I have a group of people who review our HMDA loans on a monthly basis. We have approximately 3,000/year (and growing). I am trying to determine a fair error rate per individual in order to keep our data as accurate as possible. Is 5% overall too stiff? If you have a 5% error in any one field or 10% overall you would have to re-scrub and submit.

Has anyone established this? If so and you are willing to share I would greatly appreciate it!

If you're talking about overall error rate, 5% would mean about 3 errors per file (I think there are 59 fields). That's not an effective scrub, IMO.

Our process is 100% review of all loans but we do a 32 loan sample to see where we're at in terms of error rate. I've read and understand the sample sizes and thresholds per the CFPB. What I'm asking is how do I evaluate my employees error rate? I think it should align with the CFPB standards. And if each loan has 59 potential errors.....well, there's not much room for error.

So if a sample of 32 with 6 errors equals 18.75% error rate, That would not be acceptable for an exam. They would expand the sample and if we got to 8 errors then we'd have to re-scrub and file.

I think I should hold employees to the same standard but wanted to know what other institutions do.

The FRB just completed a HMDA data integrity exam of our 2016 & 2017 Lars. Their initial sample was 66 files. They told us we were allowed ONE error in the initial sample. If they found two, the sample would be expanded by an additional 140 files. The expanded sample wasn't needed. We also conduct a 100% review of all loan files, approximately 1300 annually and growing, and that is what I credit the 'no findings' data integrity exam results to.

While a 5% error rate per employee may sound reasonable, my experience shows the expectation is the error rate will be lower than that. I've approached the error rate as the bank as a whole, not per employee, as that is how our LAR is filed. We have conducted training, training, and more training, but employees are human and will make mistakes. We've set up processes, some automated and some manual, to catch those mistakes and correct them prior to submitting our data.

Does it take time and manpower - absolutely. Is it worth the time and manpower with those regulator accuracy expectations - my management team and BOD says definitely!

Is there similar guidance on CRA Small Business/Small Farm resubmission?

Bump

The Federal Reserve has a chart on p. 854 of the following (near the end on p.85 of the CRA section): https://www.federalreserve.gov/boarddocs/supmanual/cch/cch.pdf

Thanks.

Good afternoon!

The HMDA examination guidelines have recently changed. The resubmission thresholds are now focused only on error percentages in any key fields, not an overall number/percentage of files with errors. Depending on the volume of loans, if you have between 2.5% and 10% of errors in any one key field, you would be required to resubmit that data. The new guidelines do not refer to an overall error rate any longer. Any thoughts on this out there? How would you evaluate or communicate an overall error to management or the business units when multiple fields may be involved not just an overall rate? Thoughts are greatly appreciated!

Any thoughts are greatly appreciated! Thank you!

Any thoughts would be greatly appreciated! Thank you!

No one seems to be wanting to voice an opinion, so I'll go.

My opinion is your HMDA data needs to be very accurate. The 2.5%-10% error tolerance is a guideline, not an absolute. The OCC released this in March 2019:
Proper reporting of HMDA data is important in assessing the accuracy of the HMDA data that financial institutions record and report. Where errors that exceed established thresholds are identified in an institution’s HMDA data, the OCC supervisory office has discretion in requiring the institution to correct specific errors, without requiring resubmission of the data. The supervisory office may require resubmission of HMDA data when the inaccurate data are indicative of systemic internal control weaknesses that call into question the integrity of the institution’s entire HMDA data report.

You can find it here:
https://www.occ.gov/news-issuances/bulletins/2019/bulletin-2019-12.html

I expect the other regulators to have similar a position.

I don't think you have the right to be off by 2.5%-10% and expect to not have examiner criticism.

I'll chime in since I've dealt with examiners on this...Basically what I have experienced is that even though the thresholds (and examiners) may not trigger re-submission, anything other than isolated errors identified in the examiner review will draw criticism from the examiners. Anything within their sample that shows any "systemic" problems is going to garner a comment about corrective action. Even sufficient "random" errors will likely garner comment about your controls. I would not set the expectation for accuracy to be anywhere near the level of re-submission guidelines, your HMDA data is the first impression your regulator gets for how your institution runs...it is never a good idea to have that be a bad impression. In short I agree with David, it needs to be very accurate and I wouldn't consider 2.5% error rate very accurate.

Here's my thought on this. I believe the 8/23/17 interagency HMDA Examiner Transaction Testing Guidelines still apply to all data fields, even though the guidance on the key data fields is more lenient than the 8/23/17 guidance. In the guidance for key data fields (for example, FDIC FIL 51-2017), the agencies still make disclaimers that non-key fields could be reviewed as appropriate. For example, FIL 51-2017 says this:

"However, in certain circumstances, consistent with the FFIEC Guidelines, such examination staff may determine that it is necessary to review additional HMDA data fields, as appropriate."

In addition, the March 2019 OCC guidance David cited also says this:

"In certain circumstances, however, and consistent with the FFIEC guidelines, examination staff may determine that it is appropriate to review additional HMDA data fields[/b]."

To me, this says that you can't focus solely on the key fields and just ignore the other fields. So if I were conducting internal testing, I would want to be as clean as possible and, therefore, would probably utilize the 8/23/17 guidance without (much) regard to the key fields guidance. In other words, I wouldn't just test the "key fields" but would evaluate my LAR - and possible voluntary resubmission requirements - based on the original 8/23/17 guidelines. The idea is that if you evaluate yourself under the stricter guidelines, you should be sufficiently prepared in the event that examiners use the discretion outlined in the key field guidance. That said, your primary efforts should go to ensuring the key fields are as clean as possible.

To answer your specific question:

I would recommend communicating this by following the 8/23/17 guidance and explaining the standards outlined there and how your results compared with this. You could then split out the performance in key fields and non-key fields and make a recommendation (for resubmission) based on performance/results of testing.

I'm sure you are well aware of the 8/23/17 guidance, but if you are looking for a summary of it, here is an article I wrote that summarizes that guidance: https://www.compliancecohort.com/blog/what-is-an-acceptable-hmda-error-rate

I think the communication to management has to broken down by errors per field now. An overall error rate doesn't capture risk well. If you sample 30 files and find 15 errors, that doesn't really tell you much about your risk. Neither 50% (15/30 files) or 0.45% (15/(30x110) fields) really conveys anything meaningful. If there were 5 errors each in 3 different key fields, you've got a big problem. If there were 15 errors in different non-key fields, then not so much.

We presented something like:

There were X fields out of Y total fields that had error rate above the resubmission threshold. Z of these fields were key fields.

and

There were X fields out of Y total fields that were within 2% (or whatever your risk tolerance wants to allow) of the error rate that would require resubmission. Z of these fields were key fields.