The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
orkmg raper series The Cultural Affinity Hypothesis and Mortgage Lending Decisions William C. Hunter and Mary Beth W alker Working Papers Series Issues in Financial Regulation Research Department Federal Reserve Bank of Chicago July 1995 (W P -95-8) C FEDERAL RESERVE BANK OF CHICAGO T h e C u l t u r a l A f f in it y H y p o t h e s is a n d M o r t g a g e L e n d i n g D e c i s i o n s William C. Hunter Senior Vice President and Director of Research Federal Reserve Bank of Chicago 230 South LaSalle Street Chicago, IL 60604 312/322-5800 Mary Beth Walker Policy Research Center Department of Economics Georgia State University Atlanta, GA 30303 404/651-3751 July 1995 ‘The views expressed in this paper are those of the authors and not necessarily those of the Federal Reserve Bank of Chicago, the Board of Governors of the Federal Reserve System, or the Georgia State University Policy Research Center. We thank Lynne Browne and Jeff Tootell of the Federal Reserve Bank of Boston for providing us with the Boston Fed HMDA data and acknowledge the helpful comments of Robert Avery, Jeffrey Gunther, Martin Hellwig, Kenneth Koford, Jeffrey Lacker, Leonard Nakamura, Mary Rosenbaum, Stephen Smith, Richard Sweeney, David Walker, and an anonymous referee, as well as the assistance provided by Janet Colter, Jon Hao, and Jann Howell. Th e C u l tu r a l A ffinity H ypothesis a n d M o r t g a g e L ending D ecisions A well-known result in the economics of information is that when job applicants have different productive potentials, employers will find it worthwhile to subject them to a screening process. Since productivity typically cannot be observed directly, some observable characteristic that is highly correlated with productivity may serve as a substitute. Thus, possession of a college degree or higher level of education may indicate attributes that will prove useful in certain kinds of jobs. Similarly, race or sex, whether or not justifiable on legal or ethical grounds, may be taken by employers as predictors of on-the-job performance. In credit markets, many borrower characteristics serve as signals of borrower quality or creditworthiness. Just as employers may associate the race or sex of job applicants with productivity, so may lenders associate these attributes with creditworthiness. Since the Civil Rights Act of 1964, much government legislation has been passed that aimed at eliminating the use of unalterable personal characteristics such as race in credit approval decisions. Nonetheless, the question remains whether lenders continue to use such information in that way. A recent hypothesis put forward by Calomiris, Kahn, and Longhofer (1994) suggests that continued discrimination against minorities in credit market situations could result from the lack of cultural affinity between white loan officers and minority applicants. This hypothesis states that white loan officers will rely more heavily on characteristics that can be observed at low cost when appraising the creditworthiness of minorities rather than invest marginal resources in gathering additional information about creditworthiness. Stated differently, ifthe majority of loan officers and applicants are white, white loan officers may feel they know more about white applicants than about minorities, and thus they are more likely to acquire additional information 2 about the creditworthiness of white applicants. On the other hand, we would expect these lenders to rely more heavily on basic objective loan application information in appraising the creditworthiness of minorities. This study uses mortgage application data collected under the Home Mortgage Disclosure Act of 1975 (HMDA) to provide an empirical test of an extended version of the cultural affinity hypothesis. W e examine whether loan officers perceive such objective information as credit history or reputation and financial leverage differently for minority applicants than for whites. Both characteristics would appear to provide useful information to lenders. If high-quality borrowers find it cheaper to build reputations (superior credit histories) than low-quality borrowers do, then lenders will prefer those borrowers with better reputations— their reputations derive directly from their superior abilities. Using a proxy measure of prior credit market experience in our model also allows us to examine the so-called "thicker file" phenomenon. Since accepted marginal white mortgage applicants are many times observed to have thicker loan application files than rejected marginal minorities, the common presumption is that the marginal white applicants have received special counseling or extra coaching from sympathetic loan officers while the minorities have not. The data used in this study, H M D A data for the Boston area in 1990, have been criticized for being unusually "dirty"; that is, they are missing many values and contain many coding errors.1 W e used a rigorously cleaned subset of the original data which maintains the general profile of the full original data set. The criteria used to clean the data are discussed more fully in the following section. 3 Our empirical results suggest that lenders do treat objective loan application information differently, depending on the race of the applicant. In particular, credit history and the ratio of total monthly obligations-to-total monthly income both have a substantially greater impact on the probability of loan approval for minorities than for whites. Interestingly, we do not find that gender has a significant impact on the probability of loan approval. In section 1 which follows, we present a sketch of recent legislation addressing mortgage credit issues and briefly summarize earlier studies of discrimination in mortgage credit markets. Our data sources and definitions, as well as our statistical model, are described in section 2. The empirical results are presented and discussed in section 3. A summary and conclusion follow in section 4.1 1. Previous Studies Following the passage of the Civil Rights Act of 1964, interest in the issue of discrimination in housing and mortgage credit markets increased significantly. Several additional pieces of legislation were passed in an attempt to outlaw the practice. The Fair Housing Act of 1968 (amended in 1988), the Equal Opportunity Credit Act of 1974, the Home Mortgage Disclosure Act of 1975 (HMDA), and the Community Reinvestment Act of 1977 (CRA) were enacted in part to outlaw discrimination in housing and mortgage credit markets on the basis of factors such as race, age, sex, and geographic location. While the