Thomas wrote:snood wrote:Thanks Thomas. I think the synopsis from the study answers the question that's been asked either directly or indirectly here several times, in one form or another: "Was the study fair, or did it not take the understandable risk the lenders face into account"?
I disagree. Compare the synopsis with the statistical model behind their analysis (page 13 of the paper document, page 15 of the PDF file.) When I look at the variables the model controls for, I don't really see anything like "lives in a bad neighborhood". The variables that come closest are "STATELAW2-STATELAW4" and "RURAL". But even they are very broad, and the other geographic variables are very broad.
Thus, if a bank charges higher rates from people in bad neighborhoods, and if blacks and hispanics disproportionately inhabit bad neighborhoods, the study's model would falsely record this as racial discrimination by the bank. The model doesn't really control for what the synopsis says it controls for.
But I admit that statistical models like this are closer to J_B's field than to mine. Are you still following this thread, J_B? And if so, would you mind taking a look at their model and tell me if I'm talking nonsense?
I'm still reading, but I have a couple comments from what I've read so far. They combined two separate databases to compile their data. The first database is public and the same data used by the Fed to publish their reports. The second database is 'proprietary' meaning, the data isn't available for review and their conclusions must be accepted on face value. The result of the merger is a database of 177,487 loans for review but they say they excluded any loans from the analysis where race/ethnicity was not provided. Their Tables 2 and 3 show the likihood ratios for various FICO and LTV categories of approx 67,000 loans or less than 40% of the subprime loans in their database. There is no discussion that I've seen so far (still looking) as to the breakout of the other 62% and whether the remaining data used in their analysis is an unbiased subset of the overall database.
Also, these are only subprime loans, loans made to individuals who already have credit risk factors. Making broader statements that lending institutions practice discrimination based on race is inappropriate for these data.
In the model they use, I see the same concern Thomas has, but I also see a major difference in how they analyzed the data and how the study from the Fed/CRC was presented (Table 1, pgs 8-9). The Fed/CRC study found no differences when controlling for risk factors, lending institution, FICO, and LTV ratios when looking at all loans. This study does not appear to control for lender the way the Fed/CRC study did. This, to me, would be a major question - does the same lending institution require different interest rates from borrowers based on race, all other factors being equal? There is no way to answer that question from this data.
They do a good job of stating pretty much everything I said above in the "Limitations" on pg 15 with the exception of my primary concern on the validity of the dataset representing un unbiased sample of subprime loans.
Even so, this doesn't mean the practice isn't commonplace, which I think it probably is. The paper states the "considerable leeway mortgage originators have to impose charges beyond those justified by risk-based pricing." (page 5). This was my point in my original post on this thread. The question isn't so much as 'do they' but 'why can they' in a highly regulated industry?