Residential Lending in New Haven County
Sun, Aug 1, 1993
Evidence From 1990 and 1991 Home Mortgage Disclosure Act (HMDA) Data
Prepared for: CURE
by: Morgan N. Sandquist and Harvey L. Koizim
August, 1993
©1993 Connecticut Urban Reinvestment Endowment, Inc.
P.O. Box 8905, New Haven. CT 065332-0905. All rights reserved.
Acknowledgements:
We are grateful for the assistance and wise counsel of Professor Douglas W. Rae of the Yale School of Organization and Management (SOM).
Without the technical assistance of several people in the Yale Community, this report would not have been possible. SOM Professors Edward H. Kaplan and Todd Strauss helped solve statistical problems. William L. Cunningham consulted on data management and controlled statistical oversight. Mimi Liu researched and consulted on census statistics, maps and graphs. Law students Dan Ehrenberg and Watt Alexander helped with planning and research.
Yale University paid a portion of the cost of acquiring the HMDA data and made its imt computer and research facilities available.
Connecticut Urban Reinvestment Endowment, Inc. (CURE), published, underwrote and sponsored the project.
This paper was prepared for CURE as an undertaking of the Housing and Community Development Clinic of the Yale Law School Jerome N. Frank Legal Services Organization.
Morgan N. Sandquist received a Masters in Public and Private Management from SOM in 1993.
Harvey L. Koizim, J.D. Yale 1951, was a banker from 1963 to 1985. Currently he is President of CURE, and a Tutor in Clinical Studies at Yale Law School.
Discrimination in Lending and Legislative Remedies
Because each loan approval involves several subjective determinations, underwriting biases can effect residential lending decisions. In a landmark study conducted for tile Federal Reserve Bank of Boston, the authors reported:
Many observers believe that no rational lender would turn down a perfectly good application simply because the applicant is a member of a minority group. The results of this survey confirm this perception, minorities with unblemished credentials are almost (97 percent) certain of being approved. But the majority of borrowers–both white and minority–are not perfect, and lenders have considerable discretion over the extent to which they consider these imperfections as well as compensating factors.
…The results of this study suggest that for the same imperfections whites seem to enjoy a general presumption of credit worthiness, that black and Hispanic applicants do not, and that lenders seem to be more willing to overlook flaws for white applicants than for minority applicants.
…As the supervisory agencies themselves have already recognized, under existing examination procedures, examiners can be expected to uncover only the most flagrant abuses.[1]
The results of the Fed analysis show that mortgage underwriters continue to discriminate against applicants because of personal characteristics such as gender, race, ethnicity, or income. In addition, there is broad evidence of patterns of bias against communities based on demographic characteristics like racial composition, urban location, or income profile. These practices, whether consciously or unconsciously implemented, result in disinvestment and lead to decline and decay of minority and inner-city neighborhoods.[2]
All of these forms of discrimination have been proscribed by a variety of federal and state legislation. Two laws are relent to this study: The Home Mortgage Disclosure Act of 1975 (HMDA)[3]; and the Community Reinvestment Act of 1977 (CRA)[4]. First enacted almost two decades ago, HMDA mandated the disclosure of the census tract or zip code locations of properties for which mortgage loan applications were received.[5] In 1989, having been presented with widespread evidence of continued pervasive discrimination in lending. Congress strengthened HMDA.[6] Starting with the data disclosed for 1990, a much wider variety of information was to be revealed including: loan value; race, gender and income of applicant; purpose of loan; form of government insurance; owner-occupancy of property; and, if applicable, reason for denial.[7]
Under CRA, banks are evaluated on one to three year cycles when they undergo general regulatory examinations. Bank examiners rate them with respect to their community reinvestment performance.[8] Interested groups can challenge proposed expansions, mergers, or acquisitions by regulated institutions because of alleged poor community reinvestment performance. Necessarily, examiners’ ratings also involve subjective decisions. For example, this study shows that People’s Bank (see Tables 5 and 8 as well as Appendix 4B), which recently received the highest (“Outstanding” ) rating from the examiners[10] manifests a considerable bias against minority applicants.
After the passage of the CRA in 1977, several organizations challenged bank expansions. They presented statistical analyses contesting the community reinvestment performance of banks throughout the country. Such analyses typically relied on information supplied under the old HMDA in attempts to demonstrate that certain geographic areas were receiving far fewer loans based on the racial composition, urban location, or income profile of the neighborhoods. However, these analyses on their own were unable to determine whether or not discrimination was in fact occurring because of the omission of information about mortgage applicants from the HMDA data. There could have been other explanations for the disparity in credit extension. It was possible (given the probable though incomplete correlation between residence in a low-income or minority area and reduced credit worthiness) that applicants from the contested areas were less credit worthy than applicants from other areas.[11] Furthermore, those pre-1990 analyses could focus only on geographical discrimination. No information about applicants was available so examiners could not judge applicant discrimination.
As a result, when presented with statistical evidence of discriminatory lending practices, regulators have been skeptical and have given such data little weight in their CRA evaluations. Organizations challenging CRA performance have been forced to rely on other evidence of discriminatory practices, such as outreach and marketing efforts. Therefore, evaluations of CRA performances have become focused on procedure rather than substance.[12]
Starting in l990, changes in the law[13] allowed organizations examining the results of banks’ community reinvestment efforts to control for those applicant characteristics that are now disclosed. However, the published information still does not allow control for most key determinants of an applicant’s credit worthiness, including credit history, amount of other debt and the loan-to-value ratio. Therefore, discrimination still cannot be demonstrated conclusively through solely quantitative analysis of publicly available information.[14]
The Federal Reserve Bank of Boston study[15] overcame these problems. Using the Fed’s position as a bank regulator, the writers were able to gain cooperation of lenders. They got access to all loan applications filed in the Greater Boston area in 1990. After evaluating this information, the study was able to conclude that minority applicants are indeed generally less credit worthy. However, the differences in credit worthiness did not fully explain discrepancies between the loan rejection rates for white and minority applicants. To the extent that the characteristics of credit worthiness for minority mortgage applicants in the Greater Boston Area are applicable to minority mortgage applicants elsewhere, inference can be drawn about how wide a discrepancy in loan rejection rates between white applicants and minority applicants is consistent with non-discriminatory behavior.
