CrowdfundedWealth
Articles · Research note

Mortgage Broker vs Lender: Loans Not Submitted Directly Cost $2,135 More at Closing (2025 HMDA)

By Jorge··21 min read

Some links pay us a referral fee; each one says so. Disclosure

Quick Answer

As of November 15, 2026, in the federal 2025 HMDA data for Texas and Florida, the median originated 30-year conventional home-purchase loan that was not submitted directly to the lender (the channel brokers and other third parties use) cost $8,491 in total loan costs at a 6.625% rate, against $6,356 at 6.5% for loans submitted directly: a gap of $2,135 and 0.125 of a percentage point (our arithmetic). Of the 259,263 loans in our sample, 74,635 (28.8%) came through the not-direct channel. The gap persists at the same combined loan-to-value ratio and loan size ($2,755), but inside one lender it points both ways: at Rocket Mortgage, LLC the not-direct median rate was lower (6.5% vs 6.625%), at JPMorgan Chase Bank, N.A. it was higher (6.5% vs 6.375%). HMDA records the channel, not the broker's fee or the borrower's credit score.

Update alert · free

An email when the mortgage broker vs lender numbers change

When a rate, rule or filing behind this page changes: what changed, the one number that matters, and the source to check it yourself.

Key Takeaways

  • HMDA's Submission of Application field has Code 1 “Submitted directly to your institution” and Code 2 “Not submitted directly to your institution” (2025 Filing Instructions Guide); Code 2 means someone else sent the file, which includes mortgage brokers but is not limited to them.
  • In 259,263 originated Texas and Florida conventional home-purchase loans, median total loan costs were $8,491 not direct and $6,356 direct, and median origination charges were $4,284 and $1,890 (our sample from the HMDA Data Browser).
  • Median interest rates were 6.625% not direct and 6.5% direct, and median rate spreads over the average prime offer rate were 0.257 and 0.065 percentage points.
  • Not-direct borrowers more often had lender credits (42.1% vs 29.1% of loans) and less often paid discount points (54.7% vs 58.2%), so a higher rate with lower points can be a different way of paying, not only a worse deal.
  • Inside one lender the direction varies: the not-direct median rate was lower than the direct one at Rocket Mortgage (6.5% vs 6.625%), Union Home Mortgage (6.125% vs 6.625%) and Paramount (6.625% vs 6.875%), and higher at JPMorgan Chase, PennyMac and NewRez.
  • Denial rates in the sample were 12.32% for not-direct and 10.48% for direct applications; at JPMorgan Chase they were 5.05% and 15.69%. HMDA only sees applications that reached the lender.
  • On a $350,000 loan the median rate difference costs $28.85 a month, so the closing-cost gap plus seven years of payments is $4,559 (our arithmetic); for one borrower the right comparison is two Loan Estimates, not a median.

CSV · 217 rows

Mortgage broker vs lender: 2025 HMDA medians by application channel, Texas and Florida

Rate, total loan costs, origination charges, points, lender credits, CLTV and loan size by application channel, the same comparison inside seven lenders and two lender groups, denial rates, the HMDA field definitions, Regulation Z rules and a worked example.

What HMDA records about the broker channel

Lenders that file under the Home Mortgage Disclosure Act report two facts about how each loan reached them. Regulation C, 12 CFR 1003.4(a)(33), requires, for covered loans other than purchased ones, “Whether the applicant or borrower submitted the application for the covered loan directly to the financial institution” and “Whether the obligation arising from the covered loan was, or in the case of an application, would have been initially payable to the financial institution.” The 2025 Filing Instructions Guide (FIG) turns them into two fields. For Submission of Application it says to “Indicate whether the applicant or borrower submitted the application directly to your institution”, with Code 1 “Submitted directly to your institution” and Code 2 “Not submitted directly to your institution”. For Initially Payable to Your Institution, Code 1 is “Initially payable to your institution” and Code 2 is “Not initially payable to your institution”.

Two things follow. First, the data has no field called “broker”. Code 2 on Submission of Application means someone else sent the file, which in practice includes independent mortgage brokers and can include other third-party originators; we say “not direct” for it and “broker channel” only as shorthand. Second, the rate and cost fields describe the loan at origination. FIG paragraph 4(a)(18) defines origination charges as “the total of all itemized amounts that are designated borrower-paid at or before closing”, and paragraph 4(a)(19) says of discount points “If no points were paid, leave this field blank”, which is why we report the share of loans with points and the median only among those that paid some. The combined loan-to-value ratio (CLTV) is “the ratio of the total amount of debt secured by the property to the value of the property relied on in making the credit decision”. The 2025 modified loan application register went public on March 31, 2026 with about 4,768 filers (CFPB announcement); we pulled our files from the HMDA Data Browser on November 15, 2026.

