Mark Zandi's article, "Boston Fed's Study Was Deeply Flawed" (American Banker, Aug. 19), criticizing the Boston Fed's examination of racial patterns in mortgage lending in the Boston area, combined erroneous statements and analysis to reach an unsupportable conclusion.
That Mr. Zandi made such errors is surprising as Boston Fed researchers spent considerable time going over the study with him in his role as a consultant to a lender that participated in the study.
Mr. Zandi . states that the study did not consider the state of the New England economy and housing market. In fact, the study controlled for economic conditions and any differential effects on minority applicants by including the unemployment rates of the applicants' occupations as well as extensive data on applicants' finances.
Housing market conditions were captured in the loan-to-value ratio and in a number of measures of the risks to property values in different locations.
Furthermore, contrary to Mr. Zandi's assertion that prices in the Boston area fell more for low-priced homes than for highpriced homes, Karl E. Case, a visiting scholar at the Boston Fed, pointed out in an article in American Banker (Aug. 25) that in the same period relevant to our study, "higher-priced homes suffered the biggest drop in value."
Similarly, Mr. Zandi's statement that the closure of the Bank of New England produced a credit crunch is irrelevant, since the Bank of New England failed in 1991 and the study examined 1990 data.
Mr. Zandi also wrongly claims that the presence of cosigners and several other important variables were ignored in the Boston Fed study.
The Question of Data Errors
With respect to Mr. Zandi's comments about data errors, extensive edits and call-backs to the institutions were made to verify the accuracy of the information. The example Mr. Zandi cites as an "obvious encoding error." the application with a 946% loan-to-value, was checked with the lending institution before the analysis began.
The property required rehabilitation and the loan was rejected for the very reason that the loan-to-value ratio was so high.
Mr. Zandi's final statements that the analysis should be based upon a matched sample ignores the fact that the "matching" is performed by the regressions.
Mr. Zandi did not state how he matched loans. The Boston Fed, however, ran regressions over many subsamples to confirm the robustness of its conclusions and consistently found that race entered the mortgage decision process.
Standing by Study
Most recently, in response to Mr. Zandi's assertion about "matching" files, we examined the sample of applicants seeking to purchase single-family homes and having good credit histories, total obligation ratios less than 40% (slightly above guidelines), and loan-to-values less than 80%. Matched along these dimensions, the effect of race was still statistically significant.
In conclusion, the Boston Fed's examination of mortgage lending in Boston was a careful study that was subjected to intensive review by experts before release. The results were very robust.
Mr. Zandi Replies:
Lynne Browne's comments misinterpret my criticims of the Boston Fed study.
Ms. Browne needs to examine the home price data more carefully. Karl Case's home price data show that home prices in Suffolk County, where 60% of the black and Hispanic applicants included in the Fed's study reside, fell by nearly 20% for lower-priced homes and only marginally for higher-priced homes in 1990.
The Fed study's use of the unemployment rate for massachusetts in 1989 to reflect what was occuring in the Boston econmy in 1990 drastically understates the economy's impact on mortgage lending decisions.
|Plagued with Errors'
The Bank of New England's severe downsizing in 1990, which presaged its failure in early 1991, clearly had an adverse impact on lending in Boston in 1991.
Although the Fed researchers published regression statistics including a co-singer, they did not publish results for the other variables I cited. If they had, it would have illustrated just how sensitive their results are to specific variables.
As other researchers are now discovering, the data underlying the Fed study are plagued with errors, only some of which were caught by the Boston Fed.
The Boston Fed ran a matched sample only after I challenged them on this point. Using matched samples greatly reduces race's statistical significance, a point omitted in Ms. Browne's comments.