Discrimination in Lending in New Haven County
The Connecticut Urban Reinvestment Endowment (CURE), in conjunction with the Housing and Community Development Clinic at the Yale Law School, the Yale School of Organization and Management, and others, acquired the HMDA and related census data pertaining to the years of 1990 and 1991. We have analyzed this information to determine if the data documents any discrimination in lending in New Haven County.
In an earlier study,[16] students from the Yale Law School and the Yale School of Organization and Management examined the patterns of lending in New Haven in 1987 using HMDA data then available. They found that mortgage applications from census tracts with lower incomes or greater percentages of minority residents were more likely to be rejected. Their results were based solely on the then available census tract information. The bankers attacked their inferences as indeterminate because income, loan amounts and other factors were unavailable, therefore not considered.
We have controlled for those applicant characteristics for which statistics are now available. We have used the data from 1990 to construct a model of the lending process and identify those factors which appear to affect the probability of an application being rejected. Because 1990 was the first year for which this expanded information was disclosed, it seems likely that the data is not completely reliable. Furthermore, only 2,427 usable applications were received in 1990 as compared to 15,254 in 1991. Thus, the data from 1991 was used to confirm the model we developed and to draw our conclusions about discrimination in lending.
HMDA data for 1990 and 1991 are summarized in Appendices 1 and 2. Mortgage applications were included in the database if they were received from white, black, or Hispanic applicants and contained all of the information considered in this study. Loans which respondents purchased were excluded, focusing only on loans for which these institutions made the origination decision.
In general, minorities are under-represented as mortgage applicants in both 1990 and 1991. Though minorities make up 17% of the population of New Haven county, only 13.7% of the applicants in 1990 and 6.9% of the applicants in 1991 were minorities. 76.9% of all applications received in 1990 and 76.2% of all applications received in l991 resulted in loan originations. These rates were lower for minorities in both years. In 1990, 72.2% of the applications received from black applicants and 65.8% of the applications received from Hispanic applicants resulted in loan originations. In 1991, 57.4% of the applications received from black applicants and 60.2% of the applications received from Hispanic applicants resulted in loan originations.
Applications received from low-income census tracts (defined as census tracts with median incomes below 80% of the median income in New Haven County) and minority census tracts (defined as census tracts with a combined percentage of black and Hispanic residents above 40%) were also less likely to result in loan originations. In 1990, 71.0% of the applications received from minority census tracts (non-minority tracts achieved 77.4%) and 74.9% of the applications received from low-income census tracts resulted in loan originations (non-low income tracts were 77.2% ). In 1991 with a greater volume of applications the aggregate numbers were even less favorable to minority and low income census tracts. 58.1% of the applications received from minority census tracts and 65.0% of the applications received from low-income census tracts resulted in loan originations (as opposed to the fairly constant figures of 77.2% for non-minority tracts and 77.4% for non-low-income tracts).
This is graphically represented for 1991 in Figures 4 and 5 which are maps showing that applications from minority census tracts tend to have a lower probability of resulting in origination. These maps also show that applications from low income census tracts are less likely to result in origination, but this pattern is noticeably weaker than that for minority census tracts. Finally, the maps demonstrate that applications from urban areas result in origination less frequently. It should be noted that there were few if any HMDA reported applications received from many of the large unshaded census tracts.
Logit Analysis
Though these statistics appear damning, there could be explanations other than discrimination. Thus, all factors that are relevant to the evaluation of a loan application must be controlled before any conclusions can be drawn.
One method of controlling for these factors is logit analysis. Logit equations are specified in a manner similar to a simple linear regression. However, rather than suggesting how much a given factor increases or decreases some dependent value, logits suggest how much a given factor increases or decreases the probability of a given outcome. As such, logit analysis is useful in those cases where the dependent factor is dichotomous, such as “loan originated” or “loan not originated.”
We have used a logit model with the 1990 data to control for those factors which might affect the evaluation of a loan application, with the exception of those factors for which data is not available. The characteristics that have been included in the logit model are described in Appendix 3. The effects that each of these factors had on the probability that a loan would not be originated are summarized in Table 1.
We used the same logit model with the 199l data to verify the patterns we discovered in the 1990 data The results of that analysis are summarized in Table 2.
TABLE 1: Determinants of Probability of Mortgage Loan Being Oiginated
New Haven County - 1990
| Factor | Coefficient (Std Error)[a] | Impact of Factor on Prob of Origination[b] | Mean |
|---|---|---|---|
| Government Insurance | 0.9593 (.1407) | 17.0% | .2843 |
| Home Improvement Loan | 0.6939 (.1848) | 12.3% | .1603 |
| Mortgage Refinancing | -0.1644 (.1218) | -2.9% | .2311 |
| Owner-Occupied Property | 0.4270 (.2445) | 7.6% | .9681 |
| Income-to-Loan Ratio | 0.0107 (.0237) | 0.2% | 1.3887 |
| Applicant Income | 0.0096 (.1080) | 0.2% | .5876 |
| Female | -0.1390 (.1304) | -2.5% | .1585 |
| Minority | -0.5203 (.1490) | -9.2% | .1368 |
| Median Income of Cen Tract | 1.3481 (.6150) | 23.9% | .4188 |
| Minority % of Census Tract | -0.1213 (.3683) | -2.2% | .1136 |
[a]Bolded coefficients are statistically significant at the 95% significance level.
[b]This impact is interpreted as the amount by which a change of one in a given factor will change the probability of a loan being originated for the average application. Thus, if the average application has a 76.9% chance of resulting in an origination, then an increase in applicant income of $100,000 (1 unit of applicant income) for the average applicant will increase that probability by 0.2% to a 77.1% chance of resulting in an origination.
