How to Benchmark Your CRA and HMDA Lending Against Peer Data

To benchmark CRA and HMDA lending against peers, pick a group of lenders that compete in the same markets with a similar size and product mix, pull their public HMDA and CRA data for the same year and the same geography, and compare market share, lending by tract income level, denial rates and pricing. The comparison only holds up if the peer group does.

Your own numbers mean little on their own. Say you hold a 6% share of home purchase originations in moderate-income tracts. That could be strong or weak, and nobody can tell which until it sits next to what lenders like you did in the same tracts in the same year. Examiners make that comparison as a matter of course, so it makes sense to make it first. This guide covers where the data comes from, how to build a peer group that will stand up to questions, the comparisons that matter most, and the mistakes that most often make a benchmark misleading.

Why peer comparison matters in CRA and fair lending reviews

In a CRA examination, peer performance is written into the regulation. Under the CRA regulations still in force, the agencies evaluate a bank's performance in light of its performance context, and one of the listed context factors is "the bank's past performance and the performance of similarly situated lenders" (12 CFR 228.21(b)(5), with matching text in the OCC and FDIC rules). Those are the 1995-era regulations. The Federal Reserve and FDIC confirmed in their January 2026 threshold notice that they remain in effect because the 2023 CRA final rule is enjoined.

Fair lending guidance is more guarded, and the wording matters. The Interagency Fair Lending Examination Procedures state that "market share analysis and other comparisons to competitors are insufficient by themselves to prove that an institution engaged in illegal redlining," and that an institution "cannot justify its own failure to market or lend in an area by citing other institutions' failures to lend or market there." The same passage then says that inactivity in an underserved area "where its acknowledged competitors are active would tend to support the interpretation that it intends to avoid doing business in the area." Examiners are told to develop a list of the institution's competitors and learn their level of lending in the suspected redlining area.

In practice that means a peer comparison does not clear you, but a bad one can count against you. Better to find that out from your own analysis than from the examiner's.

Where peer data comes from

Both datasets are public and free. What takes work is putting them into a shape you can compare.

HMDA data

The CFPB publishes HMDA data on the HMDA Data Publication page in several forms, and each one suits a different job:

• Modified Loan/Application Register (LAR). Loan-level data for each institution, "as modified by the Bureau to protect applicant and borrower privacy." It reflects each institution's most current submission and changes when an institution resubmits. The CFPB released the modified LAR files on March 31, 2026, a month after the March 2, 2026 deadline for filing 2025 data.
• Snapshot National Loan-Level Dataset. The national dataset frozen as of a fixed date. It does not update. The FFIEC announced the 2025 snapshot on June 23, 2026, covering data reported by 4,782 institutions, together with the dynamic dataset and the aggregate and disclosure reports.
• Dynamic National Loan-Level Dataset. The same national data, updated weekly to pick up resubmissions.
• Disclosure and MSA/MD aggregate reports. Pre-built tables for each institution (nationwide and by MSA/MD) and for each MSA/MD as a whole.
• The HMDA Data Browser, which lets you filter, aggregate and download the national datasets without handling the full files.

The CFPB itself points out that resubmissions to correct errors typically come in during the weeks after the filing deadline, "for this reason, data users may want to wait for the release of the Snapshot National Loan-Level Dataset to analyze the nationwide data" (modified LAR page). That is the difference between the "preliminary" and "official" peer data you may hear about: the spring release comes from the modified LAR, and the official one follows the snapshot.

CRA data

CRA small business and small farm data comes from banks that are not small banks. They report it annually by March 1 under 12 CFR 228.42(b) (and the parallel OCC and FDIC sections). For each geography the report gives the number and amount of loans in three size bands (at or under $100,000, over $100,000 to $250,000, and over $250,000) and the loans to businesses and farms with gross annual revenues of $1 million or less. For 2026, the Federal Reserve and FDIC set the small bank ceiling at $1.649 billion in assets; the full year-by-year table is on our CRA asset-size thresholds page.

The regulation then turns that reporting into two public products. Each reporting bank gets a CRA Disclosure Statement, which shows its small business and small farm loans in low-, moderate-, middle- and upper-income geographies, county by county (228.42(h)). The agencies also prepare aggregate disclosure statements of small business and small farm lending by all reporting institutions, for each MSA or metropolitan division and the nonmetropolitan part of each state (228.42(i)). The aggregate statements are your market; the disclosure statements are the individual peers.

The most recent CRA release came on November 13, 2025, when the agencies published the 2024 small business, small farm and community development lending data (joint release) along with aggregate reports for each MSA and county. The FFIEC also publishes flat files of census and demographic data that, in its words, "may be combined with the appropriate year(s) of HMDA and CRA data." That file is where tract income levels come from, so it is part of the peer dataset too, not an extra. You can see how past releases landed in our CRA peer data release notes.

Choosing a peer group

A peer group is an argument about who you should be compared with, and an examiner or a community group can argue for a different one. Build it on criteria you can state in one sentence and defend.

Start with geography: lenders active in your assessment areas, counted where you lend. A lender with a large national book and a handful of loans in your county is in your market, but it is not much of a peer. Then narrow by asset size, since similar balance sheets face similar limits. That is the same reasoning behind the "institutional capacity and constraints" factor in 228.21(b)(4).

