PART VI: THE STUDY DESIGN.
- Data sources: HMDA and Census Bureau.
All of the mortgage loan data used in this study, including the mortgage loan data underlying the maps prepared for the report, are based on the computerized home mortgage loan database created by the Federal Reserve Board (FRB) pursuant to the Home Mortgage Disclosure Act of 1975. The Federal Reserve Board makes this database available to the public on an annual basis in the form of a computerized data tape know as the Loan Application Register (LAR). Each record in the LAR represents an individual application for residential mortgage loan or home improvement loan (or the purchase of such a loan). Each record also indicates the census tract location of the property involved, the race and income of the applicant, and the disposition of the loan application.
The Banking Research Project used the LAR tapes for 1990 and 1991 to construct an HMDA database of 1-4 family home purchase loans originated during 1990 and 1991, for each of the 16 metro areas. This analysis included 633,850 home purchase loan application records from the 1990 HMDA database and 623,132 such records from the 1991 HMDA database.
All of the mortgage loan data used in this report refers to home purchase loan originations. (The sole exception is found in Table 5, which contains data on loan purchases rather than loan originations.) Under the HMDA reporting system, home purchase loans include mortgages to purchase 1-4 family structures and condominium units. Home purchase loans do not include refinancing, home improvement loans, loans for multifamily dwellings. The total number of home purchase loan application records in the LAR for each metro area in the study is shown in Table 7. Table 7 also indicates the number of home purchase loan application records for each metro area for which the FRB had indicated validity or quality flags. This study excluded all such application records. Table 8 provides comparable, but more limited data on the total number of home purchase loan application records in the 1990 LAR for each metro area in the study.
All of the HMDA data used in this study refers to the number of home purchase loans, rather than the dollar amount of such loans. Given the fact that home prices are generally higher in white neighborhoods than in minority neighborhoods, use of dollar amount data would show an even greater disparity in lending activity between white and minority neighborhoods than use of loan count data.
The 1991 HMDA database and, to a lesser extent, the 1990 HMDA database, were used to calculate HMDA statistics that identified any major lenders that had originated comparatively few home purchase loans in minority neighborhoods. The Banking Research Project then employed computerized mapping techniques to map the lending patterns of these major mortgage lenders.
This study also employed HMDA statistics and computer mapping to identify any major mortgage lenders that had made a comparatively large number of home purchase loans in minority neighborhoods. Such lending patterns are referred to as "affirmative lending patterns". A number of such lending patterns were identified and 8 lending pattern maps were printed to depict these "affirmative lending pattern" situations. Two of these affirmative lending patterns involved situations in which a mortgage company, identified and mapped as a worst case lender, had a bank affiliate with an affirmative, albeit somewhat limited, mortgage lending pattern.
All of the Census data used in this report are based on the 1980 Census data computer tape which the FRB makes available to the public. This computerized Census database contains population, family income, and housing stock data for each census tract within the United States. At the time the report was prepared, the computer tape for 1990 Census data was not available from the FRB. The study used Census information from the FRB's Census tape to identify the population, income, and housing stock characteristics of the census tract associated with each LAR application record.
- Selection of lenders for mapping.
The basic purpose of our research was to look for large home purchase loan originators that have extensive lending territories within a metro area, but which exclude or substantially underserve minority neighborhoods. As indicated above, such lending patterns provide prima facie evidence of unlawful racial redlining.
The process of identifying worst case lending patterns started with the 1991 HMDA data files, and to a lesser extent its 1990 HMDA data files. These files contain several different lending ratios that measure the extent of a lender's home purchase lending in minority neighborhoods. Selection of lenders was based on a set of lending ratios calculated for each lender in each of the 16 metro areas. These files were used to identify and select for mapping the major home purchase loan originators within each of the 16 metro areas that had made few loans in minority neighborhoods.
The central lending ratio employed in this selection process was the percentage of a lender's 1-4 family home purchase loan originations within a given metro area that were secured by homes located in minority census tracts. The primary definition of "minority" census tracts used in calculating this lending ratio was census tracts in which minorities comprised 75 percent or more of the population (high minority area).
