Banks lend on statistics, just as insurance companies have rates based upon the same. And insurance companies, for the most part, can assign rates based upon average accident rates, for identifiable traits.
Like one's sex. Or where they live (a common way to get statistics without 'race/colour').
So what do I mean by the chicken and the egg, here?
Well, imagine that black landowners go bankrupt more often. And sadly, this could be because black landowners go bankrupt more often!
If a specific statistical classification indicates more bankruptcy, then it becomes true. For example, bank $a is bold enough to lend to someone whom they view as, statistically, more likely to fail. Then.. a few years later?
That person needs a fiscal injection. Were that person not black, it would be easier to get funding at a time of need. However, since that person is?
Well, then they get more denials, go bankrupt, and add to the proof that "statistically, more black people go bankrupt". Or as the banks are likely doing it, "the area where people live in".
In other words, unless you can make EVERYONE, 100% not deny based upon an statistical models? Race/gender/sexual orientation/sex/where they live, etc, etc? Anyone, like bank $a lending outside of those models, will be lending to people less able to find additional assistance when/if the time comes.
Chicken and egg. It's not right, but I view this a little bit like certain regulations, where corporations can't compete against other corporations which 'cut corners', so we have to mandate all corps behave in a certain manner to level the playing field.