mortgagetape™The broker rate advantage didn't flip — it bifurcated. Brokers still beat most retail lenders, but not all.
·
Written by Mortgage Tape Team—a group of industry analysts leveraging our proprietary mortgage-domain language models to synthesize and decode housing data.
📌 Executive takeaways by role
- Loan officers & broker-side AEs: The naive “broker vs all retail” cut says brokers were 9 bp cheaper in 2021 and are 3 bp more expensive now. That framing is misleading. Once you slice retail by lender segment, brokers still beat traditional-retail IMBs (Fairway, CrossCountry, Movement, Guild) by 14-24 bp every single year including 2025 — no flip, no reversal. Where brokers now lose is against aggressive direct-to-consumer retail (Rocket, JPMC, Wells, PennyMac direct) that has repriced through the rate shock. If you compete against a traditional IMB, keep leading with rate. If you compete against Rocket or a megabank, sell service and product access. Jump to LO takeaways.
- Capital markets & MSR desks: The wholesale-vs-retail delta is no longer a channel-average signal — it’s a lender-tier signal. MSR strip pricing that averages across the “retail” bucket is losing signal in a market where retail split into a 15-bp-cheaper aggressive tier and a 15-bp-more-expensive traditional-IMB tier. Value tier-level, not channel-level. Jump to capital-markets takeaways.
- Consumers rate-shopping: Both channels can be right depending on who’s in the room. Against Rocket or a big-bank retail platform, retail may edge broker by a few bp. Against a traditional-IMB retail LO (Fairway, CrossCountry, Movement, Guild), broker is consistently 15-25 bp cheaper. A single retail quote doesn’t tell you the market. Jump to consumer takeaways.
The claim, and how to test it
A broker told one of our analysts recently that broker-originated loans deliver better pricing to consumers than retail-originated loans. The framing is common in wholesale-channel marketing and has been a signature UWM talking point through the last several years. It’s also empirically testable on our data.
Our loan-level data carries an origination-channel indicator on both GSE acquisition data (Fannie SFP + Freddie Single-Family Loan-Level Dataset, ~200M loans since 2018) and Ginnie servicing data (all FHA / VA / USDA / PIH pools, ~100M active loans). Those indicators separately identify Retail (direct-to-consumer), Broker (wholesale channel), and Correspondent (banks and IMBs selling closed loans upstream). For this analysis we focus on the head-to-head Broker vs Retail comparison — the pricing claim is specifically about the two channels a consumer can walk into.
The naive comparison is misleading. Broker-originated loans serve a different borrower profile than retail — historically higher LTV, higher UPB, more concentrated in self-employed borrowers, more concentrated in credit tiers some retail lenders decline. A raw “broker average rate vs retail average rate” comparison confounds channel-pricing with borrower-mix.
The credible comparison is within-cell. We bucket every loan into (FICO band × LTV band × origination month × occupancy × loan purpose), compute the broker-vs-retail rate delta per cell where both channels have at least 30 loans, and loan-count-weight-average across cells. That controls for the borrower-mix confound — we’re comparing broker-originated 720-759 FICO / 75-80% LTV / primary residence purchase in March 2024 to retail-originated 720-759 FICO / 75-80% LTV / primary residence purchase in March 2024, not a broker sub-680 self-employed refi in a low-LTV suburb to a retail 780 FICO primary purchase in the same window.
Scope: 30-year fixed, primary residence, purchase originations, 2020-2025, agency-eligible (GSE) or government-insured (GNMA). Between 500 and 600 paired FICO × LTV × month cells per year on the GSE side (representing ~330,000 unique broker loans and ~1.4 million unique retail loans in 2021, with cell counts stepping down alongside overall origination volume through the rate-shock cycle); 60 to 100 paired cells per year on the GNMA side (representing 80,000 to 130,000 broker loans and 200,000 to 400,000 retail loans per year).
The first cut says the broker advantage flipped
Within-cell average broker-minus-retail rate delta on GSE conforming 30-year fixed purchase originations against ALL retail:
| Vintage | Broker vs ALL retail (bp) | Read |
|---|---|---|
| 2020 | −0.7 | Roughly flat |
| 2021 | −9.2 | Broker meaningfully cheaper |
| 2022 | −5.6 | Broker cheaper |
| 2023 | +1.7 | Slight tilt to retail |
| 2024 | +3.4 | Retail cheaper |
| 2025 YTD | +1.5 | Essentially flat |
The 2021 broker advantage was extraordinarily consistent — in 96% of the 575 paired FICO × LTV × month cells that year, broker-originated loans had a lower average rate than comparable retail-originated loans. By 2023 that advantage was gone and by 2024 it had reversed.