Two additional lending ratios, based on alternative definitions of minority census tract, were taken into consideration in the selection process. These alternative definitions were:
- The percentage of home purchase loans made in census tracts in which minorities comprised 50 percent or more of the population ("minority area"); and
- The percentage of home purchase loan approvals granted in census tracts in which Black persons comprised 75 percent or more of the population ("Black census tract").
As a general rule, the Banking Research Project selected for mapping the large lenders within each metro area with the lowest percentage of their home purchase loan origination in high minority census tracts (defined as census tracts in which minorities comprised 75 percent or more of the population). In some selections, however, the two lending ratios using alternative definitions of minority census tracts were taken into consideration.
The lenders selected were drawn from among the top 20 home purchase loan originators within each of the 16 metro areas. The only cases where selected lenders ranked higher than 20th within a metro area involved two lenders whose market shares declined considerably from 1990 to 1991, and Prudential Home Mortgage Company in the Oakland metro area, which ranked 21st. In 45 of the 62 worst case lending patterns mapped for this report, the lender involved ranked within the top 10 home purchase loan originators within the metro area. Such large-volume mortgage lenders tend to have more extensive effective lending territories than smaller or even mid-sized mortgage lenders, and for this reason their geographic lending patterns are more likely to present Fair Lending issues.
The same HMDA data files were also used to identify lenders that had made a comparatively high number of home purchase loans in minority neighborhoods. Such lenders are referred to in the report as "affirmative" lenders.
- Designation of worst case lending patterns and affirmative lending patterns.
The Banking Research Project used a computer-generated map to analyze the lending pattern of each lender identified as having made comparatively few home purchase loans in minority neighborhoods. Each lending pattern map depicted a lender's home purchase loan activity within a single metro area. Whenever a pattern of exclusion or underservice was shown, the lender's pattern was designated as a "worst case" lending pattern.
The Banking Research Project printed at least one lending pattern map for each of the 62 lending patterns that were designated as worst case patterns. Color-coded maps were produced depicting these patterns. We also printed a lending pattern map for each of the 8 lending patterns that were selected as affirmative patterns.
For a lending pattern to be designated "worst case," four criteria had to be met:
- The lender had to rank among the top 20 home purchase loan originators within the metro area;
- The lender had to rank among the lowest of the major home purchase loan originators within a given metro area, in terms of the percentage of a lender's total home purchase loan originations in the metro area made in minority neighborhoods, as determined by HMDA statistics;
- The lender had to originate home purchase loans throughout most of the metro area or at least large segments of it. Making such a determination requires mapping, at least if it is to de done efficiently; and
- The lender had to exhibit a pattern of excluding or substantially underserving minority areas.
The great majority of the computer-generated lending pattern maps analyzed were designated worst case lending patterns. This was not surprising in view of the fact that each lender selected for mapping (i) was a major lender within the metro area and thus could be expected to have a large effective lending territory and (ii) had made few loans in minority neighborhoods or had underserved such neighborhoods.
The comparatively smaller number of map patterns that were not given the "worst case" designation usually involved lenders with effective lending territories of limited geographic scope, often located in suburban areas. In a limited number of cases, virtually all of the lender's home purchase loans were located in a handful of outlying census tracts -- an indication that the lender was financing particular tract developments, rather than offering home purchase loan to the public at large.
The distinction between lenders and lending patterns is important. A "lending pattern" refers to a lender's activity within a single metro area. Yet, many major lenders, including some of those designated as worst case lenders in this study, are active in more than one metro area. As indicated above, this study identified 62 worst case lending patterns, but only 48 lenders accounted for these 62 lending patterns. Eight of the 48 worst case lenders were responsible for worst case lending patterns in more than one of the 16 metro areas.
Both the lending ratio statistics used to select lenders for mapping and the lending pattern maps reviewed to identify worst case lending patterns refer to a lender's home purchase lending activity within a single metro area. Thus, when a lender was selected for mapping, it was on the basis of the lender's HMDA statistics for a given metro area -- and the selection was independent of the lender's activity in other metro areas. Similarly, when a lender's pattern was designated a worst case lending pattern, this classification was made independent of any lending patterns that the lender might have shown in other metro areas.