TABLE 2: Determinants of Probability of Mortgage Loan Being Originated
New Haven County - 1991
| Factor | Coefficient (Std Error)[a] | Impact of Factor on Prob of Origination[b] | Mean |
|---|---|---|---|
| Government Insurance | 0.0305 (.0697) | 0.6% | .1019 |
| Home Improvement Loan | -0.6266 (.0629) | -11.4% | .1602 |
| Mortgage Refinancing | -0.2758 (.0453) | -5.0% | .3838 |
| Owner-Occupied Property | -0.1964 (.0824) | -3.6% | .9363 |
| Income-to-Loan Ratio | 0.0232 (.0084) | 4.2% | 1.2844 |
| Applicant Income | -0.0598 (.0239) | -1.1% | .6632 |
| Female | -0.0255 (.0494) | -0.5% | .1849 |
| Minority | -0.6293 (.0755) | -11.4% | .0690 |
| Median Income of Census Tract | 1.0040 (.2174) | 18.2% | .4441 |
| Minority % of Census Tract | -0.7471 (.1523) | -13.6% | .0903 |
[a]Bolded coefficients are statistically significant at the 95% significance level.
[b]This impact is interpreted as the amount by which a change of one in a given factor will change the probability of a loan being originated for the average application. Thus, if the average application has a 76.2% chance of resulting in an origination, then an increase in applicant income of $100,000 (1 unit of applicant income) for the average applicant will decrease that probability by 1.1% to a 75.1% chance of resulting in an origination.
According to this logit analysis, in 1990, four factors had a statistically significant effect on the probability that a loan application would result in an origination: government insurance; home improvement; minority status of applicant; and median income of the census tract. The impact of the variable on the probability of denial is the amount by which that variable will increase or decrease the probability that the application will result in the origination of a loan for the average application. For example, this analysis implies that being a minority reduced an otherwise average applicant’s chance of getting a loan by 7.9%, from 76%.9 to 69.0%. A similar calculation can be done for the median income of the census tract. Thus, these results are consistent with discrimination on the basis of applicant race and the income of the geographic area.
Stronger results were obtained using the 1991 data to verify this model. This would be expected due to the much larger sample size. According to this logit analysis, eight factors had a statistically significant effect on the probability that a loan application would result in an origination: home improvement; refinancing; owner occupancy; income-to-loan ratio; applicant income; minority status of applicant; median income of census tract; and percent minority of census tract. These results are consistent with discrimination on the basis of applicant race and the income and racial composition of the geographic area.
Performance By Bank
The results cited above are for all of New Haven County for the years of 1990 and 1991. There are reasons to believe that loan evaluation processes vary from lender to lender so aggregate results such as these may conceal both more or less significant results for individual banks. Therefore, this logit analysis should properly be done separately for each individual lender that receives applications from New Haven County. Because there were so few applications submitted in 1990, no lender received more than 700 loans from New Haven County, with most banks receiving fewer than 200 applications. With such a small sample size, particularly given that only 13.68% of the sample is made up of minority applicants, it would be very difficult to get usable results from a logit analysis. Furthermore, because 1990 was the first year for which this expanded data was available, the reliability of this data is questionable. Thus, we applied the logit model developed above, separately, to each of the nine lenders that received more than 400 mortgage applications from New Haven County in 1991, a year in which many more applications were received.
We were able to generate a fit of the logit model for the five institutions which received more than 700 applications in l991. Those were: First Federal Bank; New Haven Savings Bank; People’s Bank; Shawmut Mortgage Company, and The McCue Mortgage Company. Results for each of these companies are summarized in Tables 3 through 7.
TABLE 3: Determinants of Probability of Mortgage Loan Being Originated
First Federal Bank - 1991
(821 Applications - 57.6% Originated)
| Factor | Coefficient (Std Error)[a] | Impact of Factor on Prob of Origination[b] | Mean |
|---|---|---|---|
| Government Insurance | 1.6557 (1.0761) | 40.4% | .0110 |
| Home Improvement Loan | -0.9528 (.2592) | -23.3% | .3191 |
| Mortgage Refinancing | 0.0935 (.2086) | 2.3% | .5043 |
| Owner-Occupied Property | 0.2103 (.5998) | 5.1% | .9854 |
| Income-to-Loan Ratio | 0.0679 (.0230) | 1.7% | 2.7770 |
| Applicant Income | -0.1291 (.0512) | -3.2% | .6827 |
| Female | -0.1246 (.1805) | -3.0% | .2192 |
| Minority | -0.5508 (.2762) | -13.5% | .0962 |
| Median Income of Census Tract | 0.6880 (.8905) | 16.8% | .4171 |
| Minority % of Census Tract | -1.9141 (.6556) | -46.7% | .1149 |
[a]Bolded coefficients are statistically significant at the 95% significance level.
[b]This impact is interpreted as the amount by which a change of one in a given factor will change the probability of a loan being originated for the average application. Thus, if the average application has a 57.6% chance of resulting in an origination, then an increase in applicant income of $100,000 (1 unit of applicant income) for the average applicant will decrease that probability by 3.2% to a 54.4% chance of resulting in an origination.
TABLE 4: Determinants of Probability of Mortgage Loan Being Originated
New Haven Savings Bank - 1991
(1,748 Applications - 81.8% Originated)
| Factor | Coefficient (Std Error)[a] | Impact of Factor on Prob of Origination[b] | Mean |
|---|---|---|---|
| Government Insurance | 0.0000 (.0000) | 0.0% | 0 |
| Home Improvement Loan | 0.0997 (.2069) | 1.5% | .1608 |
| Mortgage Refinancing | 0.4249 (.1437) | 6.3% | .3936 |
| Owner-Occupied Property | -0.0393 (.2184) | -0.6% | .8839 |
| Income-to-Loan Ratio | -0.0081 (.0167) | 0.1% | 1.8386 |
| Applicant Income | 0.1587 (.1094) | 2.4% | .7918 |
| Female | 0.0479 (.1554) | 0.7% | .2145 |
| Minority | -0.7036 (.2190) | -10.5% | .0973 |
| Median Income of Census Tract | 1.2591 (.6715) | 18.7% | .4369 |
| Minority % of Census Tract | -0.3098 (.3970) | -4.6% | .1189 |
[a]Bolded coefficients are statistically significant at the 95% significance level.