Fair Housing Act and the Equal Opportunity Credit Act addressed the general issue of access to housing and credit, H M D A and C R A (as administered) dealt more directly with mortgage credit. H M D A and C R A were passed to address the perceived problems of housing credit not flowing properly to all 4 neighborhoods within communities at large, and in particular, the failure of some mortgage lenders to adequately serve all segments of their primary trade areas. While H M D A required certain lenders to report, by census tract, the number and dollar value of home loans they made in their communities each year, C R A went one step further by encouraging (through the regulatory process) depository institutions to help meet the credit needs of their communities, including low- and moderate-income neighborhoods, consistent with safe and sound lending practices. Early studies of discrimination in mortgage markets, such as Bradbury, Case, and Dunham (1989) and Hutchinson, Ostas, and Reed (1977), examined questions related to the issue of redlining— the practice whereby lenders refuse to make mortgage loans in geographic areas characterized by heavy concentration of racial or ethic minorities regardless of the creditworthiness of the loan applicants. As noted by Benston (1981), most studies of redlining are inadequate since they fail to control sufficiently for borrower characteristics (see also Benston and Horsky 1992). In addition, as noted by Holmes and Horvitz (1994), these studies do not adequately control for the risk differences across different geographic areas. More recent studies of redlining have produced somewhat mixed results. Holmes and Horvitz (1994), in their study of redlining in Houston, Texas, fail to find clear evidence of the practice, while the paper by Canner, Gabriel, and Woolley (1991) examining nationwide data finds more evidence of it.2 Beginning in 1990, lenders were required by H M D A to publicly report the gender, income, and race of loan applicants as well as the action taken on each loan (accepted, denied, or withdrawn by the applicant). The initial H M D A report indicated that mortgage applicants from black and Hispanic households were systematically denied mortgage loans at a higher rate 5 than applicants from white households with similar incomes. H M D A data released since 1991 have showed essentially the same disparate rejection rate. Needless to say, these releases have generated much public concern. However, leading industry groups and individuals in government and academia have maintained that it would be inappropriate to draw the conclusion from these releases that mortgage lenders actively discriminate against minorities. This is because the H M D A data do not take into account information crucial to credit decisions, such as the loan applicant's credit history, other debts, and employment history. Partly in response to this debate, the Federal Reserve Bank of Boston conducted a study of mortgage denial rates in the Boston metropolitan area (Munnell, Browne, McEneaney, and Tootell, 1992) using a much wider range of loan application data. These augmented H M D A data are described briefly in section 2. As a result of taking account of the personal characteristics of the borrowers, the Boston study reduced the magnitude of discrepancy for black and Hispanic applicants from 2.7 times the white denial norm to 1.6 times. Thus, while allowing for differences in loan applicant wealth and credit history reduces race-related differences in mortgage denial rates, it apparently does not eliminate them.3 The so-called "thicker file" phenomenon, as an explanation of the remaining race-related denial rate differential, rests on anecdotal evidence that many marginal white applicants have thicker loan application files than marginal black and Hispanic applicants. It ispossible that such extra documentation, when itexists, provides mitigating information that reduces the probability that the marginal white application will be rejected. There are numerous explanations of why these thicker files might be observed. White applicants may, on average, have had more prior exposure to the credit process and therefore may come better prepared. It might also be that, as 6 a result of cultural familiarity, loan officers spend more time, on average, with white applicants, probing more deeply into whether they have evidence to offset factors that might otherwise lead to rejection.4 2. The Data and Statistical Model The data used in this study were taken from the augmented 1990 H M D A data collected by the Federal Reserve Bank of Boston. Augmentation of the H M D A data was necessary because applicant and loan characteristics collected under H M D A are severely limited. The Boston Fed, with the support of other supervisory agencies, obtained additional information on the borrowers and loan applications included in the 1990 H M D A report from the Boston Metropolitan Statistical Area. The Boston Fed data file included a random sample of 3,300 conventional applications made by whites and 722 conventional applications made by blacks and Hispanics. Boston Fed economists used the additional data, combined with census information on neighborhood characteristics, to develop a model of the determinants of mortgage lending decisions in the Boston area. The additional data collected on borrower and loan characteristics included variables measuring the borrower's ability to support the loan (net wealth, the ratio of total debt payments to income, and liquid assets), previous credit history (bankruptcy filings and number of credit reports in the file), default loss potential (loan-to-value ratios), personal characteristics of the applicant (age, marital status, number of dependents), and an estimate of the probability of being unemployed, among others. A detailed description of this data is available in Munnell, Browne, McEneaney, and Tootell (1992). 