When a lender is referred to as a worst case lender, the designation does not refer to the lender's pattern of activity in general, but rather is specific to the lender's activity in one or more metro areas where the lender is responsible for a worst case lending pattern. Thus, it is possible for a lender to be worst case lender in one metro area and an affirmative lender in another metro area. In fact, this study found that Citibank was an affirmative lender in the New York metro area, while Citibank Federal Savings Bank was a worst case lender in the Los Angeles metro area. Hence, Citicorp, the bank holding company which owns both Citibank (New York) and Citibank Federal Savings Bank (Los Angeles), could been seen as both a worst case lender and an affirmative lender.
Nonetheless, Fair Lending law obligates banking organizations to pursue nondiscriminatory mortgage lending policies in all of the markets in which they operate. Thus, the fact that a lender is an affirmative lender in one metro area does not in any way mitigate a worst case lending pattern in another metro area.
- Lending patterns and corresponding HMDA statistics.
- List of worst case lenders and affirmative lenders - Table 9.
The worst case lending patterns and affirmative lending patterns identified in this study are listed in Table 9 by the name of the responsible lender. This table indicates the metro area for each lending pattern and provides summary HMDA statistics on the lender's home purchase loan application and origination activity within that metro area. Summary HMDA statistics for 1991 are shown for each lender. In cases where two-year HMDA data (1990 and 1991) were used to create the lending pattern map for a selected lender, Table 9 also provides summary HMDA statistics for 1990 for the lender. Table 9 also indicates each lender's rank within the relevant metro area, in terms of the number of home purchase loan originations.
The Lender Type Code column in Table 9 indicates which federal agency has supervisory authority over each selected lender for HMDA purposes. To some extent, this supervisory designation provides a useful method for classifying mortgage lenders. The codes are:
Code 1: Commercial banks supervised by the Comptroller of the Currency (national banks) and their mortgage company subsidiaries.
Code 2: Commercial banks supervised by the FRB (state chartered banks) and their mortgage company subsidiaries as well as all mortgage company subsidiaries of bank holding companies.
Code 3: Commercial banks and savings banks supervised by the FDIC (state chartered banks) and their mortgage company subsidiaries.
Code 4: Savings and loans and savings banks supervised by the Office of Thrift Supervision and their mortgage company subsidiaries.
Code 7: Mortgage companies that are not subsidiaries of depository institutions or bank holding companies (independent mortgage companies).
Code C: State chartered depository institutions for which HMDA compliance authority has been delegated to state authorities.
For example, in the Boston metro area in 1991, Shawmut Mortgage Company received 1,417 applications for home purchase loans; originated 932 home purchase loans; and ranked first in the metro area in number of home purchase loan originations. Shawmut Mortgage Company is a nonbank mortgage company subsidiary of a bank holding company, Shawmut National Corporation. Thus its federal supervisor for HMDA reporting purposes is the Federal Reserve Board.
- HMDA data ratios used to select lenders for mapping - Table 10.
Table 10 provides for each worst case or affirmative lending pattern the three HMDA data lending ratios considered in selecting lenders for mapping. These ratios are:
- the percentage of the lender's total home purchase loan originations within the metro area that were made in neighborhoods in which minorities comprised 75 percent or more of the population (high minority areas);
- the percentage of the lender's total home purchase loan originations within the metro area that were made in neighborhoods in which minorities comprised 50 percent or more of the population (minority areas); and
- the percentage of the lender's total home purchase loan approvals within the metro area that were made in neighborhoods in which Black persons comprised 75 percent or more of the population (Black areas).
Table 10 also provides these same HMDA data ratios for all home purchase loan originations or approvals made in 1991 by all lenders reporting under HMDA for each of the metro areas. These ratios are shown in the row of data to the right of each metro area name. These ratios indicate the geographic distribution of the entire home purchase loan origination or approval market within the metro area and provide a useful benchmark for evaluating the lending ratios of individual lenders.
A fourth lending ratio reviewed by the Banking Research Project -- the percentage of loan approvals in Black neighborhoods that were made to Black applicants -- is also shown in Table 10. It is quite different from the other ratios in that it looks not only at the geographic location of loans, but also at the race of the borrower. Where only a small percentage of the home purchase loans originated by a lender in Black neighborhoods are extended to Black applicants, this may suggest a pattern of discrimination against Black applicants and displacement of Black residents. The value of this "displacement ratio" for several of the worst case lenders raises such a concern.
