Mortgage discrimination has been turned into today's most important civil rights issue.
What seemed compelling evidence of discrimination was offered last October by the Federal Reserve Bank of Boston, which studied Boston's mortgage market in 1990.
The bank purported to show that blacks and Hispanics were 60% more likely to be rejected for a mortgage loan than whites even after differences in financial, employment, and neighborhood characteristics were taken into account.
The alarming conclusion has received widespread attention. But regulators and others fail to appreciate how sensitive the results are to the statistical techniques employed in the study.
In fact, the results are so sensitive to technique that the conclusion seems unfounded.
Price Declines Ignored
A careful evaluation of the study was made possible this spring when the Fed released data on the applications while ensuring that no applicant or lender could be identified.
My analysis employed the same statistical techniques the Fed used. For several reasons. however. the Fed's analysis fails to adequately represent the mortgage lending process.
One critical factor the Fed failed to consider was the state of the economy and housing markets in Boston during 1990.
Home prices were declining. Homes selling at prices among the bottom third during the period fell by 19%. Midpriced homes experienced declines of 9%, high-priced homes only 2%.
With low-priced homes falling more rapidly in value than higher-priced homes, purchasers of lower-priced homes were rejected by lenders more often.
Because black and Hispanic homebuyers generally purchase lower-priced homes. Boston lenders rejected blacks and Hispanics in greater proportion than whites. This was not discrimination per se, but simply good underwriting
The potential rise in foreclosures was a serious issue for Boston lenders, many of whom were under extreme regulatory pressure to reign in runaway credit problems during this period, especially after the failure of the Bank of New England in early 1991. The result was a credit crunch of unprecedented magnitude in the region.
Other Key Omissions
The Fed study also curiously omits other variables important in explaining mortgage-lending decisions. These include whether the applicant's credit history met the lender's guidelines; whether the borrower submitted information that could not be verified; the presence of a cosigner; and the loan amount.
Including these variable in the statistical analysis reduces the rejection rate for a black or Hispanic from 60% greater than a white applicant's to 23% greater.
Wrong Question
Nor does the Fed study make adjustments for what appear to be obvious data encoding errors. One mortgage applicant was listed with a loan-to-value ratio of 946%. Correcting for this and other errors in the initial loan-to-value ratio further reduces the rejection rate for blacks and Hispanics -- to 14% greater than a that of white applicants.
Yet another problem arises from the way the Fed chose to measure the degree of discrimination. It sought to answer the question: How would the denial rate for white applicants change had they been treated like a black or Hispanic applicant?
Since the white and minority applicants included in the Fed study had very different characteristics, the study's results are substantially different than if the question was: Would denials for blacks or Hispanics decrease had they been treated like whites?
The whole issue of which question is appropriate could have been avoided had the Fed analysis been performed on a matched sample -- one in which white applicants are matched with blacks and Hispanics sharing some key characteristics.
Wrong Turn
The Fed researchers elected not to use a matched sample because they did not want to prejudge the causes of rejection. But having determined the significant factors influencing mortgage lending in their unmatched analysis, it would then seem appropriate to conduct the analysis using a matched sample.
Using a matched sample does away with the impact of race on mortgage lending decisions. The probability that a black or Hispanic will be rejected falls from 14& greater than a white applicant to a statistically insignificant 3% greater.
Thus, by integrating economic conditions into the analysis, correcting obvious coding errors, and using a matched sample, the rejection rate for blacks is just 3% greater than for whites.
I do not know if mortgage lenders discriminated against minority applicants in Boston in 1990 -- but neither does the Boston Fed.