7 In our analysis of the Boston Fed data set, only conventional loans not associated with any special housing or lending programs were included. Since the original data set had a fairly large number of data errors and missing values, we used the criteria developed by Carr and Megbolugbe (1993) to purge the raw data. This included, for example, deleting applications with a ratio of housing expenses to total monthly income greater than 100 percent and deleting applications with a ratio of the loan amount to the purchase price in excess of 3.0, among others. W e deleted approximately 1,000 observations because they contained data errors or were atypical relative to the overall sample. Of the data that remained, there were 1,726 loan applications approved and 265 denied. Among the 1,516 white applicants, the approval ratio was about .9; for the group of 475 black and Hispanic applicants, the approval ratio was .76. The characteristics of our subsample mirror those of the original data set. Our statistical analysis of cultural affinity effects in mortgage credit markets is conducted using the framework of dichotomous choice models. Essentially, cultural affinity effects can be viewed as information that is unobservable to the econometrician and statistical tests for the effects are conducted using the approach developed in the paper by Albrecht (1981). Intuitively, the approach tests whether white loan officers' decisions on white applicants depend less on formal information, such as credit history, financial obligations, and the like, than they do for minorities.5 The lender's decision of which loan applicants to approve can be modelled as accepting only those applicants whose perceived creditworthiness (or marginal productivity) exceeds some critical level. W e assume that the creditworthiness (or marginal productivity), Ik, 8 of the kth applicant as perceived by the lender can be expressed as a linear combination of borrower characteristics; that is, Ik = xk'p + ek , (1) where P is a vector of parameters and xk is the vector of informational variables for the kth applicant. The error term ek represents factors known to the lender but unobservable to the modeler; it is assumed to be random noise with mean zero and variance G 2. Of course, this continuous index of creditworthiness, Ik, is not observed. W e observe only whether a mortgage application is accepted or denied. Let yk = = 1 if Ik is > 0, i.e., the application is accepted, and 0 otherwise. Then, the probability that the kth applicant will be granted a loan can be written as P* yk = 1) = F(xk'p), (2) where F(-) is the cumulative distribution function of e, assuming that e has a symmetric distribution. Note that because the P parameters are identified only up to a positive scalar, our estimates are actually estimates of p/G. 9 3. Empirical Results W e estimate the logit model using 1,991 observations from the 1990 HMDA data file. The basic empirical results obtained from the estimation of the logistic regression are given in table 3. The dependent variable in this regression is the action taken on the mortgage application, i.e., dependent variable = 1 ifthe application was accepted, and zero if it was denied. A number of control variables other than the reputation and credit experience variables mentioned above were also used in the model. These include standard financial ratios such as housing expenses relative to monthly income, and total debt obligations relative to monthly income, as well as other information on the applicant such as marital status, employment status, and whether the loan applicant had a cosigner. The explanatory variables are described fully in table 1. Even more than in the usual logistic regression model, the estimated effects of the variables are difficult to interpret because of the multiple interactions terms. To compute the effects of the variables, we must specify the information set we use for conditioning. Suppose we write the equation that specifies the index of creditworthiness for loan applicants as Ik = xk'P + a,CHIST + tXj-ARAC + Oj-ASEX + a 4SCH + a 5THK + oCg-LOC + ctjCSCHARAC) + otg(SCH-ASEX) + a^SCHTHK) + a 10(SCHLOC) + a , ,(CHIST •ARAC) + a 12(CHISTASEX) + a 13(CHISTTHK) + a 14(CHIST LOC) + a 15(OBRAT ARAC) + error . (5) 10 Then, the direct effect of the credit history (CHIST) signalling variable on creditworthiness is given by 3I/3CHIST = a„ given A R A C = A S E X = T H K = L O C = 0. (6) This coefficient measures the effect of credit history on creditworthiness given that the applicant is black or Hispanic (ARAC=0), female (ASEX=0), does not have a "thick" credit file (THK=0), and wants to buy property in a location with a high vacancy rate (LOC=0). An alternate set of conditioning information would give a different effect. For example, the total impact of credit history on creditworthiness with a different conditioning information set is given by 3I/8CHIST = a, + a,„ given A R A C = 1, A S E X = T H K = L O C = 0. (7) Equation (6) gives the impact of credit history on creditworthiness given that the applicant is a white female with a thin credit file and the property to be financed is located in a high vacancy area. Many of the results in table 3 are interesting, though not surprising. For example, as the ratio of total debt obligations to monthly income (OBRAT) rose, the probability of loan approval fell. A 10 percentage point increase in this ratio lowered the probability of approval by more than 8 percent, and this effect was statistically significant. Applicants with prior public record defaults, i.e., bankruptcies, had a significantly lower probability of being approved. The analysis also revealed that a self-employed applicant had a probability of loan approval more than 3 11 percentage points lower than an applicant who was not self-employed but otherwise had exactly the same characteristics. It is interesting to note that some variables seemingly had no effect on loan approval; for example, the number of dependents had an insignificant impact. Our model specification provides no evidence to support the thicker file hypothesis; the estimated direct effect of T H K was insignificant regardless of the set of conditioning information. It is possible that this was due to the proxy used to gauge the thickness of the loan applicant's file. An applicant's file was considered "thick" if it contained more than two credit reports. A better measure of previous credit market experience may well have yielded different results. The location variable also produced mostly insignificant results. Our basic finding was that loan approval was somewhat easier to obtain if the property was located in a census tract with a vacancy rate higher than the median for the metro area. This result, however, was not significant at the 5 percent level. Insignificant results were also obtained when other proxies defining good and bad location were used. In particular, when the quality of location was measured using a dummy variable indicating whether the minority population in the census tract of the property exceeded 30 percent, insignificant results were found. To analyze the impact of race, we must consider the different variables that interact with the race variable. In particular, in our specification, race interacted with OBRAT, the ratio of total monthly obligations to total monthly income. Because