As shown in Table 10, in the Philadelphia metro area in 1990, PHH U.S. Mortgage Corporation made 3.3 percent of its total home purchase loan application approvals in Black neighborhoods -- a percentage ratio well above that of the other worst case lenders in Philadelphia. Yet, only 5.6 percent of the mortgage company's loan approvals in Black neighborhoods were granted to Black applicants. By way of contrast, 70 percent of all approvals granted by all lenders in the Black neighborhoods of the Philadelphia metro area in 1991 were made to Black persons. Table 10 shows roughly similar patterns for Manhattan Savings Bank in the New York metro area (1990 and 1991), Source One Mortgage Services in the St. Louis metro area (1991), and American Residential Mortgage in the Dallas metro area (1991).
- Lending patterns and HMDA market share data - Table 11.
The summary market share data for the individual lenders whose lending patterns have been mapped for this report are shown in Table 11. The 4th and 5th data columns show each lender's share of total home purchase loan originations in minority neighborhoods and in white neighborhoods. Where two-year HMDA data was used to create a lending pattern map, the market share ratios for the lender are shown separately for 1990 and 1991.
The first market share ratio (the fourth column of Table 11) indicates the individual lender's percentage share of all home purchase loan origination made by all HMDA reporters in all the minority census tracts of the metro area. High minority census tracts are defined as those in which minorities comprise 75 percent or more of the census tract population.
The second market share ratio (the fifth data column of Table 11) indicates the individual lender's percentage share of all home purchase loan originations made by all HMDA reporters in all the white census tracts of the metro area. White census tracts are defined as those in which white persons comprise 75 percent or more of the census tract population.
In contrast to the lending pattern maps, which indicate market share at the census tract level, the two market share ratios in Table 11 refer to market share within areas that represent aggregations of census tracts with similar racial concentration levels.
The market share ratios presented in Table 11 are the single best HMDA statistic for evaluating a major lender's level of activity in minority neighborhoods. This is due to the fact that the market share approach automatically adjusts for the smaller size of the loan market in minority neighborhoods. By way of contrast, the lending ratios presented in Table 10 (loan distribution ratios) -- which focus on the percentage of the lender's total loans made in minority neighborhoods -- do not adjust for the difference in market volume and thus tend to understate the impact that the geographic distribution of a lender's loans has on minority neighborhoods.
For example, as shown in Table 10, of the total number of home purchase loans originated by Shawmut Mortgage Corporation during 1991 in the Boston metro area, 5.5 percent were made in high minority neighborhoods -- a seemingly modest share of Shawmut's total origination. Yet, as shown in Table 11, Shawmut's home purchase loans in these minority neighborhoods accounted for 27.9 percent of all the home purchase loan originations that they received.
However, neither the market share ratios presented in Table 11 nor the loan distribution ratios presented in Table 10 indicate the spatial location of the lender's loans. Thus, they cannot provide information on the shape of the lender's effective lending territory or the geographic distribution of the lender's loans within the minority community.
Table 11 also identifies the lending pattern map associated with each set of lender market share ratios. The lending pattern maps incorporated into the report are assigned the map numbers shown in Appendix A. The other lending pattern maps are given the map numbers listed in Appendix C. Table 11 also indicates whether the map depicts a worst case lending pattern (WCL) or an affirmative lending pattern (AFR).
- Inventory of lending pattern maps and Census data maps.
- Lending pattern maps.
The Banking Research Project generated 80 color-coded lending pattern maps. These maps depict 71 different lending patterns: 62 worst case lending patterns, 8 affirmative lending patterns, and one pattern not classified. In 8 cases, two maps were prepared showing the same lending pattern from different perspectives. Finally, there is one map showing the local community delineated by a lender for purposes of the Community Reinvestment Act.
Of the 80 lending pattern maps produced, 30 are included and discussed in the text of the report. The entire set of 80 lending pattern maps produced for this study is listed in Appendix C, and is available from the Banking Research Project. The 30 maps actually included in this report are listed in Appendix A.
Among the 70 separate lending pattern maps generated for this study, 53 are based on 1-4 family home purchase loans originated during 1991 and reported pursuant to HMDA. The 17 other lending pattern maps are based on the lender's loan activity over the two-year period including both 1990 and 1991. Two-year HMDA data was available for mapping in these instances because the Banking Research Project had previously created a computerized mapping file for these lenders using the 1990 HMDA data and it was relatively easy to consolidate their 1990 and 1991 mapping files. In cases where two-year HMDA data was employed for mapping, the market share percentage shown in the maps refers to the lender's share of total home purchase loans originated by all HMDA reporters over the two-year period.
As a general matter, mapping two-year HMDA data provides a clearer picture of a lender's effective lending area. This is because even where lenders are actively engaged in serving a particular community, they may not lend to every census tract within the community each year, or their level of lending within individual census tracts may fluctuate greatly from one year to the next. The use of two-year data smooths out such fluctuations and produces more clearly defined effective lending territories. On the other hand, using two-year HMDA data may not be advisable if there has been a substantial change in the lender's mortgage lending pattern from one year to the next. None of the lenders for which two-year HMDA data was used in mapping worst case lending patterns had shown any major shift in the percentage of their total home purchase loans going to minority neighborhoods between 1990 and 1991.
- Census data maps.
The Banking Research Project also employed the FRB's computerized database of 1980 Census data to create a set of demographic maps, or Census data maps, for each of the 16 metro areas. The various Census data maps prepared for this study show for each census tract:
- the number of 1-4 family housing units (all 16 metro maps);
- the percentage of the census tract population represented by minorities (all 16 metro maps);
- whether the population of minority census tracts is predominantly Black or predominantly Hispanic (7 metro maps); and
- the median family income within the census tract (12 metro maps).
Used in conjunction with the lending pattern maps, these Census data maps provide a valuable framework for analyzing whether lenders have improperly excluded minority neighborhoods from their effective lending territories.
The Banking Research Project created a total of 51 color-coded Census data maps. All are listed in Appendix C, and are available from the Banking Research Project upon request. Six of the 51 Census data maps are included and discussed as part of the text of this report. A listing of all the maps included with this report can be found is Appendix A.
- Design of lending pattern maps.
- Market share concept.
To depict each lender's role in the mortgage market at the neighborhood level and the lender's overall marketing strategy, this report relies upon the market share concept in creating the lending patterns maps. A lender's market share within a census tract means the lender's share, expressed as a percentage, of the total number of 1-4 family home purchase loan originations made by all lenders within that census tract during a given year.
An alternative method for mapping the lending activity of individual lenders -- one rejected by this study -- is the loan count approach, which would display the number of loans made by each lender within each census tract. A loan count mapping is made by stratifying the census tract loan counts for an individual lender into intervals, and then using color codes to map the interval associated with the loan count for each census tract.
The market share concept is a conservative approach that tends to place lenders in a more favorable light than the loan count approach. Because lending activity in white neighborhoods tends on average to be significantly higher than in minority neighborhoods, maps using loan counts will show much higher activity in white neighborhoods than in minority neighborhoods for the great majority of lenders. Thus, the loan count approach emphasizes the disparity in lending levels between minority and white neighborhoods.
However, the loan count approach makes it is difficult to distinguish between lenders that affirmatively serve minority neighborhoods and those that underserve such neighborhoods. This happens because even lenders that have made an affirmative effort to lend in minority neighborhoods will often show higher loan counts in white neighborhoods than in minority neighborhoods.
By contrast, the market share approach automatically adjusts a lender's loan count for each census tract to reflect the size of the mortgage loan market within that census tract. This dampens the disparity in lending activity between minority and white neighborhoods, and by doing so places lenders in a more favorable light. But it also makes it easier to distinguish between affirmative lenders, break-even lenders and negative lenders.
For purposes of mapping most of the worst case lending patterns identified in this report, it would not have made much difference whether the market share approach or the loan count approach were used. Either approach would have shown that most of the worst case lenders have made no loans in the great majority of minority census tracts.
- Color-coded market share intervals.
For all of the lending pattern maps (with the exception of those including the New York City metro area) the market share values at the census tract level have been stratified into four intervals:
1) 0 percent market share (no loans within the census tract);
2) 0.01 percent - 3 percent market share;
3) 3 percent - 10 percent market share; and
4) over 10 percent market share.
The four market share intervals have been color-coded, and each census tract has been given the map color that corresponds to the lender's market share within that tract. In the case of census tracts where no home purchase loans were reported by any lenders subject to HMDA, the census tract area has been coded yellow.