Our sample: Texas and Florida, one loan type

We downloaded the row-level 2025 file of conventional loans (loan type 1) with actions originated and denied for Texas and then for Florida, one state at a time, streamed it, kept the rows described below and deleted the raw files. The rate and cost comparisons use originated loans that are first-lien, owner-occupied, site-built, one-to-four-unit home purchases with a 360-month term, not reverse mortgages, not open-end lines and not business-purpose, with Submission of Application coded 1 or 2 (rows coded exempt are dropped). That leaves 259,263 loans: 184,628 direct and 74,635 not direct. We chose Texas and Florida because their files are among the largest and the not-direct share differs between them (25.2% in Texas, 32.9% in Florida); they are not the nation.

GroupLoansMedian rateMedian total loan costsMedian origination chargesPaid points: share / medianLender credits: share / medianMedian CLTVMedian loan
Submitted directly (code 1)184,6286.5%$6,356$1,89058.2% / $3,36929.1% / $41080%$345,000
Not submitted directly (code 2)74,6356.625%$8,491$4,28454.7% / $2,42242.1% / $1,00580%$345,000
Texas, direct103,3896.5%$6,591$1,95060.0% / $3,55629.3% / $39080%$345,000
Texas, not direct34,7536.624%$7,951$3,62155.4% / $2,43240.5% / $69680%$335,000
Florida, direct81,2396.5%$6,064$1,79556.0% / $3,11628.9% / $43880%$345,000
Florida, not direct39,8826.625%$9,071$4,97554.0% / $2,40743.5% / $1,40380%$365,000

The not-direct group had a median rate 0.125 of a point higher and median total loan costs $2,135 higher, and the pattern holds in both states. The pieces differ: median origination charges were $2,394 higher, while median discount points paid (among borrowers who paid any) were $948 lower. Lender credits, which offset costs in exchange for a higher rate, appeared on 42.1% of not-direct loans and 29.1% of direct ones. Median CLTV and loan size were the same in the two groups, but the tails were not: 41.8% of not-direct loans had a CLTV of 90% or more against 38.2% of direct loans, and a debt-to-income ratio of 45% or more appeared on 31.5% against 27.7% (our arithmetic on the sample).

Same CLTV, same loan size

Borrowers with less equity pay more for credit, so we also compared a narrow cell: loans with a CLTV between 79.5% and 80.5% (the 20%-down loan) and amounts from $300,000 to $400,000.

GroupLoansMedian rateMedian total loan costsMedian origination chargesPaid points: share / medianLender credits: share / medianMedian CLTVMedian loan
Direct, CLTV 79.5-80.5%, $300,000-$400,00010,6546.49%$6,102$1,95666.8% / $4,07225.3% / $38380%$345,000
Not direct, same CLTV and loan size4,1716.625%$8,857$4,97754.5% / $2,76041.2% / $1,13280%$345,000

In that cell the median rate was 6.625% not direct and 6.49% direct, and the gap in median total loan costs was $2,755: origination charges ran $4,977 against $1,956, and direct borrowers who paid points paid more of them ($4,072 vs $2,760). The cell shows that CLTV and loan size alone do not explain the gap, but it still cannot see credit scores, so it is a market comparison and not a verdict on any lender.

The same lender on both channels

The cleanest test is a lender that takes loans both ways, because its pricing engine and licenses are held constant. Four lenders in the two states had 400 or more sample loans in each channel; we added three with at least 300 in the smaller one.