[b]This impact is interpreted as the amount by which a change of one in a given factor will change the probability of a loan being originated for the average application. Thus, if the average application has a 81.8% chance of resulting in an origination, then an increase in applicant income of $100,000 (1 unit of applicant income) for the average applicant will increase that probability by 2.4% to a 84.2% chance of resulting in an origination.
TABLE 5: Determinants of Probability of Mortgage Loan Being Originated
People’s Bank - 1991
(l,004 Applications - 74.8% Originated)
| Factor | Coefficient (Std Error)[a] | Impact of Factor on Prob of Origination[b] | Mean |
|---|---|---|---|
| Government Insurance | 0.2801 (.4562) | 5.3% | .0388 |
| Home Improvement Loan | -2.0281 (.2713) | -38.2% | .1464 |
| Mortgage Refinancing | -0.2751 (.1926) | -5.2% | .5169 |
| Owner-Occupied Property | 0.9750 (.4302) | 18.4% | .9741 |
| Income-to-Loan Ratio | 0.0732 (.0603) | 1.4% | 1.0068 |
| Applicant Income | -0.3102 (.1584) | -5.8% | .6335 |
| Female | -0.2570 (.1908) | -4.8% | .2002 |
| Minority | -0.8832 (.3060) | -16.6% | .0707 |
| Median Income of Census Tract | 1.3711 (.9290) | 25.8% | .4522 |
| Minority % of Census Tract | -0.9114 (.6092) | -17.2% | .0839 |
[a]Bolded coefficients are statistically significant at the 95% significance level.
[b]This impact is interpreted as the amount by which a change of one in a given factor will change the probability of a loan being originated for the average application. Thus, if the average application has a 74.8% chance of resulting in an origination, then an increase in applicant income of $100,000 (1 unit of applicant income) for the average applicant will decrease that probability by 5.8% to a 69.0% chance of resulting in an origination.
TABLE 6: Determinants of Probability of Mortgage Loan Being Originated
Shawmut Mortgage Company - l99l
(1,195 Applications - 76.7% Originated)
| Factor | Coefficient (Std Error)[a] | Impact of Factor on Prob of Origination[b] | Mean |
|---|---|---|---|
| Government Insurance | -0.4564 (.2279) | -8.2% | .1347 |
| Home Improvement Loan | 0.0000 (.0000) | 0.0% | 0 |
| Mortgage Refinancing | -0.6996 (.1597) | -12.5% | .4921 |
| Owner-Occupied Property | 0.0913 (.4196) | 1.6% | .9715 |
| Income-to-Loan Ratio | 0.2376 (.1594) | 4.3% | .6811 |
| Applicant Income | -0.2259 (.1225) | -4.0% | .7228 |
| Female | 0.0362 (.1743) | 0.6% | .2100 |
| Minority | -0.5133 (.2743) | -9.2% | .0703 |
| Median Income of Census Tract | 0.3291 (.7785) | 5.9% | .4588 |
| Minority % of Census Tract | -1.3092 (.6309) | -23.4% | .0847 |
[a]Bolded coefficients are statistically significant at the 95% significance level.
[b]This impact is interpreted as the amount by which a change of one in a given factor will change the probability of a loan being originated for the average application. Thus, if the average application has a 76.7% chance of resulting in an origination, then an increase in applicant income of $100,000 (1 unit of applicant income) for the average applicant will decrease that probability by 4.0% to a 72.7% chance of resulting in an origination.
TABLE 7: Determinants of Probability of Mortgage Loan Being Originated
McCue Mortgage Company - 1991
(744 Applications - 86.0% Originated)
| Factor | Coefficient (Std Error)[a] | Impact of Factor on Prob of Origination[b] | Mean |
|---|---|---|---|
| Government Insurance | 0.7373 (.2795) | 8.9% | .7782 |
| Home Improvement Loan | 0.0000 (.0000) | 0.0% | 0 |
| Mortgage Refinancing | -0.1916 (.4116) | -2.3% | .0793 |
| Owner-Occupied Property | 0.0000 (.0000) | 0.0% | 0 |
| Income-to-Loan Ratio | 0.5474 (1.0168) | 6.6% | .4285 |
| Applicant Income | 0.0153 (.9250) | 0.2% | .4425 |
| Female | 0.0499 (.2994) | 0.6% | .1761 |
| Minority | -0.9394 (.3314) | -11.3% | .1075 |
| Median Income of Census Tract | -0.2601 (1.4801) | -3.1% | .4014 |
| Minority % of Census Tract | -1.0105 (.8976) | -12.1% | .1068 |
[a]Bolded coefficients are statistically significant at the 95% significance level.
[b]This impact is interpreted as the amount by which a change of one in a given factor will change the probability of a loan being originated for the average application. Thus, if the average application has a 86.0% chance of resulting in an origination, then an increase in applicant income of $100,000 (1 unit of applicant income) for the average applicant will increase that probability by 0.2% to a 86.2% chance of resulting in an origination.
Inferences Drawn from Bank Tables
For all five of the lenders studied above, applications from minority applicants who are average in all other respects have a lower probability of resulting in the origination of a loan. In four of the five cases, this effect is statistically significant at the 955 significance level, and in the fifth case it is statistically significant at the 90% significance level. For all five banks, an increasing percentage of minorities in the area decreases the probability that an otherwise average application will result in the origination of a loan, with the effect being statistically significant at the 95% significance level in two of those cases. Finally, for four of the five banks, a lower median income in the area decreases the probability that an otherwise average application will result in the origination of a loan, though in no case is the effect statistically significant.