Since the overall level of concentration in the home purchase loan origination market varies appreciably from one metro area to the next, one would also expect to find considerable variation among metro areas in the market shares held by larger lenders. Nonetheless, in most of the metro areas under review, a market share falling within the 0.01 percent - 3 percent interval can be viewed as "moderate"; a market share within the 3 percent - 10 percent interval can be termed "average"; and a market share greater than 10 percent can been seen as "major."
The lending pattern maps for the New York City metro area differ in that they employ five market share intervals:
0.01 percent - 3 percent;
3 percent - 10 percent;
10 percent - 25 percent; and
over 25 percent.
The fifth interval (over 25 percent) was added because some of the major lenders had very high market shares in many neighborhoods. There is a reason for the high level of concentration in the mortgage market at the neighborhood level in the New York City metro area. In many neighborhoods 1-4 family structures and condo units represent a very small share of the housing stock and consequently the market for home purchase loans tends to be relatively thin.
- Demographic data overlays
Minority neighborhoods are identified in the lending pattern maps by means of diagonal line overlays. In creating these overlays, the definition of "minority neighborhoods" assumes a key role, just as it did in the calculation of the HMDA data lending ratios. For overlay purposes, the Banking Research Project selected the definition of minority neighborhoods that would enable a lending pattern map to best portray the level of minority population concentration at which the lender tended to exclude or substantially underserve minority neighborhoods.
For the great majority of lending pattern maps, minority neighborhoods were defined as census tracts in which minorities comprised 50 percent or more of the population. In some metro areas where an alternative definition of minority neighborhoods proved useful in assessing mortgage lending patterns, this alternative definition was employed in creating the diagonal overlays.
For example, in Atlanta and Detroit the diagonal overlay indicates census tracts in which minorities represent 25 percent or more of the population. In Miami the overlay shows census tracts in which Black persons represent 50 percent or more of the population. In Los Angeles, St. Louis, and Houston, the diagonal overlay on some of the lending pattern maps indicates census tracts in which minorities represent 75 percent or more of the population. As explained below, a broader perspective on the correspondence between a particular lending pattern and the location of minority neighborhoods can be gained by a side-by-side comparison of the lending pattern map and the minority population map for the metro area.
To gain further insight into the lending patterns reviewed in this study, several lending pattern maps were created in which the diagonal overlay indicates a specific median family income level, rather than minority population concentration level. For example, Map 2, the lending pattern map for Chase Home Mortgage Corporation in Chicago uses a diagonal overlay to indicate census tracts in which the median family income is greater than 110 percent of the metro area median family income -- i.e., upper income and upper-middle income neighborhoods. Other lending pattern maps with diagonal overlays specifying a median family income level are NBD Mortgage in Chicago, Bell Federal in Pittsburgh and Sears in Pittsburgh.
Even where median family income overlays have not been created, a clear perspective on the relationship between a lender's mortgage lending pattern and neighborhood's income levels can be gained by means of a side-by-side comparison of the lending pattern map and the median family income map for the metro area.
- Technical data processing and mapping issues.
The Banking Research Project has prepared a technical paper that explains in detail how the HMDA data statistics, lending pattern maps, and Census data maps were prepared for this study. This paper discusses how the FRB LAR data file was processed, which application records were excluded, and the structure of the HMDA data files created by the Banking Research Project. The paper also discusses the design of the lending pattern maps and the Census data maps and the source of the computerized census tract boundaries used to construct these maps. In addition, the paper explains the approximation procedure employed by the Banking Research Project in order to use HMDA data based on 1980 census tract boundaries to create lending pattern maps and Census data maps based on 1990 census tract boundaries.
The lending pattern maps and Census data maps prepared for Atlanta, Boston, St. Louis, and Washington DC use 1980 census tract boundaries. The maps prepared for Baltimore, Buffalo, Chicago, Dallas, Detroit, Houston, Miami, New York, Oakland, Philadelphia, and Pittsburgh use 1990 census tract boundaries. Two sets of maps -- one using 1980 census tract boundaries and the other using 1990 census tract boundaries -- were prepared for Los Angeles in order to evaluate the impact of switching from 1980 to 1990 boundaries. Appendix C indicates the year of the census tract boundaries used for each lending pattern map and census data map prepared for the report.