O B R A T is a continuous variable, the impact of race on the creditworthiness of a loan applicant will depend on the level of OBRAT. Thus, the coefficient on race alone is not very informative. W e computed the effect of race using two levels of the debt obligation ratio, one at the lowest quartile and one at the third quartile. Conditioning on "bad" levels of credit history and education (CHIST=SCH=0), 12 we found the effect of race to be small and statistically insignificant for a relatively low level of OBRAT; the coefficient was .626 with a standard error of .396. When O B R A T rose, however, we found that the coefficient became larger (1.1463) and statistically significantly different from zero (standard error was .37). These results suggest that race was relatively unimportant to mortgage lenders when the objective measure of monthly obligations indicated a high-quality applicant. As this objective measure worsened, the results suggest that white applicants were judged more creditworthy than minority applicants. Note that these calculations were performed holding all other applicant characteristics constant. Itis also interesting to compute the effects of the race variable allowing credit history and education to take different values. When we set CHIST=SCH=1, we found that race was completely insignificant for a low value of OBRAT, and only marginally significant for a higher OBRAT. Again, the results suggest that for applicants who were high-quality in terms of their objective measures, race seemed unimportant to lenders. It appears to be the marginal minority applicant whose creditworthiness was reduced because of race. With respect to the education level of the borrower, we found no evidence that education played an important role in the accept/reject decision. The only interaction variable involving education level that was statistically significant was that for education and location (the coefficient on S C H L O C was statistically significantly different from zero at a 1 percent significance level). However, this result may have been a statistical anomaly (Type I error), because we found no empirical evidence that either education (SCH) or location (LOC) had a statistically significant direct effect on creditworthiness. The lack of significance of education may be explained by the binary nature of this variable— we could only distinguish between 13 applicants who had completed high school and those who had not; we did not have sufficient information to distinguish the exact number of years of schooling beyond high school. The credit history variable (CHIST) produced a more interesting set of results. The estimated direct effect of credit history on the index of creditworthiness was positive and statistically significant for all sets of conditioning information. Not surprisingly, this means that having a good credit rating increased the probability of loan approval for both white and minority applicants and for males and females. An interesting point is that the effect was stronger (the estimated coefficient was larger and the standard error smaller) for minority applicants than for white applicants. Predicted Probabilities of Approval An interesting and somewhat different interpretation of the statistical results appears in table 4. This table shows the predicted probabilities of approval and changes in approval probabilities for applicants with varying characteristics. We focused on the changes associated with four applicant characteristics: race, credit history, obligation ratio, and location. We held the values of the other variables constant. We measured the continuous variables (e.g., HRAT and UEMP) at their means while assuming that the other categorical variables in the sample had the value held by the majority of applicants (e.g., SELF=0). The first set of entries in table 4 shows the different effects the variable credit history had on loan approval for white and minority applicants. The probability of loan approval for a white male buying property in a low vacancy census tract fell about 5 percent as his credit rating went from good to bad; even with a bad credit history, his probability of loan approval was still estimated to be slightly over 90 percent. 14 Here, we set the obligation ratio at the mean value for white applicants. In contrast, a minority male, with obligation ratio set to the mean value for minorities, buying a home in a low-vacancy census tract saw his estimated probability of loan approval fall to about 82 percent as his credit rating fell, a change of nearly 13 percent. On the other hand, a white loan applicant with the objective measures set to good values has a predicted probability of loan approval of 96 percent, only slightly higher than the 94 percent probability predicted for a minority applicant with similar characteristics. Now compare "high-quality" applicants to "low-quality" applicants. If a white male with a good credit history, average obligation ratio, seeking to purchase property in a good neighborhood, suddenly found his race changed to black or Hispanic, his probability of approval would have dropped by only 2 percentage points. If this same white male applicant had a bad credit history, average obligation ratio, and was seeking to purchase property located in a bad neighborhood, his probability of approval would have dropped by 8 percentage points if his race suddenly changed to black or Hispanic. Stated differently, if a black male with a bad credit history, purchasing property in a bad neighborhood, suddenly found his race changed to white, his probability of approval would have increased by 8 percentage points, while if his credit history was good and his neighborhood of choice was good, the change in race would have increased his probability of approval by only 2 percentage points. Finally, we focused on the impact of raising the obligation ratio on the probability of loan approval. That the obligation ratio was an important determinant of loan approval is obvious from the basic logit results in table 3. The coefficient was a substantial negative number with a small standard error—hardly a surprising result. However, when we analyzed the differential 15 impact of race, the results were surprising. First, consider a white male with poor objective measures of creditworthiness; that is, his credit history was bad, he was buying property in a high vacancy census tract, and his ratio of monthly obligations to monthly income was high, OBRAT=60.0.6 This loan applicant had a predicted approval probability of close to 70 percent. The same white male with an obligation ratio of only 30.0 (recall the mean