Have you ever wondered what is an acceptable HMDA error rate for your financial institution’s HMDA LAR? You know, how many errors can you have before the regulators would make you scrub your data and resubmit your entire LAR? Well, understanding the HMDA resubmission standards is fairly easy now, though that hasn’t always been the case.
Background on HMDA Error Rate Tolerances
For years, it was somewhat of a mystery as to what was an acceptable HMDA error rate. In fact, when I first started scrubbing HMDA file years ago, I had a very difficult time gaging how many HMDA errors were acceptable before a resubmission would be required, or worse - before a civil money penalty (CMP) would be assessed.
The challenge I faced in regards to HMDA error rates was a result of several factors. At first, I was unable to find any guidance from examiners in regards to what their HMDA error thresholds were. Then, in the late 2000’s, I noticed that the FDIC had created some internal standards which I could understand only by talking to examiners. The challenge was that during this period of time, I counted two or three changes to the way the FDIC calculated HMDA errors and how those errors could lead to a full resubmission. While this guidance was great when I had the current version, I never really knew when the HMDA resubmission standards would change and I soon found out that each regulator didn’t necessarily follow the same standards.
Then around 2013, I noticed that the CFPB had released their own standards (which seemed fairly similar to the FDIC standards), but the CFPB HMDA HMDA error rate calculations where necessarily adopted by all of the regulatory agencies. Then in early January of 2016, just a few months after the 2015 final HMDA rule changes were released, the CFPB started seeking comments in regards to the existing HMDA error rate standards and resubmission guidelines.
FFIEC Transaction Testing Guidelines
On August 23, 2017, the Federal Financial Institutions Examination Council (FFIEC) released joint guidance called HMDA Examiner Transaction Testing Guidelines. This guidance was issued jointly by the FDIC, OCC, Federal Reserve, NCUA, and the CFPB and was effective for how examiners at each agency will validate the accuracy of HMDA data collected beginning on January 1, 2018. The guidance runs seven pages and includes the following sections:
Ethnicity or Race Data Errors
HMDA Transaction Testing Sample Sizes and Thresholds
HMDA Testing Procedures
After a short introduction to the Examiner Transaction Testing Guidelines, the first part of the interagency guidance relates to testing procedures. This section includes the first eight procedures which include selecting a random sample, documenting how the sample was selected, verifying the accuracy of data entries against corresponding loan files, prioritizing data fields for review, dividing transaction testing into two stages, calculating the number of errors per data field, correcting data field entries, and including previously omitted LAR entries.
While this article is not going to explore each procedure in detail, there are a few things worth noting. First, procedure #4 clarifies that supervisory agencies may prioritize designated fields for review, meaning that some HMDA reviews may not look at all data fields within a sample. For example, FDIC FIL 51-2017 provides a list of 37 key data fields that will be focused on during HMDA reviews (though that guidance is clear that other data fields could be reviewed according to FFIEC guidance).
In addition, procedure # 5 explains that there are two stages to the testing procedures. The first stage requires examiners to review only a subset of the sample, or the initial sample, which will depend on the size of the financial institution’s LAR. If the number of errors identified in the initial sample falls below a certain threshold (which is included in the interagency guidance in Column B of the HMDA Table), no further sample review is required and examiners can conclude the transaction testing. If, however, the number of errors equals or exceeds the initial sample threshold, then examiners must proceed to Stage 2 to expand their sample size and review more loan files. Basically, if errors are discovered during the initial sample, then the sample size of the review will be increased.
The next section in the interagency HMDA testing procedures focuses on tolerances. The guidance explains that examiners should not count certain differences between data in the HMDA LAR and the loan file error for purposes of determining whether the number of errors equals or exceeds the “initial sample threshold” or the “resubmission threshold” (both found in the guidance). The differences that examiners should not count include:
Three calendar days or less in the date the application was received or the date shown on the application form.
One thousand dollars or less in the amount of the covered loan or the amount applied for.
Three calendar days or less in the date of the action taken by the financial institution, provided that such differences do not result in reporting data for the wrong calendar year.
Rounding errors in reporting the dollar amount, rounded to the nearest thousand, of the gross annual income relied on in making the credit decision or, if a credit decision was not made, the gross annual income relied on in processing the application.
Ethnicity or Race Data Errors
The next procedure (#10) listed in the interagency guidance deals with ethnicity or race data errors. Specifically, procedure #10 clarifies that for ethnicity or race, a data field consists of a group of Filing Instruction Guide (FIG) fields as follows:
The Ethnicity of Applicant or Borrower data field group—comprised of six FIG fields with information on an applicant’s or borrower’s ethnicity (FIG Data Field Numbers 19-24);
The Ethnicity of Co-Applicant or Co-borrower data field group—comprised of six FIG fields with information on a co-applicant’s or co-borrower’s ethnicity (FIG Data Field Numbers 25-30);
The Race of Applicant or Borrower data field group—comprised of eight FIG fields with information on an applicant’s or borrower’s race (FIG Data Field Numbers 33-40); and
The Race of Co-Applicant or Co-borrower data field group—comprised of eight FIG fields with information on a co-applicant’s or co-borrower’s race (FIG Data Field Numbers 41-48).
The bottom line with procedure #10 is that some ethnicity and race errors may only count as a single error, even if there are multiple FIG field errors. For example, if an applicant indicates “Hispanic or Latino” and also indicates “Mexican” in response to the question of ethnicity, this would be considered two FIG fields, though a financial institution would only have a single error if the information was not correctly listed on the LAR.
The next section of the interagency guidance used to determine an acceptable HMDA error rate relates to prospective changes. Specifically, procedure # 11 states the following:
“Examiners may direct the financial institution to make any appropriate changes in its policies, procedures, audit processes, or other aspects of its compliance management system needed to prevent the reoccurrence of errors identified within the sample that are—absent such changes—capable of repetition, even if the number of errors does not equal or exceed either the Initial Sample Threshold in Column C or the Resubmission Threshold in Column D of the HMDA Table, or even if the errors fall within the tolerances provided in paragraph 9.”
HMDA Transaction Testing Sample Sizes and Thresholds & Examples
The last two sections of the interagency guidance include a table that lists the HMDA transaction testing sample sizes and thresholds, as well as a section that provides several examples of what an acceptable HMDA error rate is and is not.
The HMDA transaction testing sample sizes and thresholds table provides different thresholds based on the LAR count of the filing institution. This table provides guidance for LAR counts up to more than 100,000 entries and also explains that financial institutions that have fewer than 30 LAR entries will have their entire LAR included in the sample.
The interesting thing about both the initial sample and resubmission thresholds is that the tolerance for the total number of errors really does not change significantly as the LAR size increases. For example, all LAR sizes have an initial sample threshold of 2 errors, while the resubmission sample threshold is either 3 or 4 total errors, depending on the size of the financial institutions LAR. The result of this is that the allowable error rate percentage continuously decreases as the LAR count increases. For example, financial institutions that have 100 or less LAR entries essentially have a 10% resubmission threshold, while financial institutions with over 100,000 LAR entries have a resubmission threshold around 2.5%.
As mentioned earlier, the interagency guidance concludes with a number of examples which help to further explain what is an acceptable HMDA error rate for a financial institution's HMDA LAR.
Locating the HMDA Examiner Transaction Testing Guidelines
While the referenced HMDA Examiner Transaction Testing Guidelines is, in fact, interagency guidance - meaning that it applies to all regulatory agencies - each regulator has released the guidance for their respective financial institutions. Therefore, the following links provide the applicable guidance for each federal regulator: