortgage
money to those neighborhoods that are predominately white." [p. 7] The study does not
mention that the two predominantly black census tracts cited contain mostly multifamily
buildings while the predominantly white tracts contain mostly single-family houses. Nor is
mention made of other lenders (including the omission of two major savings banks) nor is
any consideration given for the demand for mortgages.
Redlined." A National Survey by National Peoples Action [NPA} of Mortgage Lending
Policies in the United States [1976] is perhaps the most extensive study in terms of cover-
age of cities. Data made available by the Home Mortgage Disclosure Act of 1975 from
financial institutions in 25 cities in many states for the year ended September 3 0, 1976
were used. Specifically, dollars of conventional mortgages plus home improvement loans
made by two institutions in two neighborhoods (selected by the NPA) in each city are
presented in tables. Nothing is said about possible differences in the demand for mortgages
or of other sources of supply. Nevertheless, based only on these tables, the authors in-
correctly conclude: "In every case the data supports the existence of redlining - the refusal
to make conventional loans."
Professional Stndies of Mortgage Supply
Devine [1973] (Bronx, New York}. Where the Lender Looks First." A Case Study of
Mortgage Disinvestment in Bronx County, 1967-70 is important because of its historical
role (particularly in the hearings preceding enactment of the Home Mortgage Disclosure
Act of 1975). This monograph (a doctoral dissertation which was prepared for and pub-
lished by the National Urban League) presents a discussion of the possible relationship
between lending by financial institutions and community decline, with particular emphasis
on racial discrimination. The empirical work consists of multiple regressions relating the
number and dollar amounts of mortgages made per 1,000 occupied housing units in 1960
and 1970 and the change between these dates in 14 Community Planning Districts. The ex-
planatory variables are the proportion of blacks and Puerto Ricans (race, the proportion of
one-to-four family homes, and total or median rent in each district). (Other variables were
tried but the results are not presented.) The data were obtained from 12 savings banks:
these banks made less than 22 percent of the number and less than 18 percent of the
amount of mortgages in the Bronx in the years studied. Changes in the amount of mortgages
made in the Bronx by subsets of the institutions studied also were compared to changes in
their total savings balances, a curious comparison of a short-term flow with a stock.
The regressions show race not to be a statistically significant explanatory variable, with
two exceptions. In the 1970 "number of mortgages per 1,000 occupied units" regression,
the coefficient of race is negative and statistically significant when total rent was substituted
for median rent and when the proportion of one-to-four family homes variable was omitted.
In the 1970 "amount of mortgages per 1,000 occupied units" regression the (negative) co-
efficient of race is significant when median rent was used instead of total rent, one institu-
tions mortgages in a district were excluded, and the proportion of one-to-four family homes
variable was omitted. Nevertheless, after qualifying his conclusions ("Our statistical analysis
is neither definitive, nor exhaustive. It does not suggest that race is the only or primary vari-
able being considered by mortgage lenders."), Devine concludes: "It does, however, disprove
the assertion that race has no significant bearing on mortgage lending." [p. xii] But even
this limited conclusion is not supported by the regressions. Nor can any meaningful conclu-
sions be drawn from this study, considering the quality of the data (the community plan-
ning districts are very heterogeneous), the lack of a model or other rationale for the inde-
167
pendent variables, the small fraction of the mortgage lenders included, and the omission of
demand and risk considerations.
~ ~
Ahlbrandt [1977] (Pittsburgh). Ahlbrandt [1977] estimated an expanded version of
equation (5) (in the text above) with data from 162 census tracts in Pittsburgh. (Tracts
dominated by pub/it housing, business or industry were excluded.) The dependent variables
used were the total value and number of residential mortgages made by financial institutions
in 1973 and 1974, adjusted for the number of occupied units~ Financing by individuals, real
estate firms or other organizations or institutions (not otherwise described) was excluded.
The independent variables included median family income in 1969, percent of units owner-
occupied in 1970, percent of units vacant in 1970, change in percent of black population
1960-70, percent of black population in 1970, and crime rate in 1970. Because an inter-
active relationship was assumed, the data were transformed to logarithms. Three sets of
regressions are presented: the first with all the data (13 5 observations), the second with
census tracts having less than 20 percent black population (100 observations) and the third
with census tracts having more than 20 percent black population (28 observations). One
variable that possibly might indicate redtining (were demand accounted for and the equation
well-specified), the change in percent of black population 1960-70, had statistically signifi-
cant coefficients only ha regressions run with the third set of data - census tracts with
black population over 20 percent. Ahlbrandt concludes: "although racial concentration is
not statistically significant and racial change is positively associated to mortgage lending,
this does not conclusively refute the hypothesis that racial redlining exists." [p. 479]
Although his study is considerably better than the community groups and Devines
study, it still is seriously deficient. In addition to the factors mentioned in the text above
(in particular, a reduced-form, equilibrium equation cannot measure shortfalls in demand),
the independent variables are likely to be collinear. Family income is likely to be negatively
correlated with percent black, while owner-occupied, percent units vacant and crime rate
are likely to be positively associated with percent black. Ahtbrandt also warns: "A problem
could arise in the interpretation of the results ff there is not a reasonably good correspon-
dence between the characteristics of the mortgage applicant and those of the census tract
in which the loan is made (or turned down) .... [In particular], since the race of the appli-
cant is unknown, it is not possible to conclude the degree of racial discrimination against
mortgage applicants (sic). The interpretation of race in the regression analysis is thus limited
to the racial characteristics of the areas in which the loans are granted." [p. 480] But, con-
sidering the other problems with the study, even this interpretation is not warranted.
Hutchinson, Ostas and Reed (HOR) [1977] (Toledo). Reduced-form equations similar
to, but more elaborate than, Ahlbrandts were used by HOR to estimate the determinants
of commercial and government-insured mortgage flows and home improvement loans. They
used data provided by "the four savings and loan associations which dominate the mortgage
loan market in the Toledo, Ohio SMSA." [p. 464] Loan data for 1975 aggregated by 123
census tracts were regressed on independent variables derived from 1970 census data.
The dependent variables are the total number of mortgages on owner-occupied houses,
the percentage of conventional mortgages to the total, the number of conventional mort-
gages, the number of government-insured mortgages, the number of home improvement
loans, and the percent of home improvement loan applications accepted, all in 1975. The
independent variables are the percent black population (B), its square (B2), the change in
B from 1960 to 1970, the average age of residential structures, the unemployment rate,
median years of education, median income, number of owner-occupied one-to-four family
dwellings, median property value, percent of population over 55 years of age, average dura-
tion of residency, percent of population having moved into the tract from 1968-1970, and
2~ Furthermore, the conclusions given in the introduction to the study, on the impor-
tance of "the active presence of institutional lenders such as savings banks and savings and
loan associations in a housing market," have been cited in community activist studi, es and in
Congressional hearings as if they were derived from Devines data, even though the~ are but
assertions that are not otherwise addressed in the monograph.
168
the number of persons per household, all as of 1970 (except thevariables lneasuring changes,
as noted).
The coefficients of the independent variables representing race (percent black, B, and
its square) are statistically significant only for the percent conventional (negative B, positive
B
2),
number of insured mortgages (positive B, negative B
2),
and percent of home improve-
ment loan applications accepted (negative B, positive B~). HOR conclude that "the racial
composition of a neighborhood has no significant effect on the total number of loans made
in the neighborhood" [pp. 468-469] but "racial composition affects the percentage of
loans transacted that are conventional (PC) and the percentage of home improvement loan
applications that are accepted (PA). These results suggest that PC and PA are both signifi-
cantly lower in racially mixed neighborhoods than in racially homogeneous neighborhoods."
[p. 469]. However, the interaction of B and B
~
result in estimated minimum values for PC
at 45 percent black and PA at 28 percent black. HOR state: "These results are consistent
with the hypothesis that redlining takes place on the basis of risk aversion. However, they
would not be consistent with a redlining model based on a taste for discrimination. In a
model such as this, PC and PA would be expected to decrease as B increases." [p. 469].
The coefficients of the average age of structures variables have significantly negative signs
for all except the number of insured mortgages and number of home improvement loans
regressions, which, HOR state, "strongly support the hypothesis of redlining on the basis
of neighborhood age in the conventional and home improvement loan markets." [p. 470].
Ostas, Reed and Hutchinson (ORH) [1979] (Toledo). The data for Toledo were ex-
panded and updated in another paper by Ostas, Reed and Hutchinson. They re-estimated
the relationship between the number of mortgages made and percentage of conventional
mortgages made and percentage of conventional mortgages made by savings and loan asso-
ciations in 1975 and 1977, excluding a few of the independent variables used in their
previous paper,
a3
For the number of mortgages regression, the percentage of black in 1970,
B, and B
~
variables are statistically insignificant in 1977 as in 1975. The average age of
house variable still is significantly negative, though its magnitude is -1.03 9 in the 1977
regression and -.008 in the 1975 regression. The greatest difference with respect to the
"redlining variables" is the insignificance of the "black" (B and B
~)
variables in the 1977
regression. ORH conclude: "In general, the equation appears unstable between the two
periods, and suggests little evidence of conventional loan restriction on the basis of racial
considerations." [p. 18] However, the average age of house variable is significantly negative
and of similar magnitudes in both years regressions.
The new version of their study was improved over the previous one in three important
regards: commercial banks were included (presumably all those in the Toledo SMSA); two
new variables, the number of one-to-four family units shown in real estate brokers multiple
listing guides (NMLS) and the number of one-to-four family houses constructed in 1977
were included to replace the number of owner-occupied one-to-four family dwellings as a
proxy for demand, and dummy variables were added to estimate more complex forms of
the relationship between the percentage black and the dependent variables. The new vari-
ables (NMLS in particular) improved the fit of the equations considerably. The regressions
presented show no statistically significant relationship between the commercial bank data
and the "redlining" variables - percentage black in its various forms and average age of
house. However, significant coefficients are reported for the savings and loan data. With
respect to the percentage black variables, ORH state: "racial composition has a significant
effect on the distribution of the total loans and S&L loans. Often this effect conforms to
a risk aversion hypothesis based on a "tipping point" argument. However, it should be
noted that this result is not apparent unless NMLS data is incorporated into the model ....
In 1977 race appears to have a negative effect on the percentage of loans which are con-
ventional." [p. 25] Finally, "age of structure is observed to have a negative impact on the
percentage of loans which are conventional, but this effect is complicated by an apparent
significant quadratic relationship." [p. 26] It also is negatively related to the number of
~3 The percent of population over 55 years of age and the percent of population having
moved into the tract from 1968-1970 were excluded.
REDLINING RESEARCH
BENSTON
169
S&L loans. Calculation of the magnitudes of the possible effects (e.g., the decrease in the
number of loans associated with an increase in the percentage black over some range) is
not given in the paper.
The first Toledo paper (HOR) suffers from the same shortcomings as Ahlbrandts
Pittsburgh analysis, in that 1970 data were used to explain 1975 lending activity, demand
for mortgages was not accounted for and percent black, average age of houses and many
of the other independent variables are likely to be correlated. In addition, by including
data only from savings and !oan associations, at most all that can be considered is the lend-
ing activity of these institutions; the availability of mortgage funds to home buyers cannot
be estimated from these data. The second Toledo paper (ORH) presents a considerably
improved variable that might serve to measure demand - the number of multiple listings
(NMLS). This variable is far preferable to the number of owner-occupied dwellings, since
there is no reason to believe that the same percentage of houses in all areas are likely to
be sold.
~4
Inclusion of the number of newly constructed dwellings also improves the speci-
fication of demand. The nonlinear forms of the "percent black" variable also are positive
aspects of the analysis. However, the reduced-form equations do not permit one to deter-
mine whether fewer mortgages were made in census tracts because of racial considerations
or because of differences in demand for mortgages available from other lenders (such as
mortgage bankers and individuals), or whether race is a surrogate for other variables not
related to discrimination (as ORH suspect).
Hauser and Elkhanialy [1978] (Chicago). The lending activities of 22 of the 3 2 federal
savings and loan associations (or 86 percent of total assets)in Chicago for 1977, aggregated
by 862 census tracts, were related to 1970 census data on percent black and other variables.
When the tracts were grouped according to predominantly white (0 to 9 percent black),
racially mixed (10 to 74 percent black), and predominantly black (75 to 100 percent black),
the average number and total amount 0"f loans made clearly are negatively related to the
percentage black. The average size of mortgages is lower as the percentage black is higher.
A correlation analysis reveals the following relationships with the number of loans per
tract: median family income, .46; per centage black, -. 3 1 ; median value of home, .16; per-
cent owner-occupied, .11; built before 1949, .05; and living in the same house, .02. How-
ever, percent black is strongly correlated with some of the other variables, as follows:
median family income, -.45; owner-occupied, -.13 ; and median value of home, -.11. In a
partial correlation, the correlate between the number of loans made. and percentage black
drops from -.3 1 to .12. Similarly, the partial correlatio.n coefficient of percent black with
the dollar amount of loans is -.07 compared to -.29 for the simple correlation. The
authors also emphasize that the percent of variance jointly explained by the seven variables
(R
~)
is only 24 and 27. They state: "It is clear, then, that variables other than those con-
sidered are involved. Such variables, for which data are not available, undoubtedly include
risk of investment, differential market forces, laws and regulations, etc." [p. 3 1]
The study is valuable primarily because it points out that, in the authors words, "to
conclude that the differential lending practices are interpretable as racially discriminatory
is simplistic. In the fight of the complexities involved, it is sounder to conclude that the
disadvantaged position of blacks with respect to home ownership and financing is among
the indications of the disadvantaged economic and social condition of blacks." [p. 16] But
the study fails to consider the possibility that differences in demand for single-family house
mortgages among census tracts and alternative sources of supply (such as mortgage bankers
and commercial banks) might account for (or exacerbate) the differences found. Also not
mentioned are risk considerations which are particularly important in Illinois. The state
foreclosure law is the most severe in the country. It requires foreclosure by sale in a three-
step judicial procedure: filing a complaint, entry of a decree or judgment by the court, and
the foreclosure sale. The time between the date a mortgagor stops payments and the lender
can acquire the property was found to average 816 days.
~s
This compares to 3 98 days in
~4Benston and Horsky [1979, p. 76] found an average home ownership duration of 26
years in the central city and 14 years in a suburb of Rochester, New York.
zs Touche, Ross & Company, 1975, Table 1.
170
California and 261 days in Texas. The direct cost less revenue per loan averaged $6,03 1 in
Illinois, $61 in California and $25 in Texas.
26
Thus, the risk of default is quite a serious
consideration in Chicago, particularly for conventional mortgages.
Dingemans [1979] (Sacramento). Further evidence on the association between mort-
gage lending patterns and minority status is provided in this very good article, in which the
author shows how misleading impressions can be drawn from this correlation.
27
Dingemans
recorded data on 8124 single-family homes made in 1976 by the 50 banks and savings and
loan associations that serviced the areas, by census tract. (Loans made by mortgage com-
panies were not included, since they are not required by the Home Mortgage Disclosure Act
of 1975 or by state regulations to report their activity.) The number of mortgages made per
existing single-family housing unit, aggregated into 114 census tracts, were correlated with
13 variables that measured tract characteristics, mostly drawn from a special census con-
ducted during 1975. Statistically significant (at the .01 level) correlates with the number of
mortgages made are reported for the following variables: upper status employment (.74),
Chicano residents (.73 ), median household income (.72), blue collar employment (-.68),
distance to minority tracts (.64), black residents (-.60), nonminority residents (.60),
income change, 1970-75 (.57), age of housing stock (-.51), distance to central business
district (.44), and percentage of owned housing units (.3 5). The only statistically insignifi-
cant variables are mobility 1968-70 (-.18) and average household size (.08). Similar cor-
relates were found when these variables were correlated with the number of loans made
per owned unit. Loans on used ho~nes only (11 census tracts that received more than 50
new home mortgages, also were excluded) are significantly correlated only with blue collar
employment (-.68), Chicano residents ~-.47), median household income (.41) and black
residents (-.3 9).
However, Dingemans points out that these correlates can be very misleading. When tile
number of mortgages made per unit is regressed, stepwise, on the variables, he finds that
"sixty percent of the variance is explained, with the ~najor contribution resulting fiom just
one variable, median household income .... [The] other significant (.05 level) variables
(in order of entry) were household size, distance to CBD [central business district], and no
minority population; none of these added more than seven percent to explained variation."
[p. 23 1] When mortgages on used houses were regressed on the variables, the first variable
to enter the regression was upper status income, which explained 55 percent of the vari-
ance, followed by age of housing, household income, and household size. None of the other
variables added more than 5 percent to the 63 percent of the variance explained, [p. 23 2].
Thus, mortgage lending patterns by financial institutions appear to be related primarily to
the income status of an area, which (as Hauser and Elkhanialy also found) can be confused
with the areas minority population.
~8
Dingemans attempted to separate the joint influence of income, minority, and loca-
tional variables with case studies of four neighborhoods. In 1976, the inner city neighbor-
hood (Old City) received 2.! loans per single-family unit compared to a region-wide rate for
used homes of 2.8 percent. Between 1976 and 1977 the rate increased sharply. Few appli-
cants in 1977 (the first year that these data were available) were turned down (23 ), though
the rate was twice as high as the regional average. An analysis of lending in areas that in-
cluded a relatively high concentration of minority (black and Chicano) residents revealed
lower loans per unit, but few rejected loan applications (in 1977). A comparison of a blue
collar, middle income suburb with a white collar, upper income shows a much lower number
of loans per unit in the former than the latter. More detailed analysis reveals that mortgage
bankers are very active lenders in the blue collar and minority areas. Dingemans explains:
~8Ibid., Tables 8, 9, and 10. The California net costs were reduced considerably by the
increase in market value of the foreclosed properties.
~7 An excellent introduction and overview of writings and studies on mortgage lending
and urban change is provided in the second section of the paper.
:SSimilar analyses are pre.sented for home improvement loans. In that regression, dis-
tance to the CBD entered the ~egression first.
REDLINING RESEARCH
BENSTON
171
"Most homes in Sacramento are bought with the aid of a real estate agent. One of the ser-
vices that the agent normally performs (in return for the fee...) is help in securing a mort-
gage loan .... The agents of mortgage bankers [who are paid commissions rather than the
salaries paid to bank employees] are. extremely aggressive in maintaining ties with sales
persons who work in... moderate income neighborhoods. This is because mortgage bankers
know that moderate income neighborhoods are the most likely locations for FHA and VA
loans, and they issue primarily the federally insured loans for which a ready secondary
mortgage market exists." [p. 23 8]. His final conclusion also deserves quotation: "The lack
of data for some major loan sources and the fact that nearly identical patterns [of Home
Mortgage Disclosure Act data] are found in Sacramento - a metropolitan area where few
complaints about redlining are heard - should act as reminders that the examination of
Disclosure Act data above 1nay not be a sufficient basis for making final conclusions about
the processes and behaviors that underlie the patterns that are being found." [p. 23 9] Based
on his analysis, he should have said, "is not" rather than "may not" be sufficient.
Whalen [1976] (l~int). "An Analysis of Mortgage Lending Activity in Flint Michigan"
represents an interesting attempt to analyze the effect of race on mortgage lending after
account was taken of other explanatory variables. It was designed and conducted by the
Chief of Informational Services at the Michigan State Housing Development Authority for
the Governors Task Force on Redlining. The number and amounts of mortgages recorded in
47 census tracts over the 18 months from January 1975 through June 1976 were analyzed
with respect to the following: prudent lending variables (an occupational income index as
a measure of creditworthiness, the percentage of homes reported in a local survey to be well
maintained as a measure of the structural condition of housing, and effective demand as
measured by owner-occupant mobility and the net change in the number of households),
and redlirting variables (racial composition as measured by the percent Negro in the 1970
census, racial change as measured by the change in percent black from 1970 to 1975 from
a special survey, proximity to racial change as measured by the average racial change in ad-
joining census tracts, and the relative age of houses from the 1970 census). I believe that
two-stage regressions were computed (the text is not clear), with the dependent variables
regressed first on the "prudent lending variables" and the residuals regressed on the "red-
lining variables."
The coverage of the data, the model and method of analysis, and the measurement of
the independent variables are greatly superior to many of the studies reviewed above. The
only problems are the dependent variables. The text states them to be the number amount
and average amount of mortgages made per census tract. From personal inquiry I learned
that most of the homes mortgaged were financed with FHA insured mortgages and the
dollars (amounts and averages) are the selling prices of the houses, since these almost equal
the mortgage amounts but were more readily available. Therefore, the interpretation of
the findings is not what it appears to be. In fact, none of the coefficients of the inde-
pendent variables in the total dollars of mortgages made regression are statistically signifi-
cant. Only the percentage of housing built 1950-59 in the number of loans made regression
is statistically significant, and the percent black in 1975 and racial change variables in the
incorrectly described average mortgage amount regression are statistically significant.
29
But
since the average mortgage really is average selling price, the regression would appear to
indicate that lower house prices, not redlining, were associated with race. In addition, the
demand for mortgages is not accounted for.
Schafer [1978, Chapter 5] (New York City). Neighborhoods in three counties in New
York City, the Bronx, Kings (Brooklyn) and Queens, were studied. The data used were ob-
tained from reports filed by a subset of state-chartered financial institutions pursuant to a
regulation imposed by the New York State Banking Department. Schafer reports, "the
lenders in the sample supplied only 13 .9 percent of the dollar value of transactions in 1975."
[p. 5-10]. Consequently the findings can be interpreted, at most, as applying only to those
institutions included and not to the overall borrowing experience of property buyers in the
three New York city counties considered.
3
Standard errors are not given; only the designation "significant at the .05 level."
The model could not be applied to other areas because of data limitations.
172
Neighborhoods in the three counties were dichotomized into those alleged by commu-
nity groups to be redlined and others (a procedure similar to that used by Benston, Horsky
and Weingartner [t978]). A regression was computed for each area (redlined and other)
separately. Each observation is a census tract: 110 in the alleged redlined area and 640 in
the other. Conventional mortgage lending (dollars) per existing one-to-four family buildings
in allegedly redlined and in other neighborhoods are the dependent variables. The 3 3 inde-
pendent variables include: transactions per building (dollars), mortgage prices on conven-
t(~nal mortgages (3 variables - interest rates, maturity, and loan-to-value ratio), the stock
of mortgages held by the banks per building (4 variables - conventional and federally
insured, one-to-four family and multifamlly), neighborhood attributes related to risk of loss
(20 variables, including pending and past housing code violations, building vacancies, tax
arrearages, structural fires, per capita welfare and change in population), risk of loss in mort-
gage lending (2 variables - ratios of foreclosure and delinquencies to total loans), percentage
of housing stock built before 193 9, and racial composition (percent nonwhite in 1974 and
change in percent nonwhite, 1974 less 1970).
The computed R~s are .68 for the alleged redllned sample and .25 for the other neigh-
borhoods sample. Few of the coefficients are statistically significant, which is not surprising
considering the great likelihood of collinearity. However, it is interesting to note that the
coefficients of the "transactions" variable (a proxy for demand) is significant (at less than
the .01 level) and is five times greater for the alleged redlined than the other sample. Also,
the stock of conventional one-to-four family mortgages is significant (at less than the .01
level) for the alleged redlined sample but trivial and very insignificant for the other sample.
The coefficients from the alleged redlined census regression were multiplied by the
values of the independent variables in each census tract and then aggregated according to
neighborhood.
3 1
The same procedure was followed for the coefficients from the other
neighborhoods regression. The aggregates then were used to compute ratios, by neighbor-
hood, of predicted lending from the alleged redllned regression to predicted lending from
the other neighborhoods regression. In two of the seven alleged redlined neighborhoods
the ratio is greater than a one, indicating, as Schafer put it, "that these neighborhoods
would receive more funds if they were redlined than if they were not." [p. 5-29] (The
aggregate ratio is .67.) In two of the four other neighborhoods the ratios are slightly greater
than one, indicating again that they would receive more funds if they were redlined than
if
~
they were not, cet. par. (The aggregate ratio is 1.02.) Because the two areas with ratios
greater than one may have been incorrectly identified as redlined, data from these areas
(3 4 tracts) were removed and the alleged redlined regression rerun. The ratios computed
with the resultant coefficients are .below one for all neighborhoods, aggregating .49 for the
alleged redlined neighborhoods and .56 for the other neighborhoods.
A similar exercise was undertaken with the amount of federally assisted (FHA and VA)
mortgages per building as the dependent variable. The coefficients of the demand proxy
variable (transactions per building) are statistically significant in both regressions, but twice
as large in the other neighborhoods regression. The ratio of aggregate predicted lending
(alleged redlined/other neighborhoods) is 1.28 for the alleged redlined areas and 1.20 for
the other areas. However, as was found in the conventional mortgage analysis, one neighbor-
hood
3 3
of the six alleged redlined areas has a ratio of 3 .70, which indicates that it got three
times more dollars of federally assisted loans because it was (allegedly) redlined. One of the
five other neighborhoods also has a high ratio (2.62). The regressions were recomputed with
observations from these neighborhoods omitted. The results were similar to those from the
complete data set (aggregate ratios are 1.3 6 for the redlined areas and 1.43 for the other
areas), except that the ratio for the questioned alleged redlined neighborhood is 8.90 and
for the other neighborhood is 2.68.
3 1 A similar procedure was used previously by Benston, Horsky and Weingartner [1978,
Chapter 2] in their analysis of mortgage terms.
3 2This one is a combination of the two problematic neighborhoods in the conventional
analysis.
REDLINING RESEARCH
BENSTON 173
Conventional multifamily mortgages were available for an insufficient number of
census tracts to allow application of the two regression technique. Therefore, Schafer
computed a single regression, specified similarly to the others, with dummy variables for
the census tracts in neighborhoods alleged to be redlined. As Schafer summarizes the find-
ings: "The age of housing stock, racial composition, and neighborhood variables are not
statistically significantly related to conventional multifamily mortgage lending at even the
10 percent level. The estimated model ... does not support allegations on redlining on
multifamily buildings." [p. 5-69]
Koebel [1978] (Louisville, Kentucky). This study is probably the most complete
record of the sources of house purchase financing yet produced (or perhaps producible).
All property sales and mortgages recorded in Jefferson County between June 1, 1976 and
May 3 1, 1977 were recorded, including those sales made with a contract for deed that were
listed in the Louisville Daily Record. Transfers that were not arms length sales were elimi-
nated. Among the data reported by area of the city and county are the types of property
(the 5 percent represented by multifamily houses are located in the city), the types of
purchasers and sellers of the properties (mostly individuals to individuals), the prices of
the houses sold (lower in the city), the source and type of financing, downpayments and
interest rates, the resident status of new owners, and the incidence of second mortgages,
all reported by area and, in many instances, by the price of house sold.
The study documents that some 3 6 percent of city properties were financed by indi-
viduals (cash and personal loans) compared to 14 percent in the rest of the county (here-
after called the suburbs). The median priced city home purchased in this manner sold for
less than $8,700. Kentuckys usury law specifies an interest rate of 8.5 percent on mortgages
of $15,000 or less; at that time the prevailing rate for other loans was between 8.5 and 8.75
percent. (Koebel failed to relate these facts causally.) The median prices of suburban houses
sold for cash or financed with personal loans was about $25,000, roughly equal to the
median prices of houses otherwise financed. In general, banks (which made very few mort-
gages) and savings and loan associations financed more expensive houses in areas with rising
prices and higher incomes, primarily with conventional loans. Mortgage companies tended to
finance the purchase of average priced houses in areas with declining prices and lower in-
comes, primarily with FHA and VA loans. They also tended to serve areas with predomi-
nantly black populations. Among the institutions, downpayments were lowest for mortgage
company loans, since they made mostly FHA and VA loans. Much more data are presented,
defying a concise summary. Most of these presentations are interesting primarily to someone
concerned with the Louisville area.
As Koebel puts it, "And although this study does not prove or disprove prejudicial dis-
investment (or redlining), it thoroughly documents housing prices and the availability of
financing throughout the county." [p. 9] Actually, it documents the incidence of financing,
since nothing is known about demand, except that one can infer that mortgages on low
priced houses would be demanded but not supplied when the state usury ceiling is below
opportunity cost (market interest rates and transactions costs). However, the complete
coverage provided does show the extent to which chartered financial institutions are not the
sole source of housing finance, for whatever reason.
The Relative Risk of Lending - Defaults and Foreclosures
33
Williams, Beranek and Kenkel [1973] (Pittsburgh). The data analyzed were obtained
from a savings and loan association. Data on all 125 defaulted mortgages made from 1962
3 ~All of the papers except Barth, Cordes and Yezer [1979] include analyses of delin-
quent mortgages. Since these loans generally do not represent a net cost to lenders, and
since the studies find that delinquencies are not highly correlated with foreclosures, the
delinquencies findings are not discussed here.
However, a study by Von Furstenberg and Green [1974] of 123 6 delinquent single-
family mortgages at a Pittsburgh savings and loan association should be mentioned because
174
to 1972 and a random sample of 1405 good loans were recorded.
3 4
In addition to variables
obtained from the loan files, data from the 1960 and 1970 censuses, by tract, were included
in the analysis to account for neighborhood factors. One stage least squares (OLS) analysis
was used, where the dependent variable is dichotomous, equal to 1 if a loan is a default and
0 if it is good. The independent variables included measure quality of the loan (loan to value
over 90 percent and 85-90 percent, term to maturity, and payment to income over 3 0 per-
cent, and 22-3 0 percent), type of loan (VA, FHA, FHA-23 5, refinanced, and loan with
junior financing, all in dummy variable form), characteristics of the borrower (age over 50
or below 3 0, number of years with employer, number of children, and monthly income -
price-level-adjusted), characteristics of the property (price - price-level-adjusted, age, and
whether or not multifamily), and neighborhood characteristics (unemployment rate, and per
capita changes in crimes against property). Other variables were excluded because of corre-
lations with included variables. For example, the following sample correlations with the
unemployment rate are reported: percentage black, .75; per capita crimes against persons,
.73 ; median per capita income, -.78. The correlation between median per capita income
and percent black is -.61. Thus the unemployment rate variable carries a lot of freight.
Statistically significant (.10 level) positive coefficients were found for the following
independent variables: loan-to-value ratio over 90 percent, payment-to-income over 3 0
percent, refinanced loan, junior financing, age of mortgagor over 50, and unemployment
rate. Significant negative coefficients include FHA insured loan, number of years with
employer, and price of the property. To the extent that one can draw inferences from the
significance of coefficients, considering the effect of collinearity on measures of signifi-
cance, the most important determinant of defaults is a high loan-to-value ratio.
3 s
The un-
employment rate variable is the only included measure of neighborhood characteristics -
it is positive and positively correlated with percent black and negatively correlated with
median per capita income. But in terms of the magnitudes of the t values, at 2.12 it is tied
for next to last place among the significant variables.
Morton [1975] {Connecticut). Data on 545 randomly sampled mortgages made by
equal numbers of commercial banks, mutual savings banks, and savings and loan associa-
tions (24 in all) were analyzed. Multiple discriminant analysis was used to identify the
variables associated with a loan being current, delinquent or foreclosed. The variables that
are significant in a function distinguishing loans as foreclosed or not foreclosed (current
and delinquent), in the order they entered stepwise calculations, are: three-family property
(generally large houses converted to rental units), junior financing, five or more dependents,
loan amount/appraisal, self-employed, employed as a salesman, borrower has nonreal estate
debt, four dependents, and no dependents. The signs of all but the last variable are positive.
of the superior methodology employed. They classified mortgages into cohorts that relate
the year a mortgage is made to the year in which it became delinquent. The percentage
delinquent in that year to the number in the year the loan was made originally is the
dependent variable. This is the annual delinquency rate. Each loan Was identified additionally
as being in one of five loan-to-value categories, in one of five borrowers income classes, a
loan in the center city area (Allegheny county) or in the suburbs, and a loan on a new or
previously occupied house. The interest rate in each of the 12 years over which the data are
drawn also was specified. A logarithmic multiple regression was run (R
~
= .11, statistically
significant), from which the authors draw the following conclusions: "mortgages on existing
homes are 43 percent more risky that those on new homes," significant at the .05 level; and
"Ceteris paribus, the riskiness of a mortgage declines by over 47 percent.., when it is on a
single-family home located outside rather than inside the center city county of Pittsburgh
[significant at the .01 level]. Since Allegheny county includes many high-class residential
areas and is by no means confined to deteriorating areas in the inner city, this effect is
stronger than expected." [p. 12].
3 4 Missing numbers are said to have limited severely the usable data.
3 SThis finding isconsistent with the conclusions of a large scale study of FHA-insured
loans by Von Furstenberg [1969].
175
Among the insignificant variables, only "one-family property" and "two-family property"
are related to area. The race of the borrower was not included. The function correctly
classified 74 percent of the original sample and 72 percent of a holdout sample.
Morton identifies the three-family property variable as "a proxy for property and/or
neighborhood quality .... Many of the three-family dwellings included in this study in-
volved large older homes, in declining neighborhoods, that have been subdivided into apart-
ments." [p. 74] Thus the data are consistent with the hypothesis that neighborhood quality
affects defaults.
Schafer [1978, Chapter 3] (New York City). Data from two sourceswere used in this
study. A subset of state-chartered commercial and mutual savings banks and savings and
loan associations in Kings (Brooklyn), Bronx, and Queens counties reported the number and
dollar values of their loans foreclosed in the past five years, by census tract, to the Banking
Department. Additional data were gathered with a special study of New York State mutual
savings banks in Kings, Bronx, Queens, and downstate suburban counties, excluding Man-
hattan. Approximately 100 loans were sampled from each bank: 25 active and current
loans, 25 active and delinquent loans, 25 foreclosed loans, and 25 satisfied loans.
The data are presented in tables listing the average foreclosure rate (foreclosures over
the past five years divided by outstanding mortgages for the data reported to the Banking
Department and foreclosures divided by paid-off plus foreclosed loans for the special
study).3 6 Average rates are given for individual neighborhoods, identified as those alleged to
be redlined and other,
3 7
and also by type of mortgage - one-to-four conventional, one-to-
four family federally assisted, and multffamfly conventional. The tables of the Banking
Department data show higher than average rates
3 6
for two of the three alleged redlined
neighborhoods for conventional one-to-four family and multifamily loans, and one of the
three alleged redlined neighborhoods for the one-to-four family federally assisted loans. At
most, only one of the four other neighborhoods had higher than average rates. Schafers
special study yielded relatively few loans for several of the 11 neighborhoods scheduled.
With respect to the four allegedly redlined neighborhoods, these limited data show no
above-average-foreclosure-rate neighborhoods for the one-to-four family loans and one
above-average-rate neighborhood for the other two categories of mortgages. Among the
seven other neighborhoods, higher than average rates are as follows: one in the one-to-four
conventional category, three in the one-to-four federally assisted category, and none in the
multifamfly category. Overall, then, there is some evidence of greater foreclosures in the
alleged redlined areas.
Schafer also computed a multivariate analysis of one-to-four family (conventional and
federally assisted) foreclosures using 113 inactive (98.6 percent paid-off and 1.4 percent
foreclosed) loans from his special study.
3 9
Twenty-six independent variables were included,
but a dummy variable for alleged redlined area was not included because there were too few
observations. For some unexplained reason the borrowers race was not included, even
though it was asked for on the forms.
4
Nor was the racial composition of the census tract
in which the mortgaged property is located included. The only statistically significant (at
the .05 level or less) coefficients reported are for "poor condition of building" (positive),
borrower married (positive), multiple-wage earner household (positive), and "two-to-four-
3 6 This procedure does not link the defaults with the portfolio of loans made of which it
was a part. See Von Furstenberg [ 1969] for a careful discussion of the shortcomings of not
using the cohort analysis method.
3 7 See description of Schafers mortgage supply study above for details.
3 8Where a rate is within one significant digit above the average, I counted it as not
above the average.
agA multivariate analysis of the Banking Department data is not presented. See the
description of variables used in his analysis of mortgage terms with these data for a list of
variables that could have been used for such an analysis.
4o "Female ," though, is included as a variable in the delinquency analysis and is men-
tioned in the text on the foreclosure analysis as omitted because of too few observations.
family house" (negative). The presumably unadjusted R
2
is .40. A similar analysis was cal-
culated on 45 multifamily loan foreclosures. Among the 14 independent variables is a
dummy for "property located in area alleged to be redlined." The coefficients of this vari-
able and "initial equity" are positive and are the only ones that are statistically significant
(but at the . 11 level) and the regression is not statistically significant, even at the .3 4 level.
Sehafer [1978, Chapter 8] {Upstate New York). A similar analysis with similar data
was conducted with data gathered from state-chartered banks in the Albany-Schenectady-
Troy SMSA, Buffalo, Rochester, and Syracuse. The foreclosure rates (dollars of foreclosed
mortgages to mortgages outstandhlg) derived from the data reported to the Banking Depart-
ment are greater than the regional averages for the following number of alleged redlined
(AR) neighborhoods and other neighborhoods (ON) (number with ratios greater than the
average/total number of neighborhoods):
one-to-four one-to-four multifamily
conventionals federally conventionals
assisted
AR ON AR ON AR ON
Albany SMSA 3 /9 7/7 3 /9 2/7 1/9 1/7
Buffalo 2/6
1/10 5/6 1/10 0/6 1/10
Rochester 3 /3 3 /11 3 /3 2/11 0/3 1/11
Syracuse
1/2 3 /5 2/2 1/5 0/2 1/5
The rates calculated with Schafers special study data show similar patterns.
Two multiple regressions also were computed with the special study data. The depen-
dent variable equals one if foreclosed and zero if not. One regression for the Albany SMSA
used 117 observations and 25 independent variables. The statistically significant (.05 level,
one tail) coefficients include downpayment (negative), poor condition of building (nega-
tive), four of the six age of structure dummy variables (all less than 3 9 years, positive), one
age of borrower dummy variable (46 to 49 years, negative), multiple wage earner household
(positive), FHA mortgage (positive), and property located in area alleged to be redlined
(positive, with the largest t value). The presumably unadjusted R
2
is .54. The other regres-
sion combined observations from the other cities (150 in all) and included the same types
and number (25) of independent variables. The statistically significant coefficients include
two age of structure dummies (20-29 years and 40-49 years, both negative), and percent of
housing units built before 1940 (positive). The coefficient of the property located in an
area alleged to be redlined is positive but not statistically significant. The R
2
is .19.
In sum, al! the Schafer studies indicate some weak evidence that foreclosures are rela-
tively greater in allegedly redlined than in other neighborhoods. Variables for the mort-
gagors race and for the racial characteristics of census tract were not included in the
analysis.
Mareis and Hull [1975] {National FHA inner city subsMized mortgages). The mortgage
foreclosures analyzed were those made under the FHAs section 221(d)(2), which provides
insurance on low downpayment, single-family property mortgages purchased by low or
moderate income families, and section 23 5, a subsidy program aimed at increasing home
ownership for people who otherwise could not afford it. The mortgage data are the Federal
National Mortgage Associations 1972 foreclosures to mortgages outstanding, aggregated by
zip code area. These foreclosure rates (dependent variables) were associated first with the
average weighted family income for zip code areas in each of eight randomly selected cities.
Statistically significant negative relationships in simple regressions are reported for six cities
(Detroit, Philadelphia, Chicago, Los Angeles, St. Louis and Seattle, the first two at the .01
level, the balance at the .05 level) and negative, but not significant, relationships for two
cities (Atlanta and Dallas). Similar results are found when the percent of families with
incomes under $600 was used (except that the relationship is positive).
Since family income is only a partial measure of neighborhood quality, multiple re-
gressions were computed with data from 123 zip code areas obtained from the Department
REDLINING RESEARCH
BENSTON
177
of Commerces 1970 Census of Population and Housing. The independent variables on
which the foreclosure ratio was regressed, listed in order of their importance as measured
by beta coefficients, are the number of pending indictments for criminal violations of fed-
eral laws related to FHA mortgages, instability of population (percentage of population who
moved into the area since 1965), income (weighted average family income), and education
(percentage of population with some high school). (Other variables that were correlated at
over .7 with an included variable are said to have been discarded. They are not identified.)
The R
2
of the section 221(d)(2) foreclosure rate is .44 and all of the coefficients are signifi-
cant at the .05 level, except for education which is significant at the .10 level. The signs
indicate that foreclosures were higher in zip code areas where the average population is
less stable, has lower income ( a decrease of $1,000 is associated with an increase in the fore-
closure rate of 1 percentage point), has a higher percentage of female heads of households
(a 10 percent increase is associated with a higher foreclosure rate of 1.5 percentage points),
and lower levels of education, given the number of indictments (which are positively associ-
ated with foreclosures). For the regression of section 23 5 foreclosures, the R
2
is only .07
and only the negative coefficient of education is significant (at the .10 level). Considering
that zip code areas often are not very homogeneous and that the foreclosures are not
directly related to the original portfolio of loans made (as in a cohort analysis), this study
indicates that the population characteristics of neighborhoods are important determinants
of mortgage foreclosures where the usual lending criteria were relaxed or disregarded.
Barth, Cordes, and Yezer [1979] (National FHA Insured Mortgages). This study is by
far the most ambitious analysis of defaults related to the redlining question. The data (about
10,000 observations) are a 100 percent sample of all FHA 203 (b) insured mortgages en-
dorsed in 1974 and 1975 that were in default by the end of 1976 and a 10 percent random
sample of other mortgages from this period. This sample certainly exceeds any used by the
other researchers whose work is reviewed. However, the two year or less horizon is very
short - several years less than the three or four years that VonFurstenberg [1969, p. 468]
found was the peak for defaults.
The researchers used multiple regression (OLS) analysis. The dependent variable is
a dummy equal to 1 if a default occurs, 0 otherwise. The independent variables include
the following: loan terms (loan-to-value ratio, term to maturity, and monthly payment-to-
income ratio), borrower characteristics (number of years married, and as dummies, Hispanic,
black, not married, and female head of household - coded 1 if condition applies, 0 other-
wise); property characteristics (structure in fair or poor condition, age of structure, and
wood construction - all except "age" are dummy variables), neighborhood characteristics
(central city, rural, and blighted - all as dummies), and city characteristics (new single-
family starts in 1975-76 to owner-occupied units, pre-1940 housing to all occupied housing,
SMSA population in 1974, city population growth 1970-75, city income growth 1970-75,
SMSA 1975 per capita income, and percentage black population). Though the choice of
variables is said by the authors to be based on a theoretical model (which is developed at
length), the variables do not differ much from the variables used by other researchers who
were not quite as formal in their presentation. However, the dates of the data used to specify
the independent variables are coincident with the dates of the dependent variable, which is
a considerable improvement over much other research.
Nine combinations of independent variables are presented. The statistically significant
variables (at the two-tail .05 level) that are of interest to the redlining question (and their
coefficients adjusted for the sampling proportions of defaulted and not defaulted mortgages
to give estimates of the increased or decreased (-) probability of default, the mean of which
is .015) are black (+.012), female head of household (-.003 ), wood construction (+.002),
central city (+.001), and blighted (+.003 ). Hispanic, age of structure and percent of black
population in city (given that the borrowers characteristics are included) have statistically
insignificant adjusted coefficients of zero. It also is interesting to note that the other city
characteristics variables, which if positive indicate a growing, prosperous area, all have
statistically significant, negative signs, with the exception of city income growth.
In sum, as the authors conclude: "Our results indicate that default risk is significantly
affected by location based on both neighborhood and city characteristics. This implies that
appropriate adjustment of mortgage terms based on location is consistent with profit-
maxim~zation. This type of adjustment would not be redlining in the economic sense."
[p. 511 Considering that the FHA interest rate is not adjusted for risk, these data indicate
that the FHA and/or lenders, advertently or inadvertently, discriminated in favor of some of
the areas and people said to have been redlined.
Terms Charged to Mortgagors
Scha.fer [1978, Chapter 6] (New York CiO~). As is described below, Schafer used data
reported to the New York State Banking Department by a subset of state-chartered financial
institutions on mortgages made in 1975. These data were aggregated by and averaged over
census tracts in the counties of Kings (Brooklyn), Queens and the Bronx that were identi-
fied through interviews with and reports by community groups) as allegedly redlined. Three
equations were stated and solved simultaneously, with maturity, loan-to-value, and interest
rate as the dependent variables. The 45 independent variables include: mortgage prices
(two of the three dependent variables predicted from the fh-st state estimates with theh"
equations), neighborhood attributes (six variables on housing code violations, four variables
on the fraction of buildings vacant, four variables on tax arrearages, two variables on struc-
tural fires per building, two variables on per capita welfare, per capita income, and change in
population - most of these variables measure the current level and change from a previous
date, usually 1970), risk of loss in mortgage lending (ratio of dollars of foreclosures over
past five years to amount outstanding, and ratio of dollars of 60-day delinquencies to
amount outstanding), income and assets (1970 per capita income, percent of households
in 1970 with incomes less than $15,000, and average 1975 property value), mortgage stocks
(four variables, one-to-four family and multifamily mortgages, conventional and federally
assisted), transactions predicted from the first stage, age of housing stock (fraction built
before 1940), racial change (percent nonwhite in 1974 and change in percent, 1974 less
1970), neighborhoods (six alleged redlined and three others, in dummy variable form with
one other - not redlined - neighborhood omitted to avoid overidentification), and the
constant.
Simultaneous equations were run with data on conventional one-to-four family mort-
gages (801 observations).
41
With respect to the variables of interest for the redlining ques-
tion, the following were found to be statistically significant (at the .05 one-tail level) for
the one-to-four family regressions: interest rates - for one alleged redlined neighborhood
interest is 1.3 3 percentage point lower and for one .3 4 percentage point higher (all of the
other coefficients were not significant and less than .26); maturity - lower for fraction built
before 1940, higher for increase in nonwhite population, lower by 7 and 10 years for two
alleged redlined neighborhoods (the interest rate is significantly lower by-1.3 3 percentage
point for the latter) and by 1.6 and 1.1 years for two other neighborhoods; loan-to-value
ratio - higher for fraction built before 1940, lower for percent of population nonwhite
(a tract greater by 50 percentage points averages 7 percentage points higher, higher for
change in percentage nonwhite (a change by 50 percentage points averages a ratio lower by
11 percentage points), lower for three alleged redlined neighborhoods averaging 4, 29, and
44 percentage points (the latter two have significantly shorter average maturities of 7 and
10 years), and lower for one other neighborhood by an average of 6 percentage points. Few
of the coefficients of the other variables are statistically significant, and many have signs
the opposite of those expected.
A similar set of two-stage regressions were computed for data on conventional multi-
family mortgages (118 observations). The only change was the combination of the neighbor-
hood dummy variables into one for "alleged redlined" and two for "other." Among the
variables related to the redlining question, the only statistically significant coefficients
show lower interest rates (-1.04 percentage points) and shorter maturities (- six years) in
the alleged redlined neighborhoods and lower loan-to-value ratios (-16 percentage points)
in one of the other neighborhoods. However, considering that relatively few multifamily
houses in a census tract are mortgaged in a year, these findings are of questionable value.