These results generally mirror outcomes for New Haven County as whole for 1991. However, they tend to be less statistically significant. This is probably largely attributable to the smaller sample sizes available for individual banks. For the five lenders examined, there were between 744 and 1,748 applications received, as compared to 15,254 applications received from New Haven County as a whole. As sample sizes decrease the standard error increases, making a statistically significant result less likely, even if the parameter is the same. But the results do not vary with respect to minority status and minority percentage of census tract. It is the consistency of these results that is most telling.
These results are demonstrated graphically for the whole county for 1990 and 1991 in Figures 6 and 7. Table 8 shows how the probability of an application resulting in an origination is changed for the average application if it takes on that characteristic (if the value of that variable changes from the average value to one). For non-discrete variables, an amount of change was selected as stated in the notes to the table. Figures 6 and 7 depict the results for gender, minority status, median income of census tract, and minority percentage of census tract.
TABLE 8: Effects of Factors on Probabilities of Mortgages Being Originated
Countywide 1990-1991
Individual Lenders Originating More Than 700 Mortgages 1991
| Factors | County ‘90 | County ‘91 | First Federal ‘91 | NH Savings ‘91 | People’s Bank ‘91 | Shawmut Mtge ‘91 | McCue Mtge ‘91 |
|---|---|---|---|---|---|---|---|
| Gov’t Insurance | 12.17% | 0.54% | 39.96% | 0.00% | 5.09% | -7.10% | 1.24% |
| Home Improvement | 10.33% | -9.57% | -15.86% | 1.26% | -32.61% | 0.00% | 0.00% |
| Mortgage Refinance | -2.23% | -3.08% | 1.14% | 3.82% | -2.51% | -6.35% | -2.12% |
| Owner Occupant | 0.24% | -0.23% | 0.07% | -0.07% | -0.48% | 0.05% | 0.00% |
| Income/Loan[a] | 0.05% | 1.05% | 0.43% | -0.03% | -0.35% | 1.08% | 1.65% |
| Income[b] | 0.02% | -0.11% | -0.32% | 0.24% | -0.58% | -0.40% | 0.02% |
| Female | -2.08% | -0.41% | -2.34% | 0.55% | -3.84% | 0.47% | 0.49% |
| Minority | -7.94% | -10.61% | -12.20% | -9.48% | -15.43% | -8.55% | -10.09% |
| Median Income[c] | -3.59% | -2.73% | -2.52% | -2.81% | -3.87% | -0.89% | 0.47% |
| Minority %[d] | -0.44% | -2.72% | -9.34% | -0.92% | -3.44% | -4.68% | -2.42% |
[a]Income/Loan: for an increase in Applicant income equal to 25% of Loan Amount
[b]Income: for a $10,000 increase in Applicant Income
[c]Median Income: for a $15,000 decrease in median income of Applicant’s census tract
[d]Minority %: for a 20 percentage point increase in the minority percentage of Applicant’s census tract
Reverse Regression
Because minority status and lower credit worthiness are correlated, a model that does not control for credit worthiness will overstate the amount of discrimination that is occurring against minority applicants. However, reverse regression techniques can be used to find an underestimate of that discrimination.[17] In reverse regression, the dependent variable is assumed to be independent and the independent variable under consideration is assumed to be dependent. Thus, in this case, instead of examining the effect of race on the probability of a loan being originated, we examined the effect of a loan being originated on the probability of an applicant being a minority using the 1990 and 1991 county-wide data and controlling for all of the same factors. Tables 9 and 10 summarize the results of those analyses.
TABLE 9: Determinants of Probability of Minority Status of Mortgage Loan Applicants
New Haven County - 1990
| Factor | Coefficient (Std Error)[a] | Impact of Factor on Prob of Origination[b] |
|---|---|---|
| Government Insurance | 0.4402 (.1683) | 5.2% |
| Home Improvement Loan | -0.3939 (.2304) | -4.7% |
| Mortgage Refinancing | -0.7058 (.2128) | -8.3% |
| Owner-Occupied Property | -0.1806 (.4041) | -2.1% |
| Income-to-Loan Ratio | 0.0321 (.0149) | 0.4% |
| Applicant Income | -0.6776 (.2651) | -8.0% |
| Female | 0.5859 (.1584) | 6.9% |
| Median Income of Census Tract | 0.3034 (.8245) | 3.6% |
| Minority % of Census Tract | 4.5924 (.4227) | 54.2% |
| Loan Originated | -0.5156 (.1506) | -6.1% |
[a]Bolded coefficients are statistically significant at the 95% significance level.
[b]This impact is interpreted as the amount by which a change of one in a given factor will change the probability of an applicant being a minority for the average application. Thus, if the average application has a 13.7% chance of coming from a minority applicant, then an increase in applicant income of $100,000 (1 unit of applicant income) for the average applicant will decrease that probability by 8.0% to a 5.7% chance of being a minority.
TABLE 10: Determinants of Probability of Minority Status of Mortgage Loan Applicants
New Haven County - 1991
| Factor | Coefficient (Std Error)[a] | Impact of Factor on Prob of Origination[b] |
|---|---|---|
| Government Insurance | 0.7529 (.0979) | 4.8% |
| Home Improvement Loan | -0.3292 (.1200) | -2.1% |
| Mortgage Refinancing | -0.4876 (.0930) | -3.1% |
| Owner-Occupied Property | 0.4393 (.1625) | 2.8% |
| Income-to-Loan Ratio | 0.0274 (.0127) | 0.2% |
| Applicant Income | -0.4559 (.1242) | -2.9% |
| Female | 0.3695 (.0852) | 2.4% |
| Median Income of Census Tract | -1.4885 (.4289) | -9.6% |
| Minority % of Census Tract | 4.7870 (.2133) | 30.7% |
| Loan Originated | -0.6391 (.0769) | -4.1% |
[a]Bolded coefficients are statistically significant at the 95% significance level.
