Does anyone know what the examiners consider a "disparity" when performing statistical analysis of a bank's HMDA data against that of their peers? For example, if 14% of the subject bank's loans came from majority and high minority tracts and for their peers it was 19%, is that 5% difference considered a substantial disparity?

I can't find anything in actual examination guidelines, etc. that says the regulators will consider a difference of XX% is their threshold for being a substantial difference.

And you won't find anything that says a negative variance below X% is acceptable. It is all going to come down to a number of potential factors -
1. How many years have you underperformed against your peers (e.g., this is the first year or you have in consecutive years showing a pattern).
2. Your presence in that market.
3. Branch distribution and efforts in those markets.

The percentage will come into play but their is no bright line standard.

Also bank size and AA. I've seen community banks in major cities (e.g. LA and Houston) that have taken the city as the assessment area. The latter had 921 tracts, 2/3rds majority minority. A community bank could not adequately serve the same way as a super regional or large bank.

One other way to look at the lending is to determine the overall percentage of the bank's lending in the AA and compare the percentage in the LMI or MM tracts to see if they are underrepresented. If so, look for the reasons, such as accessibility, products, etc. If a potential applicant has to pass-by 40 competitors to reach your door, there is a chance they will not consider you. Also, larger banks have larger marketing budgets, something a smaller bank would not have.

You want to look at statistical significance more than % difference, and you have to look at volume to determine what the statistical significance is. If your peer is 19% and you are 10%, but you only have 10 apps in that AA, then you are 1 app away from peer. However, if you have a 1,000, then that 9% difference is pretty significant.

In general, a p-value less than .05 is generally the norm for determining significance, but I'd ask your examiner how they determine statistical significance and what their threshold is.

The t-Test and paired t-Test...my favorite!

I have inquired about the significance level and confidence level that was used previously and this was not shared so let me know if you ever hear something on that front.

In my experience, statistical significance of the differences in at the 5% level will always raise questions, and exam teams may use statistical significance at the 10% level if the institution's counts of applications/loans in a geography are small.