41 R~s are given, but how they were calculated for a system of simultaneous equations
is not explained.
REDLINING RESEARCH
BENSTON 179
Simultaneous equations also were computed with data on mortgages made over 25
years and collected in a special survey from a subset of mutual savings banks.
42
The 40 inde-
pendent variables include mortgage prices (two of the three dependent variables), building
characteristics (age and wood construction or not), borrower characteristics (income and
wealth), age of housing stock (percent built before 1940), neighborhoods (alleged redlined
and six other neighborhoods with the balance omitted to avoid overidentification), year of
loan (18 dummy variables), time and location interaction (each neighborhood dummy vari-
able was multiplied by a reciprocal of the average number of years from which loans were
made), and the constant.
The conventional one-to-four family regressions used 265 observations. Only two of
the "redlining" variables have statistically significant coefficients (at the .05 one-tail level):
lower interest rates in one other neighborhood (of -.3 3 percentage point) and shorter
maturity in another neighborhood (- seven years). Similar regressions were run with data
on one-to-four family federally assisted mortgages (283 observations), except that two
alleged redlined and six other neighborhoods were specified. The only statistically signifi-
cant coefficients of redlining related variables are: higher interest rates in one other neigh-
borhood (1.3 9 percentage point), shorter maturity in one alleged redlined neighborhood
(- seven years) and longer maturity in another alleged redlined neighborhood (16 years),
and lower loan-to-value ratios in two other neighborhoods (-3 3 and -7 percentage points).
If these results are taken seriously, they indicate little evidence that the alleged red-
lined neighborhoods are discriminated against with respect to the mortgage terms. How-
ever, the often theoretically unexpected signs of many of the variables and their statistical
insignificance and the quality of the data used lead to doubts about the validity of the
analysis.
Muth [1979] (Oakland, California). Muths study used data on 1903 conventional
loans made by state-chartered savings and loan associations in 1976 and 1977, as reported
to the State Commissioner.
43
This limitation of the lenders who provide mortgage funds in
Oakland to SLAs is not serious, however, since he primarily analyzed the yields on mort-
gages in a state where yields are market determined.
Interest on a loan was calculated to include points. It is the internal rate of return that
equates the initial loan amount less fees and discounts paid by the borrower to the present
value of monthly payments plus an assumed repayment of the unpaid balance after 12 years.
This dependent variable was regressed on the following independent variables: term and
amount of the loan, borrower characteristics, most of which are related to possible discrimi-
nation (black male, other minority male, female, and property not owner-occupied - all as
dummy variables), age of building, neighborhood characteristics (percentage of occupied
dwellings units black occupied, and percentage of year-round dwellings lacking some plumb-
ing facilities - all by census tract, 1978 pretest data), a dummy variable that equals 1 if a
loan was made in 1977, and the constant. The variables were transformed to natural loga-
rithms. Each observation is a loan.
4~
Most of the coefficients are statistically significant at the .05 one-tail level; the R
2
is
.3 4. Before considering the redlining related variables, it is interesting to note that the co-
efficients of the term to maturity and loan amount variables are negative and highly signifi-
cant. The former indicates that the shorter maturity and higher interest rates are comple-
ments,ex post, though they may be substitutes, ex ante. When the term to maturity variable
was removed from the regression, the coefficients of the loan amount and nonowner-
occupied variables increased considerably. With respect to borrower characteristics, the
following levels of statistically significant coefficients are reported: black male (.10), other
minority male (.04) and female (.04). In terms of signs and magnitudes, with an average
a~See description given above under "The Relative Risk of Lending - Defaults and
Foreclosures."
~ The data were screened for errors, wliich eliminated less than 100 mortgages.
~4Average contract rent of rented dwellings, median value of single-family owner-
occupied units, and percent of dwellings built prior to 1940 also were included initially, but
later discarded.
interest rate of 9.3 percent, the coefficients indicate that, on average, black male bor-
rowers (who made about 20 percent of the loans analyzed) paid .02 percentage point
more, other minority males paid .03 percentage point more, and females (who made 17
percent of the loans) paid .03 percentage point less than others, cet. par. The statistically
significant coefficient for age of house indicates that a loan on a 40-year old dwelling re-
quired a yield of about .10 percentage point more than a loan on a new dwelling. Finally,
the statistically significant (.03 level) coefficient of percentage black residents in a census
tract indicates that mortgages in all black tracts bore interest rates that were .07 percentage
point higher than rates in all white tracts. Thus the estimated effect of possible discrimina-
tion on interest rates is very small.
Muth also regressed the debt-to-equity ratio on the independent variables, excluding
the term to maturity and loan amount variables; the R
z
is .12. Three variables have co-
efficients that are statistically significant at the .05 level. They indicate that black male
borrowers experienced loan-to-value ratios of .84 compared to .79 for white male borrowers;
the ratios for borrowers in all black census tracts were roughly one-fourth less than those
of borrowers in all white tracts, and mortgages on 40-year old buildings were about 5
percentage points lower than on new buildings.
The term to maturity was regressed on the same variables; the R
~
is .21. All the co-
efficients are statistically significant. The signs are positive (longer terms) only for black
male borrowers and female borrowers. The magnitudes, though, are quite small for the
borrower characteristics, but indicate 5 percent shorter maturities for mortgages on prop-
erties in all black compared to all white census tracts. However, when two additional vari-
ables, market value per square foot of living area of the mortgaged property and the bor-
rowers monthly income, were added,
as
the percentage of black residents and the other
minority male coefficients became very small and not significant and the R
2
increased to
.26. The age of house coefficient also declined to half its former value. Muth concludes:
"This suggests to me that the higher riskiness lenders may attach to loans in black neighbor-
hoods is not due to race per se but rather to the poorer quality of dwellings and lower
borrower income on loans in such areas." [p. 15]
Muths study thus indicates that the terms on mortgages made by state-chartered sav-
ings and loan associations were not significantly worse for possibly discriminated-against
people or areas. Though he found, as did other researchers, that the institutions made far
fewer conventional, one-to-four family mortgages in census tracts with higher proportions
of black residents, his findings with respect to terms lead to the conclusion that either the
supply is demand-determined or if it is supply-determined, the demand for conventional
mortgages is highly elastic. As he concludes: "on either interpretation, the fact that fewer
conventional mortgages are made in black and other inner-city areas would not appear
injurious." [p. 171
Benston, Horsky and Weingartner [1978, Chapter 2] (Rochester, New York). The
researchers gathered data on 712 one-to-four family house mortgages made by the principal
mortgage lenders in Rochester (three mutual savings banks, two commercial banks, one
savings and loan association, and two mortgage companies) over the period September
1973 through September 1976. The mortgages were made in an area of Rochester identi-
fied by community activists as redlined (the central city) and in a control area, a suburb.
Mortgages were identified as conventional, FHA and VA. Because the terms on mortgages
can be demand- as well as supply-determined, loans with downpayments of 50 percent or
more or maturities of 10 years or less were excluded. As other researchers have found,
conventionals are predominant in the suburb (in terms of numbers, 59 percent of suburban
mortgages), but not in the central city (12 percent of central city mortgages). Within each
area and type of mortgage, there is virtually no difference in the interest rates charged. In
the suburb compared to the central city, the average loan-to-value ratios are higher on con-
ventionals by .027 but lower on FHAs and VAs by .03 4 and .025. The greatest differences
between the areas are in the number of months to maturity, as follows: conventional, 98
4STheir coefficients are significant at better than the .0001 level.
181
months; FHA, 96 months; and VA, 80 months. Because a lower loan-to-value ratio and
shorter term to maturity are alternative means of reducing the risk to the lender of defaults,
a variable that combined thei~ effect was constructed. The accepted risk of decline (ARD) is
the complement of the rate at which a houses price could decline before the end-of-period
owners equity would be zero. The higher the loan-to-value ratio and the more the months
to maturity, the higher the ARD and hence, given the interest rate, the more risk the lender
is accepting in granting the mortgage. For all three types of mortgages, the ARD is slightly
higher on average for suburban than for the central city mortgages, which indicates that
these mortgages were perceived as less risky or that lenders were biased against central city
properties.
To assess whether the differences in loan-to-value ratios, months to maturity, and the
ARD were due to the area of the properties or to other characteristics of the borrowers and
properties, a mortgage supply function was specified. For each type of loan (conventional,
FHA, and VA) in each area (central city and suburb), the loan-to-value ratio, the number of
months to maturity, or the ARD was regressed on the following variables: borrower charac-
teristics (family income, age less than 3 5, married, previously a home owner, co-borrower
used, and low or no credit rating - all but the first are dummy variables), property charac-
teristics (price of the house, appraisal/price of the house, built before 1941, built 194_1-1960,
built after 1960,
46
and private mortgage insurance on conventional mortgages - the latter
fottr are dummy variables), and census tract changes (median house price 1970/1960 and
occupancy rate 1976/1970 included for the central city observations only since the suburb,
effectively, is a single census tract). Since the observations cover three years, two ad~litional
variables were included to adjust for changes over the period - the long-term yield on U.S.
bonds and the monthly flow of savings deposits in the area. The statistically significant (.05
level) coefficients varied between the areas, implying different mortgage lending functions
for each area. They a/so differed among the terms analyzed. The regressions were rerun with
the nonsignificant variables omitted (which changed the coefficients of the significant vari-
ables but little). These coefficients were used to specify the lending functions. The coeffi-
cients estimated from, say, the central city mortgages (excluding the census tract changes)
then were multiplied by the mean values of the suburban mortgage variables to obtain the
terms that these mortgages would have gotten had the suburban lending function been
applied. The same procedures was followed for estimating the terms that the suburban mort-
gages would have obtained had the central city lending function been used.
The exercise leads to the following conclusions, first with respect to loan-to-value
ratios. Were the suburban lending function used rather than the central city lending func-
tion, the average central city conventional mortgage would have a slightly higher loan-to-
value ratio than when the central city function is used (.779 vs..760). But the average
suburban mortgage would have a higher loan-to-value ratio were the central city lending
function used rather than the suburban function (.884 vs..793 ). Similar predictions were
calculated for FHA and VA mortgages. Thus it appears that the loan-to-value ratio for each
type of mortgage is not h function of the area in which the property is located.
With respect to the number of months to maturity on conventional mortgages, appli-
cation of the suburban lending function to the average central city borrower and property
characteristics (primarily family income, age under 3 5, and the price and age of the house)
predicts a term of 285 months compared to the actual average of 247 months. When the
central city lending function is applied to the suburban borrower and property characteris-
tics (primarily the price and age of the house), 3 3 1 is the predicted number of months to
maturity, compared to the 3 45 actually experienced. Thus some 3 8 or 14 months of the
~raw" difference of 98 months are unexplained and appear due to unspecified area-related
factors. These fewer months to maturity translate into higher monthly payments by a buyer
of the average priced central city house of $5 or $2. Most suburban FHA and VA mortgages
carry 3 0-year mortgages. (The coefficient of variation is only .04.) Hence the suburban
46 Overidentification was avoided because from 8 to 42 percent of the properties were
not identified as to age. These observations were coded 0. Tests indicated that this procedure
did not bias the coefficients.
multivariate analysis explains very little, and application of the suburban lending function
to the central city values predicts almost the same number of months to maturity as the
average suburban loan, due mostly to the constant term. When the central city lending
function is applied to the suburban values (primarily the price of the house), the predicted
number of months to maturity is 3 55 compared to the central city average of 258 months.
Thus 27 months of the "raw" difference of 96 months appear due to the area. This works
out to a higher monthly payment for the average central city home buyer of $4. The VA
analysis produced similar findings. It is interesting to note that the results for the conven-
tional and the FHA and VA mortgages are almost the same, which implies that the lenders
and the government assessed the area-related characteristics similarly.
The accepted risk of default (ARD) variable provides a measure of the joint effect of
the loan-to-value ratio and term to maturity of a mortgage. Application of the mortgage
lending functions (as described above) indicates that the average central city mortgage
would have slightly better terms were the suburban rather than the central city function
used, but the suburban mortgage would have slightly better terms were the central city
rather than the suburban function applied. These results are found for all three types of
mortgages. The authors conclude: "Thus the data do not reveal any appreciable difference
in mortgage terms between the areas, even without accounting directly for differences in
risk faced by lenders in granting ~nortgages in the central city. These findings for Rochester
are completely contrary to the allegation of those who assert the existence of redlining
by mortgage lenders." [p. 73 ]
King [1979B] (Miami, San Antonio and ToledoJ. The data were derived from registers
of loan applications made in September, October and November 1978 at all the federal
savings and loan associations in the SMSAs studied. The cities studied were chosen to in-
clude those with substantial proportions of minorities, allegations of redlining or other
forms of discrimination, nonexternal statewide branching, and adequate loan volume and
the presence of one or more minority operated associations. The loan terms studied (the
dependent variables) are the interest rate, number of months to maturity, loan-to-appraisal
value ratio, and fees. Because mortgagors could have demanded the terms they received
(particularly downpayments), King also used the difference between each of the terms (with
the exception of fees) asked for and received as a dependent variable. In this important re-
gard, his study is unique. The independent variables included the following: ~nortgage char-
acteristics (FHA-VA and private mortgage insurance as dummy variables, dollar amount of
loan, and loan-to-appraisal value of property - except for loan-to-value ratio regression),
borrower characteristics (combined gross monthly income and net wealth of applicant and
co-applicant, ratio of monthly gross income of co-applicant to total for applicant and co-
applicant, age of applicant and age squared, and dummy variables equalling 1 if applicant is
black, Hispanic, other minority, female or unmarried), property characteristics (age and its
square), neighborhood characteristics (by census tracts, 1970 data - percent black, percent
Spanish-American, percent of households having incomes below the poverty level, percent
of owner-occupied units, percent owner-occupied units built before 193 9, and for Toledo,
29 dummy variables designating defined city neighborhoods compared to the suburbs), the
time in days from September 1978 to the approval of the mortgage application to account
for the upward trend in mortgage rates, and dummy variables for each savings and loan
association, less one. The sample sizes are 1960 in Miami, 559 in San Antonio, and 953 in
Toledo. About the same number of applicants had to be omitted because of missing data.
Ordinary least squares regressions were computed, from which the coefficients of the
variables that would be considered as measuring discrimination (e.g., the applicants race)
provide estimates of the magnitude of the possibly discriminatory effect. King also used
simultaneous equations, but these did not change his conclusions. The regressions indicate
that ordinary economic variables, such as the presence of mortgage insurance, other mort-
gage terms, and the borrowers income, and the dummy variables representing the individual
associations are the most important explanatory variables. With respect to each of the mort-
gage terms he found that the following are statistically significantly (.05 one-tail level)
related to possibly discrimination factors, cet. par. :
REDLINING RESEARCH BENSTON 183
Interest rates. Miami (R
~
of .28)
47
- Rates averaged 9.59 percent. They were
higher for females by 3 basis points and for census tracts with 50 percent more Spanish-
Americans by 4 basis points and lower for census tracts with 50 percent more blacks by 3
basis points and for census tracts with 50 percent more of the houses built before 193 9 by
6 basis points. San Antonio (R
z
of .58) - none. Toledo (R
~
of .51) - Rates averaged 9.86
percent. They were lower for other minorities by 17 basis points, and for three of the 29
neighborhoods by 13 , 27 and 3 5 basis points.
Number of months to maturity. Miami (R
~
of .10 for level and .02 for difference) -
Maturities averaged 3 44 months. For levels, building age showed longer maturities until age
54, then shorter maturities, and mortgages in census tracts with 50 percent more of the
houses built before 193 9 were five months shorter; but the coefficients of the later two
variables are not significant for the difference between actual and requested maturities.
San Antonio (R
~
of .3 4 for level and .03 for difference) - Maturities averaged 3 50 months.
There were shorter maturities of 10 months for Hispanics, 74 months for mortgages in
census tracts with 50 percent more blacks, and 10 months for mortgages in census tracts
with 50 percent more houses built before 193 9; but only the coefficient of the black census
tract variable is significant for the difference between actual and requested maturities, indi-
cating 16 months fewer in tracts with 50 percent more blacks. Toledo (R
~
of .46 for level
and .08 for difference) - Maturities averaged 3 21 months. There were shorter maturities of
22 months for a 50-year old house compared to a new house, of 26 months for Hispanics, of
3 7 months for other minorities, of 11 months for unmarrieds, of 21 months for applicants
age 60 compared to applicants age 40, and of 3 0 and 87 months for 2 of the 29 neighbor-
hoods specified; but only the coefficient of the age of house and of two of the neighborhood
variables are significant for the differences between actual and requested maturities.
Loan-to-value ratio. Miami (R
~
of .28 for level and .01 for difference) - Ratios
averaged .79. They were lower by 2 percentage points for females for the level but not
significant for the difference between actual and requested. San Antonio - none. Toledo
(R
z
of .42 for level and -.01 for difference) - Ratios averaged .75. They were higher for
blacks by 4 percentage points and lower by 49 percentage points for census tracts with 50
percent more Spanish-Americans, but neither of these are significant for the difference
between actual and requested; only unmarried is significantly lower, by 1 percentage point.
Fees. Miami (R
~
of .68) - Average amount of $1472. They were higher by $13 6 for
blacks and by $106 for other minorities (significant only at .10 level) but lower by $217 in
census tracts with 50 percent more blacks (significant at less than the .001 level) and by $93
in census tracts with 50 percent more Spanish-Americans (significant at the .10 level) (these
findings appear to cancel each other), and higher by $3 16 in census tracts with 50 percent
more houses built before 1979. San Antonio (R
~
of .79) - Average amount of $421. They
were lower for other minorities by $73 . Toledo (R
~
of .52) - Average amount of $626.
They were highest for applicants about 45 years old and in two of the 29 neighborhoods
specified, by $90 and $112.
In summary, few statistically significantly more onerous mortgage terms are related
to variables that measure possible discrimination. These few are of small magnitudes, with
the exception of the percentage black census tracts in San Antonio, a city with relatively
few blacks. And, at least as often statistically significant less onerous terms are associated
with these variables. Thus the findings are inconsistent with the hypothesis that the institu-
tions studied discriminated against borrowers or areas in the form of worse terms.
Appraisal Values and Purchase Prices of Properties
Schafer [1978, Chapter 4] (New York City and Nassau-Suffolk SMSAs}. The dat~i were
gathered in Schafers special survey savings banks files that included at least a 27-year period,
wherein 100 loans were sampled from each bank. Of these, complete information were ob-
are adjusted for degrees of freedom.
184
tained on the 183 2 observations of one-to-four family mortgages used for this analysis.
48
The dependent variable is the ratio of the appraisal value to the purchase price. It has a
mean value of .98 with two-thirds of the values falling between .90 and 1.06. (Values below
.4 and above 1.6 were omitted.) The independent variables include the type of the building
(percent of housing stock in census tract built before 193 9 per 1970 census, and age of
building relative to new construction expressed in six dummy variables), type of mortgage
(FHA or VA rather than conventional, expressed as dummy variables), property location
(alleged redlined and other city neighborhoods compared to Suffolk County, expressed as
dummy variables), year of transaction relative to 1951 (seven dummy variables), and the
constant. Schafer also said that he included mortgage terms and the price of the property in
the regression, but the coefficients were insignificant or of the "wrong sign."
An OLS regression was computed with the area-wide data, which yielded an R
2
of .16.
With respect to areas, the coefficients indicate a statistically significant (at the .05 one-tail
level) lower ratio of 5 percentage points for one of the three designated alleged redlined
areas and lower ratios of from 2 to 6 percentage points for five of the six designated other
(not redlined) neighborhoods. Significantly lower ratios also are shown for the age of build-
ings only between 10 and 29 years of about 2 percentage points. Thus the data are contrary
to the hypothesis that lower appraisals characterize allegedly redlined areas or older prop-
erties.
Regressions also were computed for observations in allegedly redlined New York City
neighborhoods, other New York City neighborhoods, and three suburban counties. The co-
efficients of the age of buildings dummy variables are significantly negative, but only in
the other (nonredlined) New York City regression. The coefficients of the other variables
either are insignificant or are similarly inconsistent with the hypothesis that appraisals
are lower than purchase prices in the allegedly redlined areas.
A similar analysis was undertaken for 43 multifamily house mortgages. The coefficient
for the alleged redlined areas dummy variable is positive at about the .10 level. Thus as
Schafer concludes, "The analysis indicates that appraisal practices as reflected in granted
loans do not indicate that lenders systematically under-appraise properties in areas that are
alleged to the redlined." [p. 4-46]
Schafer [1978, Chapter 9] (Upstate New York}. Regressions were run with data similar
to those used in the New York areas. The number of observations on one-to-four family
mortgages and R2s from each of the regressions are as follows: Albany-Schenectady-Troy,
467 and .09; Buffalo, 3 10 and .16; Rochester, 177 and .22; and Syracuse, 3 19 and .13 . The
coefficients for the age of building variables are not significantly different from zero in the
Albany-Schenectady-Troy and Buffalo regressions. Older houses average higher appraisal-to-
selling price ratios in Rochester and Syracuse. The coefficients of the alleged redlined and
the other city neighborhoods variables are not significantly different from zero in the
Albany-Schenectady-Troy and Syracuse regressions. In Buffalo, one of four alleged redlined
neighborhoods and one of seven other city neighborhoods averaged higher appraisal ratios
of 4 and 8 percentage points; the other coefficients are not significant. In Rochester, one
of three alleged redlined and one of three other city neighborhoods averaged lower appraisal
ratios of 4.5 and 3 percentage points; the other coefficients are not significant. Including
mortgage terms and the price of the house and running the regressions separately on obser-
vations from alleged redlined and other neighborhoods did not alter these conclusions.
Thus, these data are inconsistent with the hypothesis that savings banks appraised at lower
rates older properties or properties located in allegedly redlined areas.
Benston, Horsky and Weingartner [1978, Chapter 2] (Rochester, New York). As is
described above (see terms charged to mortgagors), mortgages made over a three year
period at almost all area institutions were sampled. The ratios of appraisals to purchase
prices of properties were calculated for conventional, FHA and VA mortgages separately on
48In Appendix B, Schafer describes the survey and states that "the total sample size
was 2,208 for the Bronx-Kings study area." [p. B-4] He also states that the data cover
"either the last 15 years or the period for which the bank had complete files, whichever
was shorter." [p. B-3 ]
properties located in an allegedly redlined area and in a "control" suburb. The ratios for
conventionals and VAs in both areas are almost exactly equal to one. The FHA ratios are
significantly higher at 1.02, but almost exactly the same in each area. Because of the close-
ness of the means between areas, a multiple regression analysis was not undertaken.
King [J979B] [Miami, San Antonio, and Toledo). The data were gathered as described
above (see mortgage terms). Two types of analyses were conducted. First, the applications
rejected because of (stated) inadequate appraisal values were counted and categorized by the
race of the applicant and the area of the property. King concludes: "It is evident from these
figures that inadequate appraisal value is a relatively infrequent explanation for rejection.
Moreover, relatively few of the rejected applications are from minorities, and there is no
evidence that the applications are clustered geographically." [p. 11 ]
Second, OLS regressions were run, with the ratio of appraisal value to property price
as the dependent variable. The independent variables include the type of mortgage (private
or FHA-VA insured), the age of the property and its square, whether or not the borrower
was black, Hispanic, other minority, female, and]or unmarried, the borrowers age and its
square, the 1970 percentage in census tracts of blacks, Spanish-Americans, people below
the poverty level, houses built before 193 9, and owner-occupied houses, and dummy vari-
ables for the S&L association and (in Toledo) specified city neighborhoods. The sample
sizes (with the percentages they represent of the. total number before observations were
omitted because of missing data in parentheses) are 283 2 (75 percent) in Miami, 765 (53
percent) in San Antonio, and 123 4 (69 percent) in Toledo. The means and standard devia-
tions (in parentheses) of the ratios of appraisal value to property price are similar to those
found by Benston, Horsky and Weingartner [1978], (the standard deviations are larger
because outlyers were included), and are as follows: Miami, .99 (.06); San Antonio, 1.02
(.09); and Toledo, 1.02 (.07). The adjusted R2s, at .04, .01 and .06, are quite low. Statis-
tically significant variables (at the .05 one-tall level) related to possible discrimination are
the following:
Miami - The relationships are contrar-y to those expected ha that blacks average 3 per-
centage points higher appraisal ratios and the ratios decrease with the age of the house until
22 years, after which they increase. (The most significant variables are the savings and loan
association dummies.)
San Antonio - The hypothesis that all coefficients are zero cannot be rejected at the
.05 level.
Toledo - The relationships are contrary to those expected: census tracts with higher
percentages of Spanish-Americans average higher appraisal ratios, older houses have higher
ratios, 8 of the 29 city neighborhoods have higher ratios (2 at the .10 level) than the subur-
ban areas of between 2 and 8 percentage points, and FHA-VA mortgage appraisal ratios
average 5 percentage points lower than conventionals.
Thus all of the analyses yield findings that are contrary to the hypothesis that lower
appraisals are associated with possibly discriminated-against people or areas.
Denials of Mortgage Applications
Black, Schweitzer and Mandell [1978] (Nationwide). A survey conducted by the
Comptroller of the Currency and the FDIC provided data from 176 banks that chose to co-
operate of 3 00 that were asked. Between September 1976 and February 1979, mortgage
applicants at the participating banks completed a form that asked for their personal charac-
teristics and mailed it to the FDIC. The banks later sent their forms that reported informa-
tion about the mortgages, which were matched with the customers forms. About half of
each group could be matched (4895 in all). Missing data reduced the number to 3 456. The
authors state "There is no indication that banks systematically avoided returning forms that
were matched with a group considered a priori as being most likely to suffer discrimination."
[p. 1871
The rejection rate overall was 2.7 percent (13 8 loans), which the authors postulate may
be dne to prescreening by banks or real estate brokers, among other reasons. The data were
analyzed with probit analysis, the dependent variable being equal to 1 for reject, 0 for
accept. The independent variables, entered linearly, include loan terms (amount requested,
downpaymcnt, loan origination fee, years to maturity, interest rate, monthly payment, and
insurance status), economic variables (total income, net worth, monthly debt, years em-
ployed, self-employed, and age of house), and personal variables (male sex, black race, and
age 55 years or older), plus the constant term. The dependent variables were regressed on the
loan terms first (Model 1). Then the economic variables were added (Model 2), and then the
personal variables (Model 3 ). In all three models the only statistically significant coefficients
at the .05 one-tail level (and their signs) are downpayment (-), interest rate (+), self-
employed (-) and age of house (+). In Model 3 , coefficient of black race is negative but
significant only at the .10 level. The authors conclude: "Note, however, that the summary
statistics show a large improvement between Models 1 and 2. The change between Models 2
and 3 is not large. Therefore, although the racial variable is statistically significant at the 90
percent level of confidence, one must interpret its overall impact on the lending decision
with some caution." [pp. 189, 191] One also should note that 16 black applicants were re-
jected and 142 were accepted. Since a probit analysis was employed, the estimated quantita-
tive relationship between black race and the probability of denial cannot be determined
from the coefficients presented.
Schafer [1978, Chapter 7] (New York City Area]. The sample used was derived from
Equal Housing Opportunity Lender forms which provide data on applicants for new con-
ventional owner-occupied mortgages on one-to-four family houses. These forms were filed
at and completed by 27 mutual savings banks in New York City and nearby counties,
between May 1976 and October 1977. Of the 11,799 applications filed for these loans, 22
percent could not be used because of missing data that could not otherwise be obtained or
estimated, leaving 9149 observations. Since he was aware of the effect of missing observa-
tions on the analysis, Schafer was careful in conducting many tests to determine whether
the exclusion of data introduced biases, such as a relatively larger number of applications
from black than white applicants, or from alleged redlined vs. other neighborhoods, and
exclusion of applications by blacks with relatively different incomes than those whose appli-
cations were not excluded. Unfortunately, all of the test results are presented as compari-
sons of the sample with the total number of applications filed for all one-to-four family
mortgages, which total includes applications for government-assisted mortgages, refinancing,
nonowner-occupied houses, etc. These excluded applications comprise an additional 22
percent of the universe of 11,799 applications. Therefore, one cannot be sure that the sam-
ple is as representative as it seems.
49
The applications analyzed were categorized as accepted as applied for (69 percent),
accepted after modifications of terms (18 percent), withdrawn (4 percent), and denied, (9
percent). The probability of being in one group rather than in any of the other was esti-
mated with OLS regressions, with the dependent variable coded 1 if the application was
denied, 0 if it was accepted with or without modification or withdrawn.
5
The independent
variables include the following: the applicants financial characteristics (five dummy variables
representing categories of income and wealth and two dummy variables for years at present
occupation), loan characteristics (a dummy variable equalling one if the loan requested ex-
ceeds two times income and the requested loan-to-value ratio), neighborhood characteristics
(percent of households with income $15,000+ in 1969, change in household income 1976
less 1969, change in population 1977 less 1960, and percent of housing built before 1940 -
all by census tract), property locations (five New York City counties), race of applicant
(black and other minority), percentage black population in census tract 1970, and the con-
stant. The regressions are highly statistically significant,
sl
The nonqnestionable characteris-
49The tests conducted, however, are quite thorough and imaginative.
STwo other dependent variables, accepted with modification and withdrawn, were
specified similarly. The findings are not summarized here because these events are not
likely to be related to redlining practices, as is explained in the text above.
slSince the deper~dent variables is 0 or 1, the R
2
is of limited meaning. Multinomial
logit estimates also were calculated, with essentially similar findings for the signs and signi-
ficance of the variables coefficients.
REDLINING RESEARCH
BENSTON 187
tics show up as one would expect (e.g., higher income applicants are less likely to be denied).
Overall, 9 percent of the applications analyzed were denied.
The coefficients of the applicants race variables, black and other minority, are positive
and statistically significant.
5~
They imply that the probability that an application for a
mortgage was denied is greater by 6.6 percentage points for a black and 2.9 percentage
points for an other minority person than a white. These findings were cross checked in sev-
eral ways, most importantly by dividing the sample into black (552 observations), other
minority (861 observations), and white (773 6 observations). The regressions were rerun to
determine whether the coefficients differed significantly. The regressions do, at the .01 level.
So do several of the coefficients, indicating that low-income blacks have a higher probability
of denial than low-income whites, more time on the job is a significantly positive factor for
blacks but not for whites, and blacks have a much higher probability of denial when the
loan requested exceeds two times their income. (None of the coefficients of the location
variables were significant.) Schafer considers the possibility that these findings of apparent
racial bias could be due to correlations of race with omitted variables, in particular socio-
economic factors, two-worker families and the applicants poor credit rating. He rejects
these explanations because he finds that income and wealth and two-worker applicants do
not differ substantially by race, and poor credit rating is stated by the banks as a reason for
denial more for whites than for blacks. Thus he calculates and concludes that "Blacks have a
13 - to 22-percent chance of having their applications denied compared to the 9- to 21-
percent chance for similarly selected whites [depending on the regression used for the calcu-
lations]. Applications from similarly situated other minorities, a group composed mostly of
Hispanic and Asian families, have a 14-percent chance of having their applications denied."