[b]This impact is interpreted as the amount by which a change of one in a given factor will change the probability of an applicant being a minority for the average application. Thus, if the average application has a 6.9% chance of coming from a minority applicant, then an increase in applicant income of $100,000 (1 unit of applicant income) for the average applicant will decrease that probability by 2.9% to a 4.0% chance of being a minority.
As the preceding tables show, a loan being originated decreased the probability that the applicant was a minority in a statistically significant manner in both 1990 and 1991. If this is in fact an underestimate of the amount of discrimination that occurred, then it appears that discrimination in lending on the basis of minority status did in fact occur in New Haven County in both 1990 and 1991. However, as discussed above, this analysis must be repeated separately for each mortgage lender using the 1991 data in order to generate conclusive results. The results of the reverse logits for New Haven Savings Bank, People’s Bank and McCue Mortgage Company can be found in Appendix 4. The reverse logits for Shawmut and First Federal did not yield usable results, probably due to insufficient sample size.
As is shown in Appendix 4, for New Haven Savings, People’s, and McCue, having a loan originated reduced the probability that the applicant was a minority in a statistically significant manner. For Shawmut and First Federal, the two lenders for which usable results were not obtained, it appears that loan origination again reduced the likelihood that the applicant was a minority though these results are weak.
Conclusions
The results of all of these analyses taken together strongly suggest that:
- In 1990 and 1991, minorities were under-represented as mortgage applicants.
- In l990 and 1991, minority mortgage applicants from New Haven County had a significantly lower chance of receiving a loan than white applicants with similar characteristics.
- Applicants from areas with a higher percentage of minorities or a lower median income were also less likely to have their applications approved, though these results are not as strong as those above.
These results appear to be stronger with larger sample sizes, so the county wide results are clearer than those for individual lenders. However, the patterns of discrimination found above do endure for individual mortgage lenders in 1991, particularly patterns involving minority status and minority percentage of census tract.
Further Research
In order to overcome the problem of small sample sizes in dealing with individual banks, we will repeat these analyses on a statewide basis. This will allow us to examine a larger number of applications for each bank, giving stronger results for those lenders examined above and results from lenders that we have so far been unable to examine. It will also allow us to determine whether or not the patterns identified above exist statewide, and whether or not there are other patterns discernible only at that level (such as inadequate credit provision to cities).
In order to confirm the results that we have for 1990 and 1991, we will repeat this entire process using the 1992 HMDA data after the October, 1993 release date. This will allow us to check the continuing validity of the model developed above. Recurrence of the patterns we have found will strengthen our resets.
Notes
[1]“Mortgage Lending in Boston: Interpreting HMDA Data” by Alicia H. Munnell, Lynn E. Browne, James McEneaney and Geoffrey M.B. Tootell, October 1992, Federal Reserve Bank of Boston, Working Paper Series Working Paper No. 92-7
[2]Myers, Samuel L. and Chan Tsze, “Racial Discrimination in Housing Markets: Implications For Middle Flight and Concentrations of Inner City Poverty,” (Draft) September 15, 1992
[3]Sec 12 U.S.C. §§ 28r02, 2803, 12 C.F.R. §203.4. HMDA reporting requirements do not apply to a bank that does not have its home office or a branch in an MSA, or a bank or mortgage lending institution with less than $10,000,000 in assets, 12 C.F.R. §203.3.
[4]Community Reinvestment Act, 12 U.S.C., §§ 290l et seq. See also Connecticut C.R.A., §§ 36-553 et seq Connecticut General Statutes
[5]Sullivan, Alane K. and Pozdena, Randall J., “Enforcing Anti-Redlining Policy under the Community Reinvestment Act,” Federal Reserve Bank of San Francisco Economic Review, Spring 1982
[6]Pub. L. 101-173, F.I.R.R.E.A., Title XII, §1212(b)
[7]Canner, Glenn B. and Smith, Delores S., “Home Mortgage Disclosure Act: Expanded Data on Residential Lending,” Federal Reserve Bulletin, November, 1991
[8]The law mandates the use of a four-tier rating system: “outstanding,” “satisfactory,” “needs improvement,” and “substantial non-compliance”
Federal Reserve Board Regulation BB, 12 C.F.R §228.7, lists twelve factors regulators should consider in assessing a financial institution’s record of performance for CRA purposes. These factors include (quoting relevant parts): (1)“Activities conducted by the… bank to ascertain the credit needs of its community, including the extent of the bank’s efforts to communicate with members of its community regarding the credit services provided by the bank”; (2)“The geographic distribution of… bank’s credit extensions, credit applications and credit denials”; (3)“The… bank’s participation, including investments in local community development and redevelopment projects or programs”; (4)“The… bank’s origination of residential mortgage loans, housing rehabilitation loans, home improvement loans, and small buttress or small farm loans within its community, or the purchase of such loans originated in its community”; (5)“the… bank’s participation in governmental-insured, guaranteed, or subsidized loan programs for housing, small business or small farms”; and (6)“Other factors that in the [Federal Reserve] Board’s judgement, reasonably bear upon the extent to which a… bank is helping to meet the needs of its entire community.”
[9]Canner, Glenn and Cleaver, Joe M., “The Community Reinvestment Act: A Progress Report,” Federal Reserve Bulletin, February, 1980
[10]Public Disclosure, August 31, 1992, “Community Reinvestment Act Performance Evaluation, People’s Bank 27334-2,” Federal Deposit Insurance Corporation, 160 Gould Street, Needham, Massachusetts 02194. “Outstanding” is defined as: “Outstanding record of meeting community credit needs”
“An institution in this group has an outstanding record of, and is a leader in, ascertaining and helping to meet the credit needs of its entire delineated community including low- and moderate income neighborhoods in a manner consistent with its resources and capabilities.”
To quote the examiners:
“Management’s internal analysis also stated ‘the data does not show a higher disparity of minority applicants denied on any income level compared to nonminority applicants.’ Examination findings found this statement to be true” Ibid, 11.