But that’s an aggregate. The moment we published the first version of this analysis, a broker source pushed back with a specific hypothesis — one worth testing carefully because it turned out to reveal a much more important segmentation the aggregate was hiding.
Testing the broker counter-hypothesis: Rocket vs. Traditional IMBs
The pushback: “Very interesting. Can you remove Rocket from retail and see how that impacts? We often price against the Fairways, CrossCountries, etc. of the world and are usually 1+ points cheaper.”
Testing both halves of that critique reveals that removing Rocket barely impacts the aggregate average, while isolating the traditional-IMB comp set flips the narrative entirely — one hypothesis fails empirically, the other holds decisively.
Rocket alone doesn’t move the needle
Within-cell broker-minus-retail delta, retail restricted to the non-Rocket cohort (everyone except Rocket / Quicken variants):
| Vintage | Broker vs ALL retail (bp) | Broker vs retail EX-ROCKET (bp) | Change |
|---|---|---|---|
| 2020 | −0.7 | −0.2 | +0.5 |
| 2021 | −9.2 | −9.3 | −0.1 |
| 2022 | −5.6 | −6.2 | −0.6 |
| 2023 | +1.7 | +1.8 | +0.1 |
| 2024 | +3.4 | +3.4 | 0.0 |
| 2025 | +1.5 | +1.9 | +0.4 |
Removing Rocket from the retail cohort moves the broker-vs-retail delta by 0.1 to 0.6 bp per year — noise. Rocket alone doesn’t explain the aggregate flip. The rest of the retail cohort (JPMC, Wells, PennyMac direct, AmeriHome, NewRez, Chase, hundreds of smaller retail lenders and IMBs, plus the megabank direct channels) is competitive on rate even without Rocket in the pool. The first half of the broker source’s hypothesis is empirically rejected.
But against the specific traditional-retail comp set: dramatically right
Now the second half. Restrict retail to just the four comp lenders the broker source named — Fairway Independent, CrossCountry, Movement, Guild — and rerun the same within-cell delta:
| Vintage | Broker vs ALL retail (bp) | Broker vs TRAD retail comp set (bp) | Difference |
|---|---|---|---|
| 2020 | −0.7 | −0.1 | +0.6 |
| 2021 | −9.2 | −14.8 | −5.6 |
| 2022 | −5.6 | −24.3 | −18.7 |
| 2023 | +1.7 | −19.2 | −20.9 |
| 2024 | +3.4 | −14.2 | −17.6 |
| 2025 | +1.5 | −15.1 | −16.6 |
Every single year including 2022 through 2025, brokers were 14 to 24 basis points cheaper than the traditional-retail comp set on comparable loans. The 2022 gap peaked at 24 bp. There is no reversal in this cut. There is no compression. In fact the broker advantage against traditional retail was widest in 2022-2023 — right in the middle of the rate-shock window when the aggregate “broker vs all retail” story was collapsing.
The broker source’s practical experience is fully consistent with what he’s seeing on his files: when he competes against Fairway or CrossCountry or Movement or Guild, he wins on rate consistently, and the margin has been 15-25 bp for years.
We can’t reproduce his “1+ points” claim in this data set — the largest cell-weighted delta in the GSE conforming purchase universe is 24 bp, not 100+ bp. The bigger number likely applies in specific non-QM or state-restricted or bank-statement configurations that aren’t in the GSE tape. But directionally he’s right at magnitudes that matter.
The aggregate trend masks a structural split: retail bifurcated into two distinct pricing tiers
The “broker vs retail flipped” headline is technically true but analytically shallow. Retail split into two very different pricing tiers between 2022 and 2024, and the split hasn’t closed:
- Aggressive-tier retail — Rocket, JPMC, Wells Fargo (until its 2023 pullback), PennyMac direct, AmeriHome, NewRez, plus a handful of large IMB retail platforms — repriced through the rate shock and now sits at rates below broker on comparable loans. This is the cohort pulling the “all retail” average down through 2023-2025.
- Traditional-retail IMBs — Fairway, CrossCountry, Movement, Guild, and a similar cluster of high-touch, LO-driven retail platforms — held their pricing posture through the rate shock. They price 15-25 bp above broker consistently.
The aggressive tier has grown as a share of retail volume through the down cycle (many traditional-IMB retail platforms contracted harder than the direct-to-consumer / megabank platforms in 2023-2024), which is what makes the “all retail” average look competitive with broker even though most of the lender universe on the retail side isn’t.
For a consumer, this means the answer to “should I use a broker or retail?” depends entirely on which retail lender is in the quote. If it’s Rocket, JPMC, Wells, PennyMac direct — the broker is likely to be within a few bp. If it’s Fairway, CrossCountry, Movement, Guild — the broker is going to be 15-25 bp cheaper.