for whites is 32.3) had a 94 percent chance of loan approval. A minority male with this same profile (bad credit history, bad location, OBRAT=30.0) had a predicted approval probability of 88 percent, not too far from the prediction for whites. But if this minority male had an obligation ratio of 60.0, his predicted loan approval probability was less than 16 percent, far lower than the 70 percent predicted for a similar white male. Implications of the Findings These findings are important for several reasons. First, we found no evidence to support the "thicker file" explanation for the disparate rejection rates for black and Hispanic applicants versus white applicants, at least in the Boston Fed mortgage study sample. Second, we also found no evidence that loan officers in the Boston metropolitan area engaged in the practice of redlining on the basis of the economic status of the neighborhood in which the property was located (status measured by vacancy rates, the percent of population at or below the poverty level, and the percentage of the population belonging to a minority group). Overall, the analysis suggests that the marginal minority applicants in our sample were more likely the "victims" of statistical discrimination resulting from the lack of a cultural affinity with white lending officers as opposed to invidious racial discrimination. This is suggested by 16 the behavior of the race variable under differing sets of conditioning information. For applicants with good credit profiles, race was not a significant factor in the accept/reject decision. However, race was very significant for those with bad credit histories, and a bad credit history had the effect of lowering the probability of approval for minorities by an amount appreciably larger than for whites. Similarly, race became an important factor for high levels of the monthly obligation ratio; our results indicate that minority applicants with high debt ratios were very much less likely to receive loan approval than similar white applicants. While these results are consistent with the cultural affinity hypothesis put forth by Calomiris, et al. (1994), it is also possible to interpret these results as the outcome of a simple rational Bayesian updating procedure used by loan officers. Under this interpretation, the precision of estimates of the probability of repayment conditional on all available information, including observable applicant characteristics, improves as loan officers process more and more applications. Clearly, if mortgage loan officers process more white applications than minority applications, the precision of the posterior probability distribution of repayment associated with white applicants will be greater than that for minority applicants. By processing more white applications and observing the outcomes of the lending decisions, the loan officer has more information with which to determine how various applicant characteristics, including less objective (and potentially mitigating) factors, impact the outcome of the loan. Thus, it is quite possible for factors other than credit history to matter for whites, possibly mitigating the impact of a weak credit history, while at the same time credit history may continue to play a dominant role in the accept/reject decision for minorities. 17 While this interpretation might be suggestive of the logic used by proponents of the "thicker file" explanation, it is not the same. The thicker file explanation rests on a presumed linkage between the thickness of white applicants' files and the possibility that they received special treatment in the form of counselling or coaching during the application process. The cultural affinity and statistical discrimination explanations do not depend on any special assistance being given white applicants. They only require that loan officers have a more informative posterior probability distribution of repayment for whites than for minorities. 4. Summary and Conclusion The search for statistical evidence of racial discrimination in economic life is exceedingly difficult. The results obtained in the literature using approaches that rely exclusively on the sign and significance of a racial dummy variable in a standard regression equation generally appear inconclusive. In this paper, we examined the cultural affinity hypothesis put forth by Calomiris, et al. (1994). This hypothesis states that white loan officers, because of a lack of familiarity with minority applicants, will rely more heavily on characteristics that can be observed at low cost (e.g., objective loan application measures) in evaluating the creditworthiness of minority applicants. An interesting feature of this hypothesis is that it allows us to test whether mortgage lenders reward objective measures of borrower quality differently for whites than for minorities. In this way the methodology allows us to conduct a more stringent test of mortgage lender discrimination on the basis of race, sex, and property location. 18 The empirical model predicts the probability that a mortgage applicant will receive loan approval. We focused on four objective borrower characteristics that are generally believed to be important factors in measuring borrower quality. These are the applicant's education level, credit history, previous credit market experience, and the ratio of total monthly obligations to monthly income. Previous credit market experience was used to test the "thicker file" hypothesis. To allow for the possibility that these factors were perceived differently by lenders depending on the race of the applicant, each of these four variables was interacted with the applicant's race. Our results, based on mortgage application data collected and reported by the Federal Reserve Bank of Boston, indicate strong support for the cultural affinity hypothesis. We found empirical evidence that certain borrower characteristics were treated differently by lenders. In particular, the loan applicant's credit history and monthly obligation ratio appeared to be assessed differently for minority (black and Hispanic) borrowers than for whites. Interestingly, for highquality borrowers (applicants with good credit histories and relatively low monthly obligation ratios), race seemed unimportant. The predicted probability of loan approval was very similar for white and minority applicants with good objective measures. For more marginal borrowers, race was much more important. In one case, a minority loan applicant with poor objective measures had only about a 16 percent chance of loan approval. A white applicant with the same poor objective measures had nearly a 70 percent