[11]Canner, Glenn, “The Community Reinvestment Act: A Second Progress Report,” Federal Reserve Bulletin, November, 1981
[12]Marisco, Richard, “A Guide to Enforcing the Community Reinvestment Act before the Board of Governors of the Federal Reserve System,” January, 1992
[13]F.I.R.R.E.A., passed in 1989, see note 6.
[14]Galster, George “Statistical Proof of Discrimination in Home Mortgage Lending,” The Review of Banking and Financial Services, November 20, 1991
[15]Munnell, Browne, McEneaney and Tootell, “Mortgage Lending in Boston”
[16]Moss, Alex and Porat, Naomi, “Residential Lending in New Haven Connecticut, Summary if Findings” Prepared for Harvey Koizim, April 27, 1989, See also: Junnarkar, Preethi and Koizim, Harvey “Bank Community Reinvestment Activity in New Haven, Connecticut,” September l990
[17]Conway, Delores A. and Roberts, Harry V., “Regression Analyses in Employment Discrimination Cases” in Statistics and the Law, edited by Morris H DeGroot, Stephen E. Fienberg, and Joseph B. Kadane, New York John Wiley and Sons, 1986
Bibliography
Canner, Glenn “The Community Reinvestment Act: A Second Progress Report,” Federal Reserve Bulletin, November, 1981.
Canner, Glenn and Cleaver, Joe M. “The Community Reinvestment Act: A Progress Report,” Federal Reserve Bulletin, February, 1980.
Canner, Glenn B. and Smith, Dolores S. “Home Mortgage Disclosure Act: Expanded Data on Residential Lending,” Federal Reserve Bulletin, November, 1991.
Convway, Delores A. and Roberts, Harry V. “Regression Analyses in Employment Discrimination Cases,” in Statistics and the Law, edited by Morris H. DeGroot, Stephen E. Fienberg, and Joseph B. Kadane, New York: John Wiley and Sons, 1986.
Galster, George “Statistical Proof of Discrimination in Home Mortgage Lending,” The Review of Banking and Financial Services, November 20, 1991.
Marsico, Richard “A Guide to Enforcing the Community Reinvestment Act before the Board of Governors of the Federal Reserve System,” January 2, 1992.
Moss, Alex and Porat, Naomi “Residential Lending in New Haven: Summary of Findings,” April 27, 1989.
Munnell, Alicia H., Browne, Lynn E., McEneaney, James, and Tootell, Geoffrey M. B. “Mortgage Lending in Boston: Interpreting HMDA Data,” Federal Bank of Boston Working Paper Series, October, 1992.
Myers, Samuel L. and Chan Tsze, “Racial Discrimination in Housing Markets: Implications For Middle Class Flight and Concentrations of Inner City Poverty,” (Draft) September 15, 1992.
Public Disclosure, August 31, 1992, “Community Reinvestment Act Performance Evaluation, People’s Bank 27334-2,” Federal Deposit Insurance Corporation, Needham, Massachusetts.
Sullivan, Alane K. and Pozdena, Randall J. “Enforcing Anti-Redlining Policy Under the Community Reinvestment Act,” Federal Reserve Bank of San Francisco Economic Review, Spring, 1982.
APPENDIX 1: 1990 HMDA Data For New Haven County
LOAN ORIGINATION BY RACE
| White | Black | Hispanic | Total | |
|---|---|---|---|---|
| Loan Originated | 1,635 | 151 | 81 | 1,867 |
| Percentage of Race | 78.0% | 72.2% | 65.9% | 76.9% |
| Loan Not 0nginated | 460 | 58 | 42 | 560 |
| Percentage of Race | 22.0% | 27.8% | 34.1% | 23.1% |
| Total | 2,095 | 209 | 123 | 2,427 |
| Percentage of Applicants | 86.3% | 8.6% | 5.1% |
LOAN ORIGINATION BY MINORITY CENSUS TRACT
| Minority Cen Tract | Not Minor Cen Tract | Total | |
|---|---|---|---|
| Loan Originated | 130 | 1,737 | 1,867 |
| Percentage of Census Tract | 71.0% | 77.4% | 76.9% |
| Loan Not Originated | 53 | 507 | 560 |
| Percentage of Census Tract | 29.0% | 22.6% | 23.1% |
| Total | 183 | 2,244 | 2,427 |
| Percentage of Applicants | 7.5% | 92.5% |
LOAN ORIGINATION BY LOW INCOME CENSUS TRACT
| Low Income Cen Tract | Not Minor Cen Tract | Total | |
|---|---|---|---|
| Loan Originated | 130 | 1,737 | 1,867 |
| Percentage of Census Tract | 71.0% | 77.4% | 76.9% |
| Loan Not Originated | 53 | 507 | 560 |
| Percentage of Census Tract | 29.0% | 22.6% | 23.1% |
| Total | 183 | 2,244 | 2,427 |
| Percentage of Applicants | 13.8% | 86.2% |
APPENDIX 2: 1991 HMDA Data For New Haven County
LOAN ORIGINATION BY RACE
| White | Black | Hispanic | Total | |
|---|---|---|---|---|
| Loan Originated | 11,001 | 362 | 254 | 11,617 |
| Percentage of Race | 77.5% | 57.4% | 60.2% | 76.2% |
| Loan Not Originated | 3,200 | 269 | 168 | 3,637 |
| Percentage of Race | 22.5% | 42.6% | 39.8% | 23.8% |
| Total | 14,201 | 631 | 422 | 15,254 |
| Percentage of Applicants | 93.1% | 4.1% | 2.8% |
LOAN ORIGINATION BY MINORITY CENSUS TRACT
| Minority Cen Tract | Not Minor Cen Tract | Total | |
|---|---|---|---|
| Loan Originated | 472 | 11,145 | 11,617 |
| Percentage of Census Tract | 58.1% | 77.2% | 76.2% |
| Loan Not Originated | 340 | 3,297 | 3,637 |
| Percentage of Census Tract | 41.9% | 22.8% | 23.8% |
| Total | 812 | 14,442 | 15,254 |
| Percentage of Applicants | 5.3% | 94.7% |
LOAN ORIGINATION BY LOW INCOME CENSUS TRACT
| Low Income Cen Tract | Not Low Inc Cen Tract | Total | |
|---|---|---|---|
| Loan Originated | 1,012 | 10,605 | 11,617 |