For loan officers, the practical implication is that the sales pitch has to fit the competitor. The “brokers save consumers” line still works — against most of the retail lender universe by count. The line stops working when the retail competitor is one of the ~10 direct-to-consumer or megabank retail platforms that repriced aggressively.
The FHA / VA story: a much bigger swing
The Ginnie-side comparison — Broker vs Retail on FHA / VA / USDA / PIH 30-year fixed purchase originations — shows a similar flip in the aggregate but with roughly 2× the magnitude on both sides:
| Vintage | Broker vs Retail (bp) | Read |
|---|---|---|
| 2020 | +1.6 | Roughly flat |
| 2021 | −18.9 | Broker much cheaper |
| 2022 | −16.3 | Broker much cheaper |
| 2023 | +2.7 | Slight tilt to retail |
| 2024 | +12.7 | Retail meaningfully cheaper |
| 2025 YTD | +13.5 | Retail advantage widening |
We can’t do the seller-level segmentation on the Ginnie side the way we did on GSE — the Ginnie servicing data is labeled by servicer, not by originator seller — so the aggregate flip is what we can report. But the same underlying dynamic almost certainly applies: within FHA / VA, the aggressive-retail cohort (large IMBs with efficient government-loan execution, plus certain megabank retail channels) has almost certainly repriced hard through the rate-shock era, and that cohort is dragging the “all retail” FHA / VA rate average down. Brokers competing against traditional IMB retail on an FHA file are very likely still winning by wide margins; brokers competing against a Rocket or a PennyMac direct FHA quote are now losing to it.
Why the flip in the aggressive-retail cohort? Three plausible mechanisms
We can’t causally identify from origination data alone what drove the flip in the aggressive-retail tier, but three mechanisms are consistent with the pattern and with what we know about the market during the window.
The wholesale price war peaked and moderated. United Wholesale Mortgage’s “All-In” strategy through 2021 and much of 2022 aggressively subsidized broker-channel pricing to grow wholesale market share. UWM’s own guidance and industry reporting confirms that the aggressive comp posture eased in 2023-2024 as the company transitioned toward margin protection. The 2021-2022 wholesale rate edge partly reflected volume-fight economics that were unsustainable at those magnitudes.
Aggressive retail lenders repriced during the rate shock. The megabank retail channels (Wells Fargo through 2022, JPMC, Chase, Bank of America) and non-bank retail (Rocket, loanDepot, PennyMac retail) got much more competitive on rate through 2023-2024, partly as origination volume collapsed 80% from 2021 peaks and each surviving retail platform fought harder for the remaining files. But this repricing was concentrated in the aggressive tier — the traditional-IMB retail cohort held its pricing posture, which is why the segmentation opened up.
Broker volume collapse itself changed pricing dynamics. GSE broker origination fell from 23 million loans in 2021 to about 2 million in 2024 — an 8-12× contraction, sharper than the retail contraction. Fewer competing wholesale bids on fewer files, combined with wholesale-channel margin pressure from the aggressive 2021-2022 comp regime, meant less pricing aggression on the remaining broker files.
Operational takeaways: pitch, price, and shop
The bifurcation lands differently on each desk it touches. Three reads, by role.
📊 For loan officers on either side of the channel debate: fit the pitch to the competitor
The one-size-fits-all pitch (rate or otherwise) is over. If you’re a broker quoting against a traditional-IMB retail LO, keep leading with rate — the 15-25 bp advantage is real and durable. If you’re a broker quoting against Rocket, a megabank, or one of the aggressive direct-to-consumer platforms, sell on service, close-timing reliability, product access, or LO relationship, not rate. If you’re a retail LO at a traditional IMB, understand that a broker competing against you is holding a real pricing advantage — decide whether to compete on service or push your firm to reprice. If you’re a retail LO at Rocket, JPMC, Wells, or PennyMac direct, rate is now a legitimate selling point on comparable loans; use it.
⚙️ For capital markets & MSR desks: price at the tier level, not the channel level
MSR strip pricing that averages across the retail bucket is now losing signal in a market where retail split into a 15-bp-cheaper aggressive tier and a 15-bp-more-expensive traditional-IMB tier. Different underlying pricing means different WAC assumptions and different prepay behavior — the tier of the underlying lender matters, not just the channel. Refresh the model. Pools dominated by aggressive-retail sellers will exhibit different prepay and refi behavior than pools dominated by traditional-IMB sellers, even when the FICO / LTV composition looks identical on paper.
💼 For consumers rate-shopping: quote both channels and multiple retail lenders
Get quotes from both channels and from lenders across the retail tier. A single retail quote from a traditional-IMB LO doesn’t tell you the market — you may be leaving 15-25 bp on the table by not also checking Rocket or a broker. A single retail quote from Rocket doesn’t tell you the market either — a broker may match or beat it on many files. The right shopping-list is at least one broker, at least one aggressive-tier retail lender (Rocket / a megabank / PennyMac direct), and at least one traditional-IMB retail LO. Three quotes, three tiers, one decision.