chance of loan approval. The empirical results regarding the applicant’s education and the thickness of the applicant's file were disappointing. Neither education nor thickness of file appeared to be a significant factor in the loan approval process. We speculate that the results on education may have been due to the fact that our data on education were quite limited; we could only determine 19 whether the applicant had completed high school. Our results on thickness o f file may have been due to a poor measure o f credit market experience. These findings, while suggestive, are based on data from the Boston area and cannot be generalized to all U.S. mortgage markets. Nonetheless, this appears to be a fruitful line of research, in that it implies that disparate loan rejection rates for whites and minorities are not the result of all minority applicants' being forced to meet higher standards than white applicants. Instead, it is among the marginal applicants where differences in the credit assessment process seem related to race. Stated differently, those candidates who, by objective measures, appear to be poor credit risks, are the ones for whom race matters. 20 Notes 1. There were also a number of criticisms leveled at the first study to use the data, Munnell, et al. (1992). We respecified the model used in the Boston Fed study in order to examine whether public information about loan applicants was systematically treated differently by mortgage lenders. 2. The methodology used by Holmes and Horvitz (1994) has received considerable criticism, making their empirical results somewhat suspect. 3. The conclusions of the Boston Fed Study have been challenged and criticized by Home (1994), Liebowitz (1994), and Zandi (1993) for containing numerous data errors, e.g., miscoded and incorrect information, among others. However, Carr and Megbolugbe (1993) conducted a detailed analysis of the Boston Fed Study database, systematically eliminating erroneous data and atypical observations, and found that race was still a significant factor in the loan application accept/reject decision. Based on their analysis, Carr and Megbolugbe concluded that their results presented an even stronger case for discrimination than was originally reported in the Boston Fed Study. A recent working paper by Bostic (1994), reexamining the Boston Fed's findings, reported evidence of significant racial effects. Minority applicants were found to face substantial negative biases in terms of debt-to-income requirements but favorable biases regarding loan-tovalue requirements relative to identical white applicants. Similar to the results reported in this paper, Bostic found that the biases held only for the marginal applicants. 4. It is also suggested that the applications of black and Hispanic applicants failed to meet the underwriting guidelines required for selling these mortgages in the secondary market more often than those of white applicants. 5. In the context of the model presented below, this will result in significantly different coefficients between whites and minorities for these coefficients, rather than different cut-off ratios.6 6. The value of 60.0 for OBRAT is higher than approximately 90 percent of the values in the sample. References Albrecht, James. "A Procedure for Testing the Signalling Hypothesis," Journal o f Public Economics 15 (1981), 123-32. Benston, George J. "Mortgage Redlining Research: A Review and Critical Analysis," Journal of Bank Research 11 (1981), 8-23. ____________________ and Dan Horsky. "The Relationship Between the Demand and Supply of Home Financing and Neighborhood Characteristics: An Empirical Study of Mortgage Redlining," Journal of Financial Services Research 5 (1992), 235-260. Bostic, Raphael. "Patterns of Racial Bias on Mortgage Lending," Working Paper, Department of Economics, Stanford University, 1994. Bradbury, Katharine, Karl E. Case, and Constance R. Dunham. Mortgage Lending in Boston, 1982-1987," New "Geographic Patterns of England Economic Review (September/October 1989), 3-30. Calomiris, Charles W., Charles M. Kahn, and Stanley D. Longhofer. "Housing-Finance Intervention and Private Incentives: Helping Minorities and the Poor," Journal o f Money, Credit and Banking 26 (August 1994), 634-674. Canner, Glenn B., Stuart A. Gabriel, and J. Michael Woolley. "Default Risk and Mortgage Redlining: A Study of the FHA and Conventional Loan Markets," Southern Economic Journal 58 (July 1991), 249-262. Carr, James H., and Isaac F. Megbolugbe. "The Federal Reserve Bank Study on Mortgage Lending Revisited," Fannie Mae Working Paper, Office of Housing Research, 1993. Holmes, Andrew, and Paul Horvitz. "Mortgage Redlining: Race, Risk, and Demand," Journal o f Finance 49 (March 1994), 81-99. Home, David K. "Evaluating the Role of Race in Mortgage Lending," FDIC Banking Review 7 (Spring/Summer 1994), 1-15. Hutchinson, Peter M., James R. Ostas, and J. David Reed. "A Survey and Comparison of Redlining Influences in Urban Mortgage Lending Markets," AREUEA Journal 5 (1977:4), 463-472. Liebowitz, Stan. "A Study That Deserves No Credit," Wall Street Journal (September 1, 1994), A-14. Munnell, Alicia H., Lynne E. Browne, James McEneaney, and Geoffrey Tootell. "Mortgage Lending in Boston: Interpreting the Data," Working Paper, Federal Reserve Bank of Boston, 1992. Zandi, Mark. "Boston Fed's Bias Study was Deeply Flawed," American Banker (August 19, 1993). Table 1 D e s c r ip t io n of E x pl a n a t o r y V a r ia b l e s HRAT Ratio of monthly housing expenses to monthly income. OBRAT Ratio of total monthly obligations to monthly income. MHIST Mortgage payment history: 1 if two or fewer mortgage payments are recorded as late, 0 otherwise. PUB Public records credit history: 1 if there are public record defaults, 0 otherwise. SELF Self-employed: 1 if applicant is self-employed, 0 otherwise. CHIST Consumer payments credit history: 1 if there is no history of delinquent credits (defined as one or more accounts 60 days or more past due), 0 otherwise. UEMP Estimate of probability of unemployment by industry. MULTI Type of property purchased: 1 if applicant is purchasing a two- to four-family home, 0 otherwise. COSIGN Co-signer of application: 1 if co-signer, 0 otherwise. MS Marital status: 1 if married, 0 otherwise. LNPR Loan-to-price ratio. DEP Number of dependents. SCH Years of education: 1 if applicant has more than 12 years in school, 0 otherwise. THK Thickness of file: 1 if there are more than two credit reports in the file, 0 otherwise. ARAC Race of applicant: 1 if applicant is white, 0 otherwise. ASEX Sex of applicant: 1 if applicant is male, 0 otherwise (i.e., female or information