Product mix matters too. A lender that mostly refinances will look nothing like a purchase lender on home purchase share, so compare like products or at least break results out by loan purpose. Lender type is the last cut. Banks, credit unions and independent mortgage companies use different channels and answer to different regulators. Mixing them can be right for a market view and wrong for a peer view, so decide which question you are asking before you build the list.

It also pays to run more than one group: the one you think is fair, and the one an examiner might pick instead. If you hold up against both, you have a strong answer. If you only hold up against one, you should know that before the exam does.

The comparisons that matter

Market share by tract income level

Tract income levels come straight from the CRA regulation. A low-income geography has a median family income under 50% of the area median, moderate is 50% to under 80%, middle is 80% to under 120%, and upper is 120% or more (12 CFR 228.12(m)). The most useful single comparison is your share of originations in low- and moderate-income tracts against your share of the whole market. If, say, your overall share is 4% and your LMI share is 2%, the gap is the finding, whatever your raw counts look like.

Denial rates and pricing by demographic group

Public HMDA data carries action taken, applicant race, ethnicity and sex, and pricing fields, so you can set your denial-rate ratios and pricing differences by group next to the same ratios for your peers in the same market. Two cautions apply. A peer gap is a reason to look harder at your own files, not a conclusion, and the interagency procedures quoted above say as much. And a raw denial-rate comparison does not control for credit factors, which public HMDA data only partly carries. For what controlling for those factors can and cannot show, see what regression can and cannot prove in fair lending.

Small business lending by tract income

For CRA reporters, the disclosure and aggregate statements give you small business and small farm lending by geography income level, plus the share going to businesses with revenues of $1 million or less. Comparing your distribution to the aggregate for the same MSA or county shows whether your small business lending reaches LMI tracts at the rate the market does.

Common pitfalls

• Mismatched geographies. Under the CRA regulations a geography is a census tract "delineated by the United States Bureau of the Census in the most recent decennial census" (228.12(k)). When tracts are redrawn after a census, a tract number in one year may not cover the same ground in another. Make sure your data and your peers' are geocoded to the same tract set, and that your assessment area is defined the same way on both sides. Drawing the assessment area and shading its tracts on a map, which is what Comply Mapping does, makes a mismatch easier to spot than a table of tract numbers.
• Different reporting years. Your current-year LAR set against last year's market data compares two different markets. The FFIEC publishes its census and demographic files by year, so classify tracts with the file for the same year as the lending data.
• Small cells. A peer with two applications in a tract has a denial rate of 0%, 50% or 100%, and one decision moves it half the scale. Report counts next to every percentage, and do not draw conclusions from cells too thin to bear them.
• Preliminary versus official data. The spring data reflects submissions as they stood, and the snapshot later picks up resubmissions. Benchmarks you will put in front of a board or an examiner should be re-run on the snapshot.
• Comparing the wrong products. HMDA covers dwelling-secured loans, and CRA small business data follows the Call Report definition of loans to small businesses (228.12(v)). Neither is a full picture of your lending, so say which one each comparison uses.

Doing it in Comply Peer-2-Peer

In Comply Peer-2-Peer, the report wizard follows the same order as this guide. It starts by choosing a comparator from a filterable list of every reporting institution for the year, showing respondent ID, agency, application count, assets and loan amount. You can compare a single institution or an institution group, which covers the holding company question (each charter alone, or the group combined). The next step builds the peer group with filters for agency, lender codes, city and state, rate of higher-priced loans, assets and the assets-to-loan-amount ratio. A group is saved as a definition, so next year's run uses the same peers instead of a group rebuilt from memory.

The ranking step offers Top Peer and Closest Peer rankings, by application count, loan amount or assets. Closest Peer shows the institutions just above and below you, which is usually the more useful view. The report families are Individual Comparator, Loan Summaries by Census Tract Characteristics, Ranking Reports, and Peer Summaries of market penetration or market share at the state or MSA/MD level or by any LAR field. They can be batched into one run for a board pack. Because it reads the same Comply database as your filing, your side of the comparison is the data you actually reported.

If you want to try the peer-group step against a market you know, run a peer comparison in a demo.

Sources

• Federal Reserve Regulation BB as in effect March 29, 2024, eCFR, retrieved October 13, 2026: 12 CFR 228.21, Performance tests, standards, and ratings, in general
• Federal Reserve Regulation BB as in effect March 29, 2024, eCFR, retrieved October 13, 2026: 12 CFR 228.42, Data collection, reporting, and disclosure
• Federal Reserve Regulation BB as in effect March 29, 2024, eCFR, retrieved October 13, 2026: 12 CFR 228.12, Definitions
• Federal Reserve Board and FDIC, 91 FR 509, January 7, 2026, retrieved October 13, 2026: Community Reinvestment Act Regulations Asset-Size Thresholds
• FFIEC member agencies, August 2009, retrieved October 13, 2026: Interagency Fair Lending Examination Procedures
• CFPB, retrieved October 13, 2026: HMDA Data Publication
• CFPB, entries of March 31 and June 23, 2026, retrieved October 13, 2026: HMDA News and Updates
• FFIEC Announcement 2026-04, June 23, 2026, retrieved October 13, 2026: FFIEC Publishes 2025 Data on Mortgage Lending
• FDIC, Federal Reserve Board and OCC, November 13, 2025, retrieved October 13, 2026: Federal Bank Regulatory Agencies Release 2024 Small Business, Small Farm, and Community Development Lending Data
• FFIEC, retrieved October 13, 2026: Community Reinvestment Act data products

Photo by Daniel Miksha on Unsplash.