[p. 7-114]
Alleged area redlining was tested with the five New York City counties dummy vari-
ables (applications on suburban properties were coded zero). Only the coefficient of subur-
ban NYC county (Richmond) of-.03 is significant. Since these are crude proxies for alleged
redlining, regressions were run with 3 3 3 6 observations from Bronx, Kings, and Queens coun-
ties. The regressions run for the entire sample were modified as follows. The five county
dummy variables were replaced by 20 variables measuring specific neighborhood problems
(six measures of housing code violation, four measures of vacancies, four measures of tax
arrearages, two measures of welfare) and 10 neighborhood dummy variables (seven alleged
redlined and three others). The only coefficient of the alleged and other neighborhood vari-
ables that is statistically significant is for an alleged redlined neighborhood (Park Slope). It
indicates that applicants in this neighborhood are less likely to be denied than applicants in
nonspecffied, presumably favored neighborhoods by 8.9 percentage points. The coefficients
of the percent black in census tract variables are not statistically significant, but the co-
efficients of the black and other minority variables are significant, and at +.04 and +.05 are
similar to those described above. The coefficients of percent of housing built before 1940
variables also are significant, and indicate that an applicant for a mortgage on a house in a
census tract where 50 percent of the houses were built before 1940 compared to one with
no old houses would have a probability of denial that is 2 percentage points greater. Schafer
concludes that while "the race of the applicant is crucial.., geographic location, however,
does not increase the likelihood that an application will be denied although there are some
areas in which applications are more likely to be modified. In fact, if geography is a major
consideration in lending decisions, it is probably used as a surrogate measure for actual
risks of loss associated with the subject property because of the condition of its surround-
ings." [p. 7-112]
Schafer [1978, Chapters 11, 12] (Upstate New York]. Multinomial logit estimates
were used for analysis of data from the Albany-Schenectady-Troy SMSA, Buffalo, Ro-
chester and Syracuse. The source of the data and method of checking for the effect, of
~Denials as percentages of the total number of applicants are 12 percent for whites,
18 percent for blacks, and 12 percent for other minorities. The number of denials are 627
whites, 97 blacks, and 106 other minorities.
188
excluded observations are similar to those described above for Schafers New York City
analysis. The equations specified are similar to the Bronx, King, and Queens counties
regre.ssions, except that the 20 variables measuring specific neighborhood problems could
not be included. The equations are statistically significant.
For the Albany-Schenectady-Troy SMSA 6173 applications were analyzed, 74 percent
of the universe of 83 22.
53
Eight percent of the applications were denied. With respect to
race, there were 450 denials of whites (8 percent of their total), 20 of blacks (17 percent),
and 19 of other minorities (9 percent). The coefficients of two of the eight neighborhoods
alleged to be redlined are significant and positive, indicating that applicants for mortgages
on these properties have a higher probability of denial than suburban residents.
~4
The esti-
mated probabilities are 81 and 73 percent compared to 10 percent for a similar average
suburban application. The only other statistically significant property location coefficient
is for the nonalleged-redlined City of Troy, for which denials are estimated to be 6 per-
cent higher than a similar average suburban applicant,
ss
The coefficients of the other vari-
ables of interest (race and areas with old houses) a_re not significant.
The Buffalo SMSA analysis used 7404 applications, 87 percent of the universe.
~6
Eight percent of the applications analyzed are denials. In terms of race, there are 520 denials
of whites (8 percent of their total), 45 of blacks (19 percent), and 24 of other minorities
(7 percent). Six of the city areas have significant negative coefficients, indicating lower
probabilities of denial in these city areas than in the suburbs. Of these, one is alleged to be
redlined and three others are said to be redlined by some community organizations. The
coefficient of the old house tract variable (percent of housing built before 1940) is signifi-
cant and positive but has a very small effect on the probability of denial. The coefficient
of the percent black census tract is not significant. Finally, black, but not other minorities,
have a significantly greater probability of denial. Compared to similar average white appli-
cants, the probability of denial is 12 percentage points greater.
The Rochester SMSA sample of 2951 applications is 84 percent of the universe,
s7
Rochester mutual savings banks denied 3 percent of the applications. With respect to the
applicants race, 72 white applicants were denied (3 percent of their total), 6 blacks (10
percent), and 4 other minorities (4 percent). None of the city neighborhood dummy vari-
ables (alleged redllned and other) have significant coefficients, though the signs of all but
one are negative, indicating some smaller probability of denial than applications for mort-
gages on suburban properties. The coefficient of the old house tract variable is significant
and positive. The racial composition of a census tract is not significant. But the average
black applicant was computed to have a significantly greater probability of denial of 9
percentage points compared to a similar white applicant.
The Syracuse sample of 2563 applications is 82 percent of the universe,
s8
The mutual
savings banks denied 6 percent of the applications. In terms of the applicants race, 147
white applicants were denied (6 percent of thek total), 5 blacks (14 percent), and 5 other
minorities (7 percent). Neither of the two alleged redlined neighborhoods has statistically
significant coefficients (though both are negative). Only a county area has a significant
SSTests for noninclusion bias included an additional 51 percent of the universe, com-
prised of definitionally excluded applications. Hence the apparent absence of noninclusion
bias cannot be determined.
s4 One neighborhood (Hudson/Park) is in Albany. For the three alleged Albany neighbor-
hoods there are 15 denied applications. The other neighborhood (Hillside) is in Troy. For
the three alleged redlined neighborhoods in Troy there are five denied applications.
ss Sixteen denied applications came from this area.
S~Test for noninclusion bias include 3 63 5 other applications in addition to the 1082
excluded because of incomplete information.
STAs is discussed above, the tests presented for estimating noninclusion bias include an
additional 203 2 applications, 56 percent of the universe.
SSThe noninclusion bias analysis includes 2579 other applications in addition to the 609
that were excluded because of missing information.
positive coefficient. The coefficients of the age of housing and percentage black population
are not statistically significant. But the coefficient of black is significant and positive,
indicating that an average black applicant would have a 15 percent chance of denial com-
pared to a 7 percent chance for a similar white applicant.
With respect to the upstate areas studied, Schafer concludes: "In general, objective
lending criteria, such as applicant income, net wealth, and requested loan-to-value ratio,
are major determinants in the lending decision in all four areas. The race of the applicant
is also crucial in three of the four metropolitan areas (Buffalo, Rochester, and Syracuse).
Property location in an alleged redlined neighborhood is a significant impediment to a
favorable decision in only one metropolitan area (Albany-Schenectady-Troy)." [p. 12-155]
Evaluation of the Schafer Studies. These studies of five municipal areas are done with
considerable care and imagination. The validity of the findings, though, are in doubt for the
following reasons. First, though considerable effort was given to checking for noninclusion
bias, the inclusion of a relatively large number of applications for mortgages not in the uni-
verse (e.g., FHA mortgages and mortgages on nonowner-occupied buildings) in the tests de-
stroyed their usefulness,
s~
(Nevertheless, the data used need not be biased with respect to
the hypotheses tested.) Second, despite the large number of observations available, a valida-
tion sample was not used. Third, it is likely that many of the variables (such as the many
measures of neighborhood quality and the applicants income) are related to the variables
of interest, property location and the applicants race. Consequently, there is no way of
determining the statistical significance of the coefficients. This problem is partic!~larly acute
for the New York City property location data. Fourth, the omitted variable problem may be
particularly important to the measured relationship between the applicants race and the
probability of denial (as Schafer acknowledges in the quotation given in the introduction to
this section in the text above). In particular, race may be a determinant of foreclosures,
given an applicants income, etc., if blacks are discriminated against in employment or are
more likely to suffer from illness or marital problems and hence are more likely to default
on their mortgages.
6
Should this be the case, society rather than lenders bears much, if not
all, of the responsibility for the higher denial rates for black applicants reported. Fifth, the
number of blacks whose applications were denied are small: 97 in the New York City area,
45 in Buffalo, 6 in Rochester and 5 in Syracuse. Based on these numbers, Schafer concludes:
"The race of the applicant as opposed to tlie racial composition of the neighborhood sur-
rounding the property, is a crucial factor in the lending decisions of all but one of the
metropolitan areas... [This finding is] consistent with allegations of discrimination against
black applicants." [p. 13 -111 ]. In any event, "black" could be a proxy for other, risk-related
variables and not a discriminator, as such, with reasoning similar to interpretations of the
positive relationship between "black" and high defaults. Finally, because all the data were
obtained from mutual savings banks, the alternative hypotheses that they discriminate
against blacks or tend to favor blacks cannot be distinguished, if it is the case that the other
variables used in the model are inadequate measures of the quality Of the factor specified.
For example, the quality of earnings may be lower for blacks because they are discriminated
against in employment. But if real estate agents know that mutual savings banks (or the ones
sampled which, it should be noted, were willing to give Schafer their data) tend to favor
blacks, these banks will get a disproportionate number of applications (relative to other
lenders). In this event, though the probability of a black being accepted is higher at the
subject banks than at other lenders, the regression analysis will indicate the reverse.
Sgln total, over the five municipalities, applications excluded because of missing data
are 20 percent of the universe (conventional mortgages on owner-occupied houses) and
applications included in the tests that are definitionally not part of the universe are 3 8
percent of the universe.
60 Barth, Cordes, and Yezer [1979] report a statistically significantly higher probability
of defaults on FHA loans by blacks of 1.2 percentage points. (The overall average default
is 1.5 percent.) As they state, this finding does not mean that blacks necessarily are poorer
risks but that "black" may be associated with other, omitted variables that are positively
related to defaults.
Warner and Ingram [1979] (Columbia, South Carolina). The data on conventional
mortgages applied for were obtained from all four of the savings and loan associations and
three of the four commercial banks with headquarters in the area. (One bank was excluded
because it concentrated its conventional mortgage lending outside the area.) Over the period
studied (1976 and 1977) all the 250 denied applications were analyzed and a randomly
selected sample of 500 accepted applications was drawn from a population of about 4600.
(The denial rate of 5.4 percent is lower than found in the New York mutual savings banks
data. No mention is made by Warner and Ingram of modifications.) A validation sample
was derived by selecting every other observation of each group.
A multiple discriminant model was used first to estimate only those factors that were
considered to represent ordinary, nondiscriminating risk and return considerations (step I).
Then the function was reestimated with a set of factors defined as discriminatory added to
the first set (step II). As the authors explain: "While neither function may be a very effec-
tive model of the loan offer function, the key to the discrimination analysis rests in the fact
that unless there is a significant improvement in discriminating power moving from Step I
to Step II the discrimination variables representing particular categories of borrowers are
eliminated as possible residual variables (with respect to the first discriminant function)."
[p. 41
The first (step I) discriminant function includes the following variables (listed in the
order of the relative discriminating power estimated): applicants credit rating, loan-to-value
ratio requested, annual percentage (Regulation Z) interest rate, applicants tenure in current
occupation, applicants total monthly payments-to-income ratio, applicants occupation
(percentage of income earned from sales commissions), term of mortgage applied for, and
co-borrowers tenure in current occupation (0 if no co-borrower).
~1
It is statistically signifi-
cant at the .01 level. The additional, "discrimination" variables include the following (listed
in the order of relative discrimination power estimated): applicants age, neighborhood
median family income in 1970, marital status (unmarried or separated = 1, married = 0),
age of the dwelling, applicants sex (1 for female or other than married couple applying
jointly, 0 otherwise), applicants race (1 for non-Caucasian, 0 otherwise), and co-applicants
income as a percentage of total income.
62
A comparison of the discriminating power of the
step I and step II functions reveals no significant difference at the .01 level. (Quadratic dis-
criminant functions also were computed, leading to the same conclusion.) Furthermore,
the coefficients estimated were used to classify the validation sample. The step I linear
model classified 68 percent correctly, as did the linear step II model which included the
"discrimination" variables. The quadratic step I model classified 76 percent correctly, com-
pared to 66 percent for the quadratic step II model. Thus, as the authors conclude, "...
there is no evidence of mortgage lending discrimination in the data." [p. 3 0]
King [1979B] (Miami, San Antonio, and ToledoJ. The data were obtained from regis-
ters of loan applications kept at 45 savings and loan associations for the months of Septem-
ber through November, 1978. The sample sizes are 23 97 for Miami, 546 for San Antonio
and 83 1 for Toledo. About half as many observations had to be omitted because of missing
data in the San Antonio and Toledo samples and about a fourth as many in the Miami sam-
pie; however, a comparison of the means of some variables of the used and omitted samples
indicates no evidence of bias. The dependent variable equals 0 if an application was accepted
and 1 if it was rejected. Ordinary least squares and logit analyses were used, with little sub-
~lThe applicants net worth was specified by the model but the data could not be
obtained.
~2Data limitations precluded inclusion of four discrimination variables specified:
applicants religion, applicants national origin, public assistance income, and applicants
previous exercise of consumer protection rights. The following are the means of the
"discrimination" variables in the accepted and rejected applications samples: age, 3 7 and
3 4 years; unmarried or separated, 11 and 12 percent; building age, 7 and 10 years; female
or other than married, 21 and 11 percent; non-Caucasian, 11 and 20 percent; andco-
applicants income percentage, 9 and 8 percent~
191
stantive difference. The independent variables included the following: mortgage characteris-
tics (FHA-VA and private mortgage insurance as dummy variables, and the ratio of the re-
quested loan amount to appraisal value), borrower characteristics (the applicants credit
rating, ratio of monthly payment to the applicants gross monthly income and the ratio
squared, ratio of the purchase price of the house to the applicants gross monthly income
and the ratio squared, ratio of the requested mortgage loan to the applicants net worth,
ratio of applicants existing monthly debt to their gross monthly income, the applicants age
and age squared, ratio of second income to total income of applicants, and dummy variables
equaling 1 if the applicant is black: Hispanic, other minority, female, or unmarried), prop-
erty characteristics (age of house and age squared), neighborhood characteristics (by census
tracts, 1970 data - percent black, percent Spanish-American, percent of households having
income below the poverty level, percent owner-occupied units, percent owner-occupied units
built before 193 9 and, for Toledo, 29 dummy variables designating defined neighborhoods),
and dummy variables for each savings and loan association, less one.
The most important variables generally are those representing the mortgage characteris-
tics, the borrowers characteristics of bad credit rating, other debt to income, payments to
income (except for San Antonio and some savings and loan dummy variables).
Because the applicants credit rating variable was recorded only for declined applica-
tions, all accepted applications were considered to be from applicants with good credit
ratings. But, since this coding may not be correct, King reports regressions including and
excluding this variable. The statistically significant coefficients of the variables that may
measure discrimination (e.g., black, Hispanic, other minority, female, unmarried, census
tract black, Spanish-American, and units built before 193 9) found are as follows:
Miami - The denial rate averaged 12 percent. Inclnding bad credit rating (R
~
of .22),
the average Hispanic applicant had a higher denial rate of 3 percentage points and census
tracts with 50 percent more Spanish-Americans are associated with denial rates averaging 6
percentage points higher; excluding bad credit rating (R
2
of .11), the two variables were not
changed much but blacks also averaged high denial rates of 7 percentage points (significant
at the .10 level).
San Antonio - The denial rate averaged 5 percent. Including bad credit rating (R
~
of
.3 0), the average Hispanic applicant had a higher denial rate of 3 percentage points, which
increased to 5 percentage points when the bad credit rating variable was omitted (R: of .11).
Toledo - The denial rate averaged 7 percent. Including bad credit rating (R
2
of .23 ) the
probability of denial declines until houses are 21 years of age and then increases, and 2 of
the 29 city neighborhoods have lower denial rates of 22 and 19 percentage points; excluding
bad credit rating (R
~
of .15), the age of house and neighborhood coefficients are largely
unchanged and the average Hispanic applicant had a higher denial rate of 20 percentage
points, but this finding is based on only five observations.