not available). LOC Characteristic of location: 1 if tract vacancy is less than the MSA median, 0 otherwise. SCHRAC SCH*ARAC SCHSEX SCH*ASEX SCHTHK SCH*THK SCHLOC SCH*LOC CHRAC CHIST*ARAC CHSEX CHIST* ASEX CHTHK CHIST*THK CHLOC CHIST*LOC OBRAC OBRAT*ARAC Table 2 S u m m a r y S ta tistic s fo r V ariables Variable . Sia n I I U I in M odel HRAT 25.227 7.523 M liiiiiS ■l l l l l i l l 0.52 OBRAT 32.719 8.259 0.52 79.00 MHIST 0.975 0.155 0.00 1.00 PUB 0.931 0.254 0.00 1.00 SELF 0.119 0.323 0.00 1.00 CHIST 0.802 0.398 0.00 1.00 UEMP 3.776 2.054 1.80 10.60 MULTI 0.136 0.342 0.00 1.00 COSIGN 0.029 0.167 0.00 1.00 MS 0.612 0.488 0.00 1.00 .821 .166 DEP 0.713 1.065 0.00 7.00 SCH 0.727 0.446 0.00 1.00 THK 0.104 0.306 0.00 1.00 ARAC 0.761 0.426 0.00 1.00 ASEX 0.775 0.418 0.00 1.00 LOC 0.533 0.499 0.00 1.00 SCHRAC 0.585 0.493 0.00 1.00 SCHSEX 0.558 0.497 0.00 1.00 SCHTHK 0.076 0.266 0.00 1.00 SCHLOC 0.402 0.490 0.00 1.00 CHRAC 0.639 0.480 0.00 1.00 CHSEX 0.628 0.484 0.00 1.00 CHTHK 0.080 0.271 0.00 1.00 CHLOC 0.438 0.496 0.00 1.00 OBRAC 24.594 0.00 79.00 LNPR 'In thousands of dollars. 15.50 .011 72.00 2.57 L ogit R esults o n Table 3 M ortgage A cceptances Dependent Variable = 1 if mortgage was accepted, 0 if mortgage was denied. §ji VARIABiUMl jName $$$$$$$. H S E stimated! ! ! ! tandard E R illte HRAT 0.019223 0.012428 1.5468 OBRAT -0.12174 0.018657 -6.5251 MHIST 0.75848 0.36495 2.0783 PUB -1.5735 0.21758 -7.2320 SELF -0.64513 0.21603 -2.9862 CHIST 1.3570 0.37768 3.5929 UEMP -0.093505 0.033145 -2.8211 MULTI -0.60786 0.19700 -3.0855 COSIGN 1.0994 0.56227 1.9552 MS 0.45762 0.17267 2.6502 LNPR -1.1461 0.48565 -2.3599 DEP -0.091568 0.073145 -1.2519 SCH -0.56899 0.42101 -1.3515 THK -0.24733 0.52209 -0.47372 ARAC -0.99226 0.79523 -1.2478 ASEX -0.42984 0.44151 -0.97357 LOC -0.69419 0.36295 -1.9126 SCHRAC -0.30177 0.36058 -0.83690 SCHSEX 0.72755 0.42585 1.7085 SCHTHK 0.053618 0.49735 0.10781 SCHLOC 0.72478 0.34161 2.1216 CHRAC -0.39771 0.34889 -1.1400 CHSEX -0.23831 0.37715 -0.63186 CHTHK 0.11223 0.47534 0.23610 CHLOC 0.23793 0.33181 0.71707 OBRAC 0.0578 0.0202 2.8702 Log-likelihood = -780.94 Log-likelihood function = -613.33 Likelihood ratio test = 335.228 with 26 d.f. Table 4 P r e d ic t e d P ro ba b ility o f L o a n A ppr o v a l P redM ^ M PR O B ^P® P O F A P lI if ttl ARAC=W, CHIST=good, LOC=good, OBRAT=RW .9629 ARAC=W, CHIST=bad, LOC=good, OBRAT=RW .9087 ARAC=M, CHIST=good, LOC=good, OBRAT=RM .9453 ARAC=M, CHIST=bad, LOC=good, OBRAT=RM .8166 ARAC=W, CHIST= good, LOC=good, OBRAT=RM .9587 ARAC=M, CHIST=good, LOC=good, OBRAT=RM .9453 ARAC=W, CHIST= bad, LOC=bad, OBRAT=RM .8962 ARAC=M, CHIST=bad, LOC=bad, OBRAT=RM .8119 ARAC=W, CHIST= bad, LOC=bad, OBRAT=60 .6929 ARAC=W, CHIST= bad, LOC=bad, OBRAT=30 .9379 ARAC=M, CHIST=bad, LOC=bad, OBRAT=60 .1551 ARAC=M, CHIST= bad, LOC=bad, OBRAT=30 .8762 -.0542 -.1287 -.0134 -.0843 .245 .7211 1 W=White; M=Minority; good and bad credit histories are defined in Table 1; RM—mean of obligation ratio (OBRAT) variable for minorities=34.06; RW—mean of obligation ratio (OBRAT) variable for whites=32.3. Note that RM and RW are not statistically significantly different. 2 Change in the approval probability associated with an applicant moving from the set of characteristics given in the top row of each panel to those given in the bottom row. Working Paper Series A series of research studies on regional economic issues relating to the Seventh Federal Reserve District, and on financial and economic topics. REGIONAL ECONOMIC ISSUES Estimating Monthly Regional Value Added by Combining Regional Input With National Production Data WP-92-8 Philip R. Israilevich and Kenneth N. Kuttner Local Impact of Foreign Trade Zone David D. Weiss Trends and Prospects for Rural Manufacturing WP-92-9 WP-92-12 William A, Testa State and Local Government Spending—The Balance Between Investment and Consumption WP-92-14 Richard H. Mattoon Forecasting with Regional Input-Output Tables WP-92-20 P.R. Israilevich, R. Mahidhara, and G.J.D. Hewings A Primer on Global Auto Markets Paul D. Ballew and Robert H. Schnorbus WP-93-1 Industry Approaches to Environmental Policy in the Great Lakes Region WP-93-8 David R. Allardice, Richard H. Mattoon and William A. Testa The Midwest Stock Price Index-Leading Indicator of Regional Economic Activity WP-93-9 William A. Strauss Lean Manufacturing and the Decision toVertically Integrate Some Empirical Evidence From the U.S. Automobile Industry WP-94-1 Thomas H. Klier Domestic Consumption Patterns and the Midwest Economy WP-94-4 Robert Schnorbus and Paul Ballew 1 Workingpaperseriescontinued To Trade or Not toTrade: Who Participates in RECLAIM? WP-94-11 Thomas H. Klier and Richard Mattoon Restructuring & Worker Displacement in the Midwest Paul D. Ballew and Robert H. Schnorbus Financing Elementary and Secondary Education in the 1990s: A Review of the Issues WP-94-18 WP-95-2 Richard H. Mattoon Community Development-Fiscal Interactions: A Review of the Literature WP-95-6 William H. Oakland and William A. Testa Community Development-Fiscal Interactions: Theory and Evidence from the Chicago Area WP-95-7 William H. Oakland and William A. Testa ISSUES IN FINANCIAL REGULATION Incentive Conflict in Deposit-Institution Regulation: Evidence from Australia WP-92-5 Edward J. Kane and George G. Kaufman Capital Adequacy and the Growth of U.S. Banks Herbert Baer and John McElravey WP-92-11 Bank Contagion: Theory and Evidence WP-92-13 George G. Kaufman Trading Activity, Progarm Trading and the Volatility of Stock Returns James T. Moser WP-92-16 Preferred Sources of Market Discipline: Depositors vs. Subordinated Debt Holders WP-92-21 Douglas D. Evanoff An Investigation of Returns Conditional on Trading Performance WP-92-24 James T. Moser and Jacky C. So The Effect of Capital on Portfolio Risk atLife Insurance Companies WP-92-29 Elijah Brewer ///, Thomas H. Mondschean, and Philip E. Strahan 2 Workingpaperseriescontinued A Framework forEstimating the Value and InterestRate Risk of Retail Bank Deposits WP-92-30 David E. Hutchison, George C. Pennacchi Capital Shocks and Bank Growth-1973 to 1991 Herbert L Baer and John N. McElravey WP-92-31 The Impact of S&L Failures and Regulatory Changes on the CD Market 1987-1991 WP-92-33 Elijah Brewer and Thomas H. Mondschean Junk Bond Holdings, Premium Tax Offsets, and Risk Exposure atLife Insurance Companies WP-93-3 Elijah Brewer 111and Thomas H. Mondschean Stock Margins and the Conditional Probability of Price Reversals Paul Kofman and James T. Moser IsThere Lif(f)e After DTB? Competitive Aspects of Cross Listed Futures Contracts on Synchronous Markets WP-93-5 WP-93-11 Paul Kofman, Tony Bouwman and James T. Moser Opportunity