| Percentage of Census Tract | 65.0% | 77.4% | 76.2% |
| Loan Not Originated | 544 | 3,093 | 3,637 |
| Percentage of Census Tract | 35.0% | 22.6% | 23.8% |
| Total | 1,556 | 13,698 | 15,254 |
| Percentage of Applicants | 10.2% | 89.8% |
APPENDIX 3: Factors Affecting the Probability That a Loan Will Be Originated
| Variable | Descripion |
|---|---|
| GVTINS | Whether or not the application was insured under a govemrnent mortgage insurance program (FHA or VA). |
| HOMIMPRV | Whether or not the application was for a home improvement loan. |
| REFIN | Whether or not the application was a mortgage refinancing. |
| OWNOCC | Whether or not the application was for an owner-occupied property. |
| INCLOAN | The income-to-loan ratio for the application. |
| APPINCOM | The annual income of the applicant (in hundreds of thousands of dollars). |
| FEMALE | Whether or not the applicant was female. |
| APPMINOR | Whether or not the applicant was either black or Hispanic. |
| MEDINCOM | The median income of the census tract where the property is located (in hundreds of thousands of dollars). |
| PERMIN | The percentage of residents of the census tract where the property is located that are either black or Hispanic. |
APPENDIX 4A: Determinants of Probability of Minority Status of Mortgage Loan Applicants
New Haven Savings Bank - 1991
(1,748 Applications - 9.7% Minority)
| Factor | Coefficient (Std Error)[a] | Impact of Factor on Prob of Origination[b] |
|---|---|---|
| Government Insurance | 0.0000 (.0000) | 0.0% |
| Home Improvement Loan | -0.2687 (.3433) | -2.4% |
| Mortgage Refinancing | -0.5531 (.2371) | -4.9% |
| Owner-Occupied Property | 1.1822 (.3910) | 10.4% |
| Income-to-Loan Ratio | 0.0209 (.0354) | 0.3% |
| Applicant Income | -0.0868 (.2071) | -0.8% |
| Female | 0.8154 (.2184) | 7.2% |
| Median Income of Census Tract | -0.7173 (1.0832) | -6.3% |
| Minority % of Census Tract | 5.59545 (.5259) | 49.9% |
| Loan Originated | -0.7043 (.2243) | -6.2% |
[a]Bolded coefficients are statistically significant at the 95% significance level.
[b]This impact is interpreted as the amount by which a change of one in a given factor will change the probability of an applicant being a minority for the average application. Thus, if the average application has a 9.7% chance of coming from a minority applicant, then an increase in applicant income of $100,000 (1 unit of applicant income) for the average applicant will decrease that probability by 0.8% to a 8.9% chance of being a minority.
APPENDIX 4B: Determinants of Probability of Minority Stattus of Mortgage Loan Applicants
Peoples Bank - 1991
(1,004 Applications - 7.1% Minority)
| Factor | Coefficient (Std Error)[a] | Impact of Factor on Prob of Origination[b] |
|---|---|---|
| Government Insurance | 1.2569 (.4633) | 8.3% |
| Home Improvement Loan | -2.3036 (.7722) | -15.1% |
| Mortgage Refinancing | -0.7515 (.3272) | -4.9% |
| Owner-Occupied Property | -0.5842 (.6658) | -3.8% |
| Income-to-Loan Ratio | 0.1568 (.1397) | 1.0% |
| Applicant Income | -1.2188 (.6760) | -8.0% |
| Female | -0.2194 (.3628) | -1.4% |
| Median Income of Census Tract | -0.7395 (1.8392) | -4.9% |
| Minority % of Census Tract | 4.4270 (.8811) | 29.1% |
| Loan Originated | -0.8477 (.3140) | -5.6% |
[a]Bolded coefficients are statistically significant at the 95% significance level.
[b]This impact is interpreted as the amount by which a change of one in a given factor will change the probability of an applicant being a minority for the average application. Thus, if the average application has a 7.1% chance of coming from a minority applicant, then an increase in applicant income of $50,000 (1⁄2 unit of applicant income) for the average applicant will decrease that probability by 4.0% to a 3.1% chance of being a minority.
APPENDIX 4C: Determinants of Probability of Minority Status of Mortgage Loan Applicants
McCue Mortgage Coompany - 1991
(744 Applications - 10.8% Minority)
| Factor | Coefficient (Std Error)[a] | Impact of Factor on Prob of Origination[b] |
|---|---|---|
| Government Insurance | 0.5868 (.4882) | 5.6% |
| Home Improvement Loan | 0.0000 (.0000) | 0.0% |
| Mortgage Refinancing | -0.1406 (.6841) | -1.3% |
| Owner-Occupied Property | -3.4660 (1.1617) | -33.2% |
| Income-to-Loan Ratio | -0.2289 (1.4196) | -2.2% |
| Applicant Income | 1.9867 (1.3400) | 19.1% |
| Female | 0.8980 (.3453) | 8.6% |
| Median Income of Census Tract | -1.2401 (2.0244) | 11.9% |
| Minority % of Census Tract | 6.8909 (1.0903) | 66.1% |
| Loan Originated | -0.9211 (.3352) | -8.8* |
[a]Bolded coefficients are statistically significant at the 95% significance level.
[b]This impact is interpreted as the amount by which a change of one in a given factor will change the probability of an applicant being a minority for the average application. Thus, if the average application has a 10.8% chance of coming from a minority applicant, then an increase in applicant income of $100,000 (1 unit of applicant income) for the average applicant will increase that probability by 19.1% to a 29.9% chance of being a minority.