In summary, the findings show evidence of adverse selection only for Hi~panic bor-
rowers in Miami and San Antonio averaging about 3 percentage points higher rates of denial
(relative to average rates of 12 and 5 percent). The data show no evidence of discrimination
against black, other minorities, females, old, or unmarried applicants, applicants for mort-
gages on older houses or in neighborhoods occupied by blacks, Spanish-Americans, or in
Toledo, city versus suburb.
Tyree and Yeager [1979, Chapter 10] (St. Louis). All 194 loans referred in 1978 to
the St. Louis Savings Service Corporation, an organization established and financed by area
saving and loan associations to consider "high risk" residential mortgage applications that
previously were reviewed and rejected, were analyzed. Multiple discriminant analysis was
employed to determine whether use of possible "discrimination" factors played a role in
the corporations rejection of 3 4 percent of the applications. These factors include location
of the property, and the applicants race, marital status, age, and sex as well as other factors,
such as the price of the house and housing expense. The authors conclude: "The most im-
portant conclusion from this analysis is the insignificance of the entire model. The 18 fac-
tors, when considered as a group, provide no improvement over pure chance in determining
the likelihood of an application being accepted or rejected .... The conclusion from this
analysis is that decisions by the Savings Service Corporation are primarily influenced by
other factors, and not by factors relating to discrimination, particularly redlining." [pp.
119-120] This finding is gratifying, but considering the purpose for which the Corporation
was established, not surprising.
The Demand for Mortgages by Actual and Potential Home Buyers
Benston and Horsky [1979] (Rochester, New York). Surveys of present (1976) home-
owners and buyers of houses within the previous three years were undertaken to determine
the mortgage-related difficulties they encountered in selling and buying their homes and in
obtaining home improvement loans. The experiences of people in the allegedly redlined area
(identified by anti-redlining community group leaders, essentially as the central city) were
contrasted with those of people in a control, suburban area. Based on a pilot study, 500
central city and 400 suburb homeowners were selected randomly for interviews. The inter-
views were conducted by a professional market research firm to reduce the possibility of
bias. The homeowners were asked whether they had attempted to sell their homes within
the last five years and/or currently were trying to sell their homes. Twelve percent of the
central city owners (45 people) and 5 percent of the suburban owners (16 people) contacted
said yes.
63
Of these, no potential buyer had inspected the homes of 3 6 percent of the cen-
tral city owners and 13 percent of the suburban owners. The owners reported that, of those
potential buyers who had inspected the house, a somewhat greater proportion of those in
the central city than in the suburb said they encountered mortgage problems (28 vs. 21
percent). However, none of these problems was stated to be related to the area of the house.
Buyers of homes in both areas over the three years 1974 through 1976 were identified
by sampling sales reported in the Rochester Daily Record; 260 people in the central city and
273 people in the suburb who had purchased homes for their own use were interviewed and
asked about their experiences in obtaining financing. Mortgages were obtained by 92 per-
cent of the central city buyers and 99 percent of the suburban buyers. All but 0.4 percent
(one person) who wanted a mortgage did not get one; the balance said that they did not
want a mortgage. Relatively more suburban buyers than central city buyers obtained a mort-
gage from a financial institution (99 vs. 87 percent). Of these, a mortgage was obtained from
the first institution contacted by 84 percent of the central city and 89 percent of the subur-
ban mortgagors. Most of the balance chose to use another institution. Mortgages were
refused by a lending institution to 2 percent of the central city buyers (four people) and
none of the suburban buyers. One of the four buyers said that the area of the home was the
reason for refusal.
The buyers reasons for obtaining conventional or FHA mortgages also were ascertained.
(Conventionals were held by 29 percent of the central city and 62 percent of the suburban
home buyers; FHAs were held by 55 percent of the central city and 18 percent of the subur-
ban home buyers.) The principal reason given for obtaining FHA mortgages was lower down-
payment and the principal reason for the choice of conventional mortgage was lower interest
rate. The only reason that might indicate a forced choice, "only type available," was given
for FHAs by the same percentage in each area (2 percent). Conventionals were said to be
the only type available by 3 percent of the central city and 7 percent of the suburban home
buyers. Finally, all but 3 percent of the central city and 2 percent of the suburban buyers
said they were dissatisfied with the financing they obtained. Also no area-related problems
with respect to home improvement loans were reported.
Tyree and Yeager [1979, Chapter 8] (St. Louis, Missouri). This study was based on
interviews with leaders of neighborhood organizations and follow-ups of their allegations of
redlinlng. All of the leaders of the 194 organizations listed by the citys Community Devel-
opment Agency that were still in existence were contacted. Repeated attempts to obtain
interviews resulted in 104 interviews. The interviews were standardized and controlled to
minimize bias. The real estate practices of real estate lending institutions were said to have
6aWhen adjusted for the total estimated number of selling attempts over the five-year
period, no buyers inspected the house in 14 percent of the central city attempts and 2 per-
cent of the suburban attempts.
REDLINING RESEARCH
BENSTON 193
a negative effect by 3 8 percent of the community leaders, and a positive or neutral effect
by 58 percent (4 percent replied "dont know"). The agreement by community leaders
who thought that lenders had a negative effect compared to those who said the institutions
had a neutral or positive effect to the following series of statements provides insight into the
basis for their belief. "An applicant who would be able to obtain financing on property out-
side the city of St. Louis would find financing for St. Louis property quite difficult or near-
ly impossible," 3 0 vs. 5 percent. "Interest terms in this neighborhood are less favorable than
terms in other parts of the city," 3 8 vs. 16 percent, "and other parts of the county," 54 vs.
3 3 percent. "The general attitude of lenders regarding real estate financing in their neighbor-
hood is discouraging," 46 vs. 14 percent, and "is less favorable than five years ago," 20 vs.
14 percent. When asked if they knew of any lending institutions that actively tried to get
real estate loan business in the neighborhood during the last three years, "yes" was given
by 41 percent of the "negative" leaders vs. 61 percent of the "positive and neutral" lenders.
The interviewers also asked questions about redlining. When asked whether they knew
of "any specific cases of lending institutions not approving loan applications because of
the location of the property," 41 percent of the community leaders who said that lending
institutions have a negative effect replied "yes" compared to 18 percent of those who said
the institutions have a positive or neutral effect. When asked "Do you know ~3 f any specific
cases of lending institutions purposely discouraging applicants from buying in this neighbor-
hood," yes was answered by 16 percent of the first (negative) group and 4 percent of the
second group. When asked about the institutions discouraging applications, 8 percent of
the both groups said they knew of specific cases. In all instances where the respondent
answered "yes" to these three questions, the interviewers asked for the number of instances
and the names and addresses of the people involved. (One person expressed knowledge of 25
cases.) Repeated attempts were made to obtain this information. The researchers report:
"In general, neighborhood association leaders were cooperative and attempted to assist in
[the] search for redlined parties. Many discussed our search during their neighborhood
organization meetings and went so far as to contact individuals who may have had diffi-
culties in seeking financing." [p. 99] However, only seven alleged cases could be found.
Two cases included people who could not be located; one was said to have been denied a
mortgage five years earlier and one a year earlier. In four cases the people denied they had
had a problem and in one case the problem appeared to be lack of creditworthiness because
of the husbands low income, though the people got financing elsewhere and then divorced.
(Details are provided in the study.) Thus little evidence of actual redlining in St. Louis
was reported and none was documented.
References
Agelasto, Michael II and Listokin, David. 1977. "Redlining in Perspective: An Evaluation of
Approaches to the Urban Mortgage Dilemma." In A Decent Home and Environment:
Housing Urban America, edited by Donald Phares, Calnbridge, Mass.: Ballinger Pub-
lishing Co.
Ahlbrandt, Roger S., Jr. "Exploratory Research on the Redlining Phenomenon."AREUEA
Journal 5 (Winter 1977), 473 -481.
Barth, James R., Cordes, Joseph J., and Yezer, Anthony M. 1979. "Financial Institution
Regulations, Redlining and Mortgage Markets," paper presented at the Conference on
Financial Institution Regulation, sponsored by the National Science Foundation and
the Federal Reserve Bank of Boston, October.
Benston, George J. 1977. "Urban Decline and Redlining." In Innovations and New Direc-
tions in the U.S. Financial System: Implications for Real Estate Markets and Invest-
ments, Real Estate Chair Lecture Series, edited by John M. Clapp, Housing Real Estate
and Urban Studies Program, Urban Resources Center, Graduate School of Management,
University of California, Los Angeles, pp. 62-84.
__. 1978A. "The Persistent Myth of Redlining." Fortune, March 13 , 66-69.
194
__. 1978B. "The Urban Financing Dilemma Reconsidered." Urban Housing: Proceedings
of the Fourth Annual Conference, Federal Home Loan Bank of San Francisco, Decem-
ber 7-8,102-149.
___ and Horsky, Dan. 1979. "Redlining and the Demand for Mortgages in the Central City
and Suburbs." Journal of Bank Research, Vol. 10, No. 2, (Summer), 72-87.
__, Horsky, Dan and Weingartner, H. Martin. 1978. An Empirical Study of Mortgage
Redlining. Monograph 1978-5, Monograph Series in Finance and Economics, New
York University Graduate School of Business Administration, Solomon Brothers
Center for the Study of Financial Institutions.
Berry, Brian J.L. 1976. "Ghetto Expansion and Single-Family Housing Prices: Chicago
1968-72." Journal of Urban Economics, 3 97-423 .
Black, Harold, Schweitzer, Robert L. and Mandell, Lewis. 1978. "Discrimination in Mort-
gage Lending," American Economic Review, 68 (May), 186-191.
The Center for New Corporate Associates. 1975. Where the Money Is: Mortgage Lending,
Los Angeles County. Reprinted in United States Senate, Hearings, 1063 -1120.
Devine, Richard J. with Rennie, Winston O. and Sims, N. Brenda. 1975. "Where the Lender
Looks First: A Case Study of Mortgage Disinvestment in Bronx County, 1960-70," re-
produced in United States Senate, Hearings, 1267-1476.
Dingemans, Dennis. 1979. "Redlining and Mortgage Lending in Sacramento." Annals of the
Association of American Geographers, 69 (June), 225-23 9.
Erie County Citizens Organization, Inc. 1975. Redlining and Disinvestment in Buffalo.
Hauser, Philip M. and Elkhanialy, Hekmat. 1978. Lending Practices of Savings and Loan
Associations in Chicago - 1977, report prepared for Federal Savings and Loan Council
of Illinois, Chicago: Leo J. Shapiro and Associates, September.
House Ownership and the Baltimore Mortgage Market. 1975. Home Ownership Development
Programs, Department of Housing and Community Development, reprinted in United
States Senate, Hearings, 1478-1623 .
Hutchinson, Peter M., Ostas, James R. and Reed, J. David. 1977. "A Survey and Compari-
son of Redlining Influences in Urban Mortgage Lending Markets." AREUEA Journal, 5
(Winter 1977), 463 -472.
King, A. Thomas. 1979A. "Redlining: A Critical Review of the Literature with Suggested
Research." Research Working Paper No. 82, Office of Economic Research, Federal
Home Loan Bank, February.
__. 1979B. "An Analysis of Mortgage Lending in Three SMSAs." Office of Economic
Research, Federal Home Loan Bank Board (preliminary report).
Koebel, Theodore. 1978. Housing in Louisville: The Problems of Disinvestment. Louisville,
Ky.: Urban Studies Center, University of Louisville (July).
Marcis, Richard G. and Hull, Everson W. 1975. "Analysis of the Socio-Economic Deter-
minants of Foreclosures on 221(d)(2) and 23 5 Mortgages," Federal National Associa-
tion Meeting.
Morton, T. Gregory. 1977. "A Discriminant Function Analysis of Residential Mortgage
Delinquency and Foreclosure," AREUEA Journal, 3 (Spring), 73 -88.
Muth, Richard F. 1979. "Yields on Inner-City Mortgage Loans." Federal Home Loan
Bank of San Francisco. Unpublished paper (January).
Ostas, James R., Reed, J. David and Hutchinson, Peter M. 1979. "An Intertemporal Com-
parison of Urban Mortgage Lending Patterns in the Toledo, Ohio SMSA: 1977 vs 1975."
Unpublished paper, Bowling Green State University.
Schafer, Robert. 1978. Mortgage Lending Decisions: Criteria and Constraints, two volumes,
Cambridge, MA: Joint Center for Urban Studies of the Massachusetts Institute of
Technology and Harvard University, (December 7).
Touche, Ross & Company. 1975. The Costs of Mortgage Loan Foreclosure: Case Studies of
Six Savings and Loan Associations. Washington, D.C.: Federal Home Loan Bank Board.
Tyree, Donald A. 1979. "Urban Residential Mortgage Financing: Lending Practices in St.
Louis." St. Louis, MO.: Department of Finance, School of Business Administration,
St. Louis Universtiy (June 6).
United States Senate, Committee on Banking, Housing, and Urban Affairs. 1975. Hearings
on the Home Mortgage Disclosure Act of 1975. S.1281, May 5, 6, 7, 8.
195
Von
Furstenberg, George M. 1969. "Default Risk on FHA - Insured Home Mortgage as a
Function of the Terms of Financing: A Quantitative Analysis." Journal of Finance,
XXIV (June), 459-477.
__and Green, R. Jeffrey. 1974. "Estimation of Delinquency Risk for Home Mortgage
Portfolios." AREUEA Journal 2 (Spring), 5-19.
Warner, Arthur E. and Ingram, Franklin J. 1979. "Mortgage Lending Discrimination: A
Stepwise Discriminant Analysis Approach." College of Business Administration, Center
for Real Estate Urban Economic Studies, University of South Carolina. Unpublished
paper, September.
Whalen, Thomas A. 1976. "An Analysis of Mortgage Lending Activity in Flint, Michigan,"
in Final Report of the Governors Task Force on Redlining. Richard J. Francis, Chair-
person, Department of Commerce, State of Michigan, December, 51-58.
Williams, Alex O., Beranek, William and Kenkel, James. 1973 . "Factors Affecting Mortgage
Credit Risk in Urban Areas - A Pittsburgh Prototype Analysis," a paper presented at
the American Real Estate and Urban Economics Association Meeting, December 29.
Reprinted in United States Senate, Hearings, 628-657.
Discussion
Robert Schafer*
It is important to emphasize that the profitability of a particular approach
to evaluating credit risk is not an appropriate test of whether the society, through
its government, is justified in regulating the credit evaluation process. Private
property is an aggregate of rights created by society, and society has always
placed limits on its use. Similarly, society can add a requirement of equal treat-
ment of similarly situated applicants as a constraint on the market. Such rules
define the context within which business must be conducted. They can, and
frequently do, alter the profitability of business. Society need only believe
reasonably that the extra cost of doing business within the new constraints is
less than the social benefits resulting from the fair and equal treatment of
applicants.
In essence, the fair housing regulations require lenders to use objective
measures of risk of loss for each application and prevent them from attributing
the average behavior of some group to the individual. These rules are merely the
reflection of an American ideal: each person should be judged on his or her own
merits. In particular, fair housing laws prohibit discrimination on the basis of
race, color, national origin, sex, marital status, age, receipt of income from
public assistance programs, and good faith exercise of rights under the Consumer
Protection Act. As a result, a black applicant for a mortgage must receive the
same treatment and have the same chance of success as a white applicant who
has the same income, net wealth, credit history and value of other objective
measures of creditworthiness. Other regulations place restrictions on the use
of neighborhood. For example, the Federal Home Loan Bank Board (FHLBB)
prohibits "arbitrary decisions based on age or location of the dwelling.
~
Al-
though the Community Reinvestment Act does not place direct constraints on
the lending process, it requires the federal financial regulatory agencies to
encourage lenders "to help meet the credit needs of the local communities in
which they are chartered consistent with the safe and sound operation of such
institutions."