Cost and Prudentiality: A RepresentativeAgent Model of Futures Clearinghouse Behavior WP-93-18 Herbert L Baer, Virginia G. France and James T. Moser The Ownership Structure ofJapanese Financial Institutions Hesna Genay Origins of the Modem Exchange Clearinghouse: A History of Early Clearing and Settlement Methods atFutures Exchanges WP-93-19 WP-94-3 James T. Moser The Effect of Bank-Held Derivatives on Credit Accessibility Elijah Brewer 111, Bernadette A. Minton and James T. Moser Small Business Investment Companies: Financial Characteristics and Investments WP-94-5 WP-94-10 Elijah Brewer III and Hesna Genay 3 Workingpaperseriescontinued Spreads, Information Flows and Transparency Across Trading System WP-95-1 Paul Kofman and James T. Moser The Cultural Affinity Hypothesis and Mortgage Lending Decisions WP-95-8 William C. Hunter and Mary Beth Walker MACROECONOMIC ISSUES An Examination of Change in Energy Dependence and Efficiency in the Six Largest Energy Using Countries-1970-1988 WP-92-2 Jack L. Hervey Does the Federal Reserve Affect Asset Prices? WP-92-3 Vefa Tarhan Investment and Market Imperfections in the U.S. Manufacturing Sector Paula R. Worthington WP-92-4 Business Cycle Durations and Postwar Stabilization of the U.S. Economy Mark W. Watson WP-92-6 A Procedure for Predicting Recessions with Leading Indicators: Econometric Issues and Recent Performance WP-92-7 James H. Stock and Mark W. Watson Production and Inventory Control atthe General Motors Corporation During the 1920s and 1930s WP-92-10 Anil K. Kashyap and David W. Wilcox Liquidity Effects, Monetary Policy and the Business Cycle WP-92-15 Lawrence 7. Christiano and Martin Eichenbaum Monetary Policy and External Finance: Interpreting the Behavior of Financial Flows and InterestRate Spreads WP-92-17 Kenneth N. Kuttner Testing Long Run Neutrality WP-92-18 Robert G. King and Mark W. Watson 4 Workingpaperseriescontinued A Policymaker's Guide to Indicators ofEconomic Activity WP-92-19 Charles Evans, Steven Strongin, and Francesca Eugeni Barriers to Trade and Union Wage Dynamics Ellen R. Rissman WP-92-22 Wage Growth and Sectoral Shifts: Phillips Curve Redux WP-92-23 Ellen R. Rissman Excess Volatility and The Smoothing of Interest Rates: An Application Using Money Announcements WP-92-25 Steven Strongin Market Structure, Technology and the Cyclicality of Output Bruce Petersen and Steven Strongin The Identification of Monetary Policy Disturbances: Explaining the Liquidity Puzzle WP-92-26 WP-92-27 Steven Strongin Earnings Losses and Displaced Workers Louis S. Jacobson, Robert J. LaLonde, and Daniel G. Sullivan Some Empirical Evidence of the Effects on Monetary Policy Shocks on Exchange Rates WP-92-28 WP-92-32 Martin Eichenbaum and Charles Evans An Unobserved-Components Model of Constant-Inflation Potential Output WP-93-2 Kenneth N. Kuttner Investment, Cash Flow, and Sunk Costs WP-93-4 Paula R. Worthington Lessons from the Japanese Main Bank System forFinancial System Reform in Poland WP-93-6 Takeo Hoshi, Anil Kashyap, and Gary Loveman Credit Conditions and the Cyclical Behavior of Inventories Anil K. Kashyap, Owen A. Lamont and Jeremy C. Stein WP-93-7 5 Workingpaperseriescontinued Labor Productivity During the Great Depression Michael D. Bordo and Charles L Evans Monetary Policy Shocks and Productivity Measures in the G-7 Countries WP-93-10 WP-93-12 Charles L Evans and Fernando Santos Consumer Confidence and Economic Fluctuations John G. Matsusaka and Argia M. Sbordone WP-93-13 Vector Autoregressions and Cointegration Mark W. Watson WP-93-14 Testing for Cointegration When Some of the Cointegrating Vectors Are Known WP-93-15 Michael T. K. Horvath and Mark W. Watson Technical Change, Diffusion, and Productivity Jeffrey R. Campbell Economic Activity and the Short-Term Credit Markets: An Analysis of Prices and Quantities WP-93-16 WP-93-17 Benjamin M. Friedman and Kenneth N. Kuttner Cyclical Productivity in a Model of Labor Hoarding Argia M. Sbordone The Effects ofMonetary Policy Shocks: Evidence from the Flow ofFunds WP-93-20 WP-94-2 Lawrence J. Christiano, Martin Eichenbaum and Charles Evans Algorithms for Solving Dynamic Models with Occasionally Binding Constraints WP-94-6 Lawrence J. Christiano and Jonas DM. Fisher Identification and the Effects of Monetary Policy Shocks Lawrence J. Christiano, Martin Eichenbaum and Charles L Evans WP-94-7 Small Sample Bias in G M M Estimation ofCovariance Structures Joseph G. Altonji and Lewis M. Segal WP-94-8 6 Workingpaperseriescontinued Interpreting the Procyclical Productivity ofManufacturing Sectors: External Effects ofLabor Hoarding? WP-94-9 Argia M. Sbordone Evidence on Structural Instability in Macroeconomic Time Series Relations James H. Stock and Mark W. Watson WP-94-13 The Post-War U.S. Phillips Curve: A Revisionist Econometric History Robert G. King and Mark W. Watson WP-94-14 The Post-War U.S. Phillips Curve: A Comment WP-94-15 Charles L Evans Identification of Inflation-Unemployment Bennett T. McCallum The Post-War U.S. Phillips Curve: A Revisionist Econometric History Response to Evans and McCallum W P-94-16 WP-94-17 Robert G. King and Mark W. Watson Estimating Deterministic Trends in the Presence of Serially Correlated Errors WP-94-19 Eugene Canjels and Mark W. Watson Solving Nonlinear Rational Expectations Models by Parameterized Expectations: Convergence to Stationary Solutions WP-94-20 Albert Marcet and David A. Marshall The Effect of Costly Consumption Adjustment on Asset Price Volatility WP-94-21 David A. Marshall and Nayan G. Parekh The Implications of First-Order Risk Aversion for Asset Market Risk Premiums WP-94-22 Geert Bekaert, Robert J. Hodrick and David A. Marshall Asset Return Volatility with Extremely Small Costs of Consumption Adjustment WP-94-23 David A. Marshall 7 Workingpaperseriescontinued Indicator Properties of the Paper-Bill Spread: Lessons From Recent Experience WP-94-24 Benjamin M. Friedman and Kenneth N. Kuttner Overtime, Effort and the Propagation of Business Cycle Shocks WP-94-25 George J. Hall Monetary policies in the early 1990s~reflections of the early 1930s WP-94-26 Robert D. Laurent The Returns from Classroom Training forDisplaced Workers WP-94-27 Louis S. Jacobson, Robert J. LaLonde and Daniel G. Sullivan Isthe Banking and Payments System Fragile? WP-94-28 George J. Benston and George G. Kaufman Small Sample Properties of G M M for Business Cycle Analysis Lawrence J. Christiano and Wouter den Haan WP-95-3 The Fed Funds Futures Rate as a Predictor ofFederal Reserve Policy Joel T. Krueger and Kenneth N. Kuttner WP-95-4 Capital Utilization and Returns to Scale WP-95-5 Craig Burnside, Martin Eichenbaum and Sergio Rebelo 8
@FedFRASER