Lender and channelLoansMedian rateMedian total loan costsMedian origination chargesMedian CLTVMedian loan
Rocket Mortgage, LLC, direct6,8296.625%$7,082$3,74184%$315,000
Rocket Mortgage, LLC, not direct6,3626.5%$6,913$3,55090%$315,000
JPMorgan Chase Bank, N.A., direct5,1486.375%$4,040$50080%$395,000
JPMorgan Chase Bank, N.A., not direct3,0096.5%$6,908$2,94183.636%$275,000
Union Home Mortgage Corp., direct1,0556.625%$7,547$3,38780%$305,000
Union Home Mortgage Corp., not direct6936.125%$6,954$2,28480%$355,000
Paramount Residential Mortgage Group, direct6636.875%$9,822$6,22690%$375,000
Paramount Residential Mortgage Group, not direct1,6566.625%$8,002$3,23485%$395,000
PennyMac Loan Services, direct3126.499%$7,223$2,66880%$380,000
PennyMac Loan Services, not direct3,3766.624%$8,886$3,36880%$485,000
NewRez LLC, direct3096.625%$6,599$2,36880%$335,000
NewRez LLC, not direct1,5996.75%$8,273$3,74280%$345,000
Bank of America, N.A., direct2,5496.125%$4,690$1,53880%$335,000
Bank of America, N.A., not direct3005.125%$1,488$1,01298.39%$265,000

The direction is not consistent. At Rocket Mortgage, LLC the not-direct median rate was 6.5% against 6.625% direct, with similar total loan costs ($6,913 vs $7,082) but a median CLTV of 90% against 84%. At Union Home Mortgage Corp. and Paramount Residential Mortgage Group the not-direct median rate was also lower. At JPMorgan Chase Bank, N.A. it was higher (6.5% vs 6.375%), with total loan costs of $6,908 against $4,040 and origination charges of $2,941 against $500; at PennyMac Loan Services and NewRez LLC the not-direct rate was higher too. Bank of America's not-direct group is only 300 loans with a median CLTV of 98.39%, a different kind of borrower from its direct group, which is why its median rate of 5.125% should not be read as a price. We read one of these lenders in full in our Rocket Mortgage review with HMDA denials and complaints.

Wholesale-only lenders against retail lenders

Many third-party loans go to lenders that take only third-party loans. In our sample, five lenders with 1,000 or more loans had 95% or more of them submitted not directly, and United Wholesale Mortgage alone is 29,481 of the 36,412; thirty-five lenders with 1,000 or more loans had 5% or fewer. The retail group is a mix of banks, national retail lenders and builder-affiliated mortgage companies, not a single product. Nationwide, United Wholesale Mortgage reported 262,457 originated conventional loans of all purposes in 2025 and Rocket Mortgage 330,110 (HMDA aggregations).

Lender or groupLoansMedian rateMedian total loan costsMedian origination chargesMedian CLTVMedian loan
Wholesale-only group (5 lenders)36,4126.624%$8,523$4,36286%$345,000
Retail group (35 lenders)82,2596.375%$6,355$1,92780%$345,000
United Wholesale Mortgage29,4816.624%$8,261$4,20689.52%$335,000
The Loan Store2,1996.625%$8,865$3,74285%$385,000
A&D Mortgage1,9276.75%$11,927$7,39480%$355,000
CrossCountry Mortgage7,2326.875%$8,861$4,29380%$375,000
Fairway Independent Mortgage4,2156.75%$7,471$2,84980%$355,000
loanDepot.com3,1105.625%$6,864$2,66680%$305,000
Guild Mortgage3,0296.625%$7,515$3,61180%$305,000
CMG Mortgage4,7756.625%$7,841$3,47980%$365,000
PrimeLending3,2166.375%$7,877$2,94580%$405,000

The wholesale-only group had a median rate of 6.624% against 6.375% for the retail group and median total loan costs of $8,523 against $6,355, with a median CLTV of 86% against 80%. Individual retail lenders ranged from 5.625% (loanDepot.com) to 6.875% (CrossCountry Mortgage), a wider spread than the average gap between the channels, so the lender you end up with matters at least as much as the door you walk through. Our FHA lenders ranked with HMDA page does the same kind of ranking for government-insured loans.

The second field: initially payable to the lender

The second field splits the not-direct loans again.

Submission / initially payableLoansMedian total loan costsMedian origination chargesMedian rate
Direct, initially payable to the lender (1 and 1)181,746$6,351$1,8906.5%
Not direct, initially payable to the lender (2 and 1)55,595$8,936$4,8766.625%
Not direct, not initially payable to the lender (2 and 2)19,040$7,598$2,9486.624%
Direct, not initially payable to the lender (1 and 2)2,882$6,694$1,8006.624%

Loans not submitted directly and not initially payable to the lender had median origination charges of $2,948 against $4,876 for not-direct loans that were initially payable to the lender. We have only the FIG's code definitions for what sits behind each pattern, so we report the split and do not explain it; the Loan Estimate and closing disclosure show who the creditor is and who is paid what.