It should be no surprise that these regulations might increase the costs of
evaluating credit applications. Using the average risk of loss for various groups
t 12 CFR 53 1.8(c)(6), 43 Federal Register 223 3 9 (May 25, 1978).
*Robert Schafer is currently practicing law in Boston. From 1971 to 1979 he was an
Associate Professor of City and Regional Planning at Harvard University and a faculty
associate at the MIT-Harvard Joint Center for Urban Studies.
196
DISCUSSION
SCHAFER
197
of applicants is usually less expensive than researching each applicants credit
history.
The next section presents my comments on the Barth, Cordes and Yezer
paper. The following section discusses the Benston literature survey.
Barth-Cordes-Yezer
The Barth-Cordes-Yezer papers distinctions between a legal and economic
definition of redlining are misleading. Any differences that do exist are consider-
ably less than they suggest, especially regarding the anti-redlining regulations.
The most important way in which the Barth-Cordes-Yezer paper misleads
its readers is the suggestion that the fair housing regu!ations require lenders to
make loans that are unsafe. This follows from their economic definition of
redlining: differential treatment of neighborhoods that does not correspond to
differences in costs and risks. Increased costs due to a more individualized credit
evaluation process should not be a part of the definition of redlining.It is not
clear why the costs of processing and servicing should vary across mortgages that
have equal risk of loss. Then, the major question turns on the implications of fair
housing laws for the ability of lenders to use risk of loss. None of the fair housing
regulations restrict a lenders ability to use risk of loss in deciding whether and
on what terms to approve a mortgage application. These laws only require that
lenders use objective measures of each applications risk of loss. Such measures
must have a causal relationship with risk, and not just a correlation. In this sense,
the laws can be credited with keeping economists honest; that is, requiring the
models to be specified correctly. Anti-redlining laws even allow some aggregate
neighborhood measures to be employed. For example, the FHLBB regulations
only prohibit the arbitrary use of location. These conditional rules recognize the
role of housing market externalities; for example, the condition of neighboring
buildings is likely to affect the risk of loss on the property that is securing a
mortgage.
Although the fair housing laws contain unconditional prohibitions on the
use of race, sex and other personal characteristics, the anti-redlining laws can be
read as asking for help in defining what is an unfair use of neighborhood factors.
Here is a clear opportunity for economists to make a valuable contribution.
Anti-redlining laws are begging for a workable definition of the "arbitrary" use
of neighborhood. How can property externalities be reasonably used in the
lending process? What should regulators do to determine what is arbitrary and
what is not? What data are needed and what techniques are appropriate for its
analysis? Lenders need to know what neighborhood factors are legitimate credit
evaluation criteria and how they can incorporate them into an effective and
legally defensible credit evaluation process.
It is relatively easy to delineate some neighborhood factors that cannot be
utilized. These are the aggregate measures of the personal characteristics
which the fair housing laws prohibit unconditionally. For example, the racial
composition of a neighborhood is clearly prohibited. The laws tell lenders to
find objective measures of risk which are reasonably related in a causal sense
with risk of loss. Racial composition may be correlated with risk of loss but
society has correctly stated that any such relationship is spurious and not founded
on a causal link. Economists can and should develop the causal relationships for
use by lenders and regulators. A very likely candidate for a causal link with risk
of loss is the incidence of housing code violations and vacant buildings. A useful
study would be an assessment of the relative strengths of various measures of
housing condition (e.g., census tract, census block, and adjacent property) as
indicators of risk of loss in mortgage lending. The results would provide a firm
foundation for defining the area impacted by housing market externalities.
It is also unfortunate that Barth, Cordes and Yezer misinterpret a clause
of the Community Reinvestment Act that requires federal financial regulatory
agencies to "assess the institutions record of meeting the credit needs of its
entire community, including low- and moderate-income neighborhoods, con-
sistent with the safe and sound operation of such institution" in reviewing
applications for deposit facilities. The authors imply that this provision of the
CRA might prevent lenders from using neighborhood income as a criterion.
Their interpretation is uncalled for and unfounded. It is far more reasonable to
read this section of the CRA as requiring lenders to participate in appropriate
public programs to assist low-income neighborhoods. These programs would
provide the risk adjustment necessary to make the ventures safe and sound for
the lender. Thus, the statute would be consistent with economic reality.
Furthermore, Barth, Cordes and Yezer suggest that the economic but not
the legal definition provides a standard against which lender behavior can be
compared. I cannot agree. The law defines the boundaries of economic activity.
Economic activity is not possible without some agreement about the procedures
for conducting activity involving more than one party. This is the essence of law.
In this case the law requires equal treatment of similarly situated applications.
Economics has a role to play in defining the objective measures that need to be
held constant to assess whether or not there is discrimination as defined by
the law.
Barth, Cordes and Yezer also mislead the reader by incorrectly claiming that
the law views redlining as occurring when local lenders export deposit funds.
Many community organizations espouse this view, but it is not the law. The
closest these groups have come to incorporating this view into the law is the
Home Mortgage Disclosure Act of 1975, but it only requires disclosure of
lending activity. It is only a source of data for use in evaluating whether lenders
treat all neighborhoods equally. The problem is that the data are wholly inade-
quate for the task. Another problem is that economists, especially those working
for regulatory agencies have not provided sufficiently strong backing for the
appropriate data source: namely, detailed information on all serious mortgage
inquiries. These data should be even more detailed than the information available
under California and New York law. The Home Mortgage Disclosure Act of 1975
should be repealed and replaced with a law requiring the maintenance of detailed
records on all serious inquiries about mortgages. Then, all parties will have access
to data capable of answering the basic empirical question: Does discrimination
occur and, if so, who hre adversely affected?
DISCUSSION
SCHAFER 199
In the area of methodology, I have several points on the Barth-Cordes-Yezer
paper. First, a logit model is preferred to the OLS technique used because the
logit has the appropriate functional form. This choice is not a matter to be
resolved on the basis of sample size. Of course, the simultaneous equation model
has less econometrical foundation in the logit formulation because there is no
known way of introducing an error term into the logit function. It should be
noted that under certain assumptions the binomial choice OLS coefficients can
be transformed into consistent logit coefficients.
Second, the specification of the models is bothersome. The housing expen-
diture equation appears to be a confusing mixture of a consumption function
and a hedonic price index. The maturity equation does not include the age of
the borrower which would seem to be a relevant and important preference vari-
able. Similarly, the loan-to-value equation excludes two relevant and important
preference variables: net wealth and age of borrower.
Third, I believe that the authors misinterpret the coefficients of variables
which are unconditionally prohibited as lending criteria. A positive and signifi-
cant coefficient on such a variable (e.g., race) indicates that defaults are higher
for this type of person but is not a defense against or a valid basis for objecting
to regulations prohibiting discrimination. Instead, it is an indication that objec-
tive measures of default have been excluded from the model. In an appropriately
specified default model, it should be unnecessary to brand an individual with the
average behavior of a group.
Fourth, city characteristics are an inappropriate level of aggregation to
study redlining. If lenders redline, they do it at the neighborhood level. In other
words, redlining occurs, if at all, across neighborhoods within one metropolitan
area and not across metropolitan areas. This is a serious weakness.
Finally, I believe that their discussion of a sample selection bias is incomplete.
They argue that sample selection bias exists because the observed defaults are
drawn from a population with an ex ante default which is less than a critical
value. This is not true because lenders do not know the ex ante default proba-
bilities; instead,.lenders use estimates, or predicted values, of default prob abilities.
I believe this results in less efficient but not biased estimates when the dependent
variable is ex post.
The Barth-Cordes-Yezer paper makes several valuable contributions:
First, it tests defenses against marginally discriminatory factors such as age
of neighborhood. Community organizations argue that older neighborhoods are
redlined, that is, receive less funds than other neighborhoods with equal risk of
loss. Lenders have responded that such variations are spurious because neighbor-
hood age is highly correlated with risk of loss. The Barth-Cordes-Yezer results
indicate that this defense is invalid; they find that default is significantly less
likely in older neighborhoods.
Second, it allays lenders fears concerning the impact of fair housing laws.
For example, any lender who may have relied on the age of the building or the
racial composition of the neighborhood should not be concerned by their prohi-
bition as lending factors; the Barth-CordesoYezer results indicate that they do
not have a significant impact on default once other factors are taken into account.
In fact, other more objective measures of risk (e.g., the structural condition of
the building and of nearby buildings) are even better and not subject to charges
of discrimination.
Third, it contributes to the growing literature that provides a foundation for
the development of a credit evaluation system to make lending decisions and to
systematically adjust the terms (e.g., interest rate) for risk differentials. These
models must be based on default models that exclude all variables that measure
prohibited lending criteria.
Benston
Benston has tried to summarize the empirical literature available on the
issue of redlining. This is a tremendous task requiring careful reading of many
studies using widely different approaches. One valuable improvement would be
a table summarizing the results of each study with respect to redlining; it is very
difficult to assimilate these findings across the textual summary of the studies.
Persons interested in literature reviews of these studies should also consult the
excellent one by Thomas King of the Federal Home Loan Bank Board.
Benston asserts three conditions that studies of redlining must meet to have
valid conclusions. He does not require each study to meet all three conditions.
One of these conditions is plainly incorrect. He asserts that if the focus of the
study is on the effects of redlining on consumers or neighborhoods, data from
all mortgage lenders must be included. While this appears reasonable on its face,
it is inconsistent with the requirements of fair housing laws. These laws do not
require that on average a particular group of consumers such as blacks receive
adequate mortgage funds. They require that no lender discriminate against them
in the lending process. A lender that discriminates against a black applicant or
some other protected group will not find a successful defense in the fact that
another nondiscriminating lender granted that person a loan.
Bank specific studies are precisely the type necessary for regulators to make
determinations relevant to the requirements of fair housing laws. Economists
should assist them by developing pragmatic and sound methodological ap-
proaches. These analyses should also assist lenders by providing them with
credit evaluation systems that are based on objective factors with causal links
to risk of loss. The most difficult part of this challenge lies in developing a work-
able definition of redlining. Redlining based on the racial composition of the
neighborhood or some other neighborhood aggregate measure of an individually
protected status would have to pass the same standards as the discrimination
against a member of one of these protected groups. However, redlining based
on the age of the neighborhood or the geographic boundaries presents a more
complex case. Yet even here, fair housing laws will probably be read as requiring
lending rules based on a causal link between the variables and risk of loss for any
areas within each lenders reasonable lending market. The definition of this
market area is another place for economists to assist the lawyers.
It is true, however, that studies of the aggregate supply and demand for
mortgages should include most if not all sources of mortgage funds. This does
not, however, eliminate the usefulness of models based on aggregate data. In my
DISCUSSION SCHAFER 201
own study of three counties in New York City, I went to great lengths to point
out the importance of interpreting my results as only applying to those institu-
tions covered by the data set.
2
In his discussion of default models, Benston argues that meaningful findings
are only possible if the data were not prescreened by lenders. This is incorrect
because, as I have already pointed out in my discussion of the Barth-Cordes-Yezer
paper, lenders are making decisions on the basis of estimated and not the known
probabilities of default. As a result, econometric procedures can provide valid
estimates of the effect of variables on default. There is a similar, but unrelated
issue that should be mentioned. In any econometric technique, the predictions
weaken the further the situation is from the range of values of the independent
variables available in the data used to estimate the model.
In discussing default studies, Benston concentrates on foreclosures. The
wide variation in bank policies towards the institution of foreclosure proceedings
suggests that delinquency may be a better measure of risk of loss. In my own
studies the crude foreclosure and delinquency rates suggested that risk was
higher in areas alleged to be redlined, but the multivariate analyses indicated that
many factors other than location may be responsible for the higher default rates
in these areas and that lenders can develop credit evaluation systems using
information specific to the property and applicant without recourse to crude
neighborhood-level rules of thumb.3
In discussing the studies of the lender decisions on applications for mort-
gages, Benston states that "all of the [se] studies found no evidence of differential
denial rates related to" the propertys location. This is incorrect. My own study
of the Albany-Schenectady-Troy metropolitan area found strong evidence that
two neighborhoods were redlined by savings banks.
4
In his summary, he makes the further statement that "the weight of the
evidence is contrary to the hypothesis that lenders discriminate against minority
borrowers or areas by denying mortgage applications." This clearly incorrect
statement could seriously mislead the reader and reveals a bias that runs through-
aUnfortunately, Benstons summary is written so that the reader may have the mis-
understanding that this is his caveat while it is, in fact, my own. In addition, Benston states
that my study of the quantity of mortgage money provided by reporting institutions
(Chapter 5) found "no evidence of redlining." This is incorrect. The report concluded with
respect to the analysis of Chapter 5 that "the community organizations may be incorrect
about the existence of redlining in [the Central Brooklyn and Park Slope] neighborhoods,
while they may be correct in their allegations concerning the South Bronx, Crown Heights,
East Flatbush and Southeast Queens." (p. 5-78).
3 My delinquency models include variables that identify whether or not the property is
located in a neighborhood that is alleged to be redlined. Benston also points out that the
borrowers race and the racial composition of the neighborhood were absent from my
default models. The reasons are that the race of the borrower was not available (even though
we sought it) from the banks records and the racial composition is only available for 2
years of the 27-year period covered by the mortgages in the samples used to study default.
4In addition, the chances of modification of the requested terms prior to approval
were significantly higher in the allegedly redlined Park Slope neighborhood in New York
City.
202
out the literature review. Although the evidence indicates that redlining exists in
only some metropolitan areas, the same studies find that minority applicants,
especially blacks, are twice as likely to be denied a mortgage as are similarly
situated white applicants. These findings are very statistically significant, and the
differentials are large. And contrary to Benstons assertions, the studies are not
beset by problems of multicollinearity and the models are specified so as to
control for risk-related and other objective lending criteria.
5
These findings can-
not be swept aside so easily. Lenders and society must address these findings
honestly and forthrightly. We need to develop lending procedures and a lending
officer training process that eliminates racial discrimination.
A bias against findings consistent with allegations of discrimination is evident
to anyone familiar with the studies being reviewed. If the care that was devoted
to identifying deficiencies in the studies with findings consistent with discrim-
ination had also been devoted to the other studies, the latter would have been
found to be seriously flawed relative to the former. For example, the Muth
study of mortgage terms analyzes an obvious simultaneous equation system with
an entirely inappropriate econometric tool. Another example is the differential
insertion of the effect of potential multicollinearity. This criticism appears in
the discussion of findings consistent with discrimination allegations whether or
not it is appropriate. However, it is absent from other studies even though
appropriate. The Warner-Ingram two-step procedure, for example, is entirely
inappropriate and invalid in the presence of multicollinearity.
None of the research adequately deals with the question of whether lenders
discriminate illegally in their response to initial inquiries, the prescreening
process. Homeowner or borrower surveys shed no meaningful light on this
issue. Some agency should conduct an audit study of this portion of the lending
process. Such a study would have matched pairs of applicants that differ on only
one characteristic such as race make inquiries of the same lender. Comparison of
their respective treatment would be a significant addition to the literature.
Among other things, such a study would assist financial regulatory agencies in
deciding whether or not to make such audits a regular feature of their equal
credit opportunity enforcement efforts.
s For example, Benston suggests that race may be a determinant of foreclosure. Of
course, that could not be a causal relationship. In any case, the models he is criticizing
at that point include a direct measure of the foreclosure rate as well as the delinquency
rate.

Mortgage Redlining Research: A Review and Critical Analysis Discussion

Uploaded by

Copyright:

Available Formats

Mortgage Redlining Research: A Review and Critical Analysis Discussion

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Mortgage Redlining Research: A Review and Critical Analysis Discussion

Uploaded by

Copyright:

Available Formats

Mortgage Redlining Research: A Review and

Critical Analysis Discussion

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.