ABA Bank Compliance - January/February 2014

The answer is by utilizing “proxy” information. A proxy is a substitute; in this case, it means making assumptions about a person’s demographic characteristics based on what the bank already knows: the customer’s name (both first and last) and address.

First names can be used as predictors of gender, and surnames can predict race and ethnicity because certain surnames are more common within particular demographic groups than in others. Geographic location can also be an indicator of race and ethnicity. Pinpointing the census tract where a consumer lives, then identifying the minority percentage of the population within that tract provides a proxy for the consumer’s race and ethnicity. For instance, if a consumer lives in an area identified by the Census Bureau as 80 percent or more African-American, chances are high that consumer is African-American, according to logic.

Is this Reliable?

But how predictive is this method, really? For small portfolios or data sets, it’s hard to say. But for larger sample sizes, the law of large numbers increases the predictability (relative to a smaller sample) of the proxy data. Some questions remain unanswered even when dealing with a large sample. What about someone named Chris McNeal?

Is Chris a Christopher or a Christine?

And is McNeal a white or African-
American surname (statistically, McNeal
is about 50/50)? In some cases, names
cannot reliably be assigned to any group.
In these situations race and ethnic-
ity could be proxied by the minority
percentage of the census tract, or the
application or loan could be excluded
from the analysis, further shrinking the
dataset. But not considering records isn’t
unusual. HMDA analyses normally ex-
clude applications where the consumer
refused to complete the Government
Monitoring Information (GMI) section
of the application.

Once a dataset of loan information is assigned proxy information, the same types of fair lending analyses (disparity ratios, regression, matched pairs, and so forth) traditionally performed on mortgage files can be done.

Do We have to Do this?

But is using proxies the right way to go?

Is it a regulatory expectation that banks actually do this? Increasingly the answer is yes. In a letter authored by Richard Cordray in response to questions from Congress, and in a blog posting,

the Consumer Financial Protection Bureau (CFPB) affirmed its commitment to utilizing proxies in fair lending exams. The CFPB commented on its methodology when examining indirect auto lenders and how it determines whether lenders’ practices of paying dealer markup disproportionately affect minorities. Since banks don’t collect GMI for auto loans, proxies are used to establish the race, ethnicity, and gender of borrowers.

Not an Indirect Auto lender?
But you don’t make auto loans, you say.
And even if you do, you’re not in the
indirect business where you would pay a
dealer a markup. It doesn’t matter. The
letter to Congress makes the point that
“we have found frequent instances where
lenders had robust fair lending compli-
ance programs for mortgage lending,
but weak or non-existent fair lending
compliance programs for other types of
consumer lending.” While the letter (and
blog post) doesn’t require proxies per se,
how else could analyses be performed?
The letter goes on to state that the “Bu-
reau has encouraged lenders who are not
currently doing so to select a reasonable
proxy method and to monitor their data
for fair lending risk.”
While this all comes from the CFPB
and is not interagency guidance, more
generally focused attention on fair lend-
ing issues will naturally result in a subtle
“expectation” for all lenders that this be
done on non-HMDA data where fair
lending risk has been identified. This
is especially true for banks whose most
critical products are not HMDA-report-
able mortgage loans.