Methodology and caveats
Data sources. GSE side: Fannie Mae Single-Family Performance Data + full Freddie Mac Single-Family Loan-Level Dataset (deduplicated against Freddie’s STACR reference-pool disclosures to avoid double-count), consolidated via Mortgage Tape’s analytics instance; universe restricted to 30-year fixed-rate primary-residence purchase originations delivered through broker or retail channels; 94–96 paired FICO × LTV × month cells per year. Ginnie side: Ginnie Mae loan-level servicing disclosure (the standard monthly snapshot as of June 2026, covering FHA / VA / USDA / PIH pools), same universe restrictions applied via Ginnie’s origination-channel indicator for broker vs retail delivery; 60–100 paired cells per year.
Controls. Every reported delta compares broker vs retail (or a retail sub-cohort) within the same (FICO band × LTV band × origination month) cell, with both sides required to have at least 30 loans in the cell. This handles the borrower-mix confound the naive comparison implicitly relies on. It does not control for property-type or state, which we’ll add in a v2 if the demand exists — the FICO × LTV × time control is the standard analyst cut and captures the dominant sources of borrower-profile variation.
Traditional-retail comp set. Restricted to the four lenders the broker source named: Fairway Independent Mortgage Corporation, CrossCountry Mortgage LLC, Movement Mortgage LLC, and Guild Mortgage Company. Total unique loans in this cohort across 2020–2025: ~482,000 (30-year fixed, primary-residence purchase originations retained by the filter). Canonical seller-name rollups were used to catch minor name variants across the reporting record. Adding additional traditional-IMB names (loanDepot retail, Homebridge, Guaranteed Rate) is plausible but changes the analysis less than one might expect — the four-lender cluster is already a strong signal.
Rocket / Quicken detection. The Rocket Mortgage canonical seller name and legacy Quicken Loans variants were folded together as “Rocket” for the ex-Rocket cut.
GNMA snapshot bias. GNMA rates for older vintages (2020-2022) are drawn from the June 2026 servicing snapshot, which means fast-prepaying loans have dropped out of the active pool. That survivorship affects both channels comparably, so the broker-vs-retail delta is still valid, but the absolute rate levels for early vintages read slightly higher than they were at origination.
What this does not measure. Rate is not the full cost of a loan. Discount points, lender credits, origination fees, and any borrower-paid compensation on the broker side all move the true APR-equivalent cost. GSE loan-level data doesn’t carry borrower-cost detail; HMDA does (from 2018 onward), and a follow-up piece using HMDA’s borrower-cost fields for total closing costs, discount points, and lender credits would let us test whether the rate-based flip is partly an accounting artifact of channels pricing differently across rate versus points. Our prior is that it isn’t — the 15–25 bp broker-vs-traditional-IMB gap is large enough to survive normal points variance — but the check is worth running. Capital-markets and compliance readers should read every basis-point figure in this piece with that boundary in mind.
We could not reproduce the “1+ points” magnitude. The broker source referenced a 100+ bp broker rate advantage in his practical experience. The largest cell-weighted delta in our GSE conforming purchase universe against traditional retail is 24 bp. That doesn’t mean the broker source is wrong — the “1+ points” gap likely applies in specific non-QM, state-restricted, or bank-statement configurations that aren’t in the GSE tape, or in specific rate-sheet snapshots on a given day rather than as vintage-averaged behavior. Our analysis is bounded to what agency-eligible loan-level data can prove.
Non-QM excluded. Both loan-level surfaces here cover only agency-eligible originations (GSE conforming + Ginnie government). Non-QM (DSCR, bank-statement, alt-doc) is a broker-heavy segment where the broker channel likely still delivers meaningful savings by virtue of product access alone (many retail lenders don’t offer these programs at all). This analysis does not cover non-QM and shouldn’t be extrapolated to it.
Revision history. v2 (2026-08-20): added the retail-segmentation analysis after a broker source pushed back on the aggregate v1 finding. The first-cut chart is unchanged; the new chart in the middle shows the three-way retail cut that reveals the segmentation.
Informational, not advice.
Try this yourself. On our chat, ask:
- “Broker vs retail rate delta on 30yr FHA purchase originations in 2024”
- “Broker vs Fairway/CrossCountry rate delta on GSE conforming purchase 2023”
- “Which retail lenders have the lowest average rate on 30yr conforming purchase in 2024?”
The full within-cell comparison method is reproducible on our loan-level GSE and Ginnie data in the platform.