Denial rates by channel

Denial rates need applications, not only originations. We used the same filters on originated and denied loans, without the 360-month term, and computed denied divided by originated plus denied; approved-not-accepted, withdrawn and incomplete files are not in this count, so the rates differ from the all-actions definition used on our other lender pages.

GroupOriginatedDeniedDenial rate
Both states, direct184,62821,61210.48%
Both states, not direct74,63510,48612.32%
Texas, direct103,38910,9299.56%
Texas, not direct34,7534,01610.36%
Florida, direct81,23910,68311.62%
Florida, not direct39,8826,47013.96%
Rocket Mortgage, direct6,8291,25715.55%
Rocket Mortgage, not direct6,3621,29416.90%
JPMorgan Chase Bank, direct5,14895815.69%
JPMorgan Chase Bank, not direct3,0091605.05%

Across the two states, 12.32% of not-direct applications in the sample were denied against 10.48% of direct ones. The difference runs the other way at JPMorgan Chase (5.05% not direct vs 15.69% direct) and is small at Rocket Mortgage. United Wholesale Mortgage, which in this sample has only not-direct loans, had a denial rate of 10.69%. The first listed denial reason was debt-to-income ratio on 45.8% of direct denials and 35.3% of not-direct denials. HMDA cannot show what a broker told a borrower before the file went to the lender, so a low lender denial rate on broker files can reflect screening before submission.

How a mortgage broker is paid: Regulation Z

The cost fields above are tied to how originators are paid. Regulation Z, 12 CFR 1026.36(d)(1)(i), as published in the eCFR on November 10, 2026, says that in a consumer credit transaction secured by a dwelling “no loan originator shall receive and no person shall pay to a loan originator, directly or indirectly, compensation in an amount that is based on a term of a transaction”. Paragraph (d)(1)(ii) adds that the amount of credit extended is not a term of a transaction “provided that compensation received by or paid to a loan originator, directly or indirectly, is based on a fixed percentage of the amount of credit extended”, and that such compensation may be subject to a minimum or maximum dollar amount. A broker can be paid a percentage of the loan, but not a percentage that rises with the interest rate.

Paragraph (d)(2)(i)(A) matters for the borrower's choice: if a loan originator receives compensation directly from a consumer, “No loan originator shall receive compensation, directly or indirectly, from any person other than the consumer in connection with the transaction”. In plain terms, on a given loan the broker is paid either by the borrower, which shows up in origination charges, or by the lender, which is built into the price and can mean a higher rate or fewer credits. Paragraph (e)(1) bars steering: a loan originator may not direct a consumer to a transaction because the originator will receive greater compensation from the creditor in it than in others the originator offered or could have offered, “unless the consummated transaction is in the consumer's interest”.

On licensing, paragraph (f) requires a loan originator, when required by law, to be “registered and licensed in accordance with those laws, including the Secure and Fair Enforcement for Mortgage Licensing Act of 2008”. Paragraph (g) requires the loan originator organization to put its name and NMLSR ID, and the name and NMLSR ID of the individual loan originator, on the credit application, the Loan Estimate and closing disclosures, the note and the security instrument. You can look up the broker company, the lender and the individual originator by those numbers in NMLS Consumer Access, which also lists the states where each is licensed.

A worked example on $350,000 (our arithmetic)

The example takes a $350,000 30-year loan at the two median rates and adds the median closing-cost gap, once with all-sample medians and once with the same-CLTV cell.

Line (our arithmetic)All-sample mediansSame CLTV and size cell
Median rate, direct vs not direct6.5% vs 6.625%6.49% vs 6.625%
Monthly principal and interest on $350,000, direct$2,212.24$2,209.94
Monthly principal and interest on $350,000, not direct$2,241.09$2,241.09
Difference per month$28.85$31.15
Difference over 84 months (7 years)$2,423$2,617
Median total loan costs, not direct minus direct$2,135$2,755
Seven-year total, payment gap plus closing-cost gap$4,559$5,372
Payment gap over all 360 payments$10,386$11,215

On all-sample medians the not-direct loan costs $28.85 more a month, which is $2,423 over seven years, and carries $2,135 more in median total loan costs: $4,559 in total (our arithmetic). Median differences are not one borrower's difference. A borrower who takes lender credits for closing costs at a higher rate does not pay the $2,135 gap, and one who pays points to reach a lower rate does not pay the medians either. The exercise shows the shape of the comparison: a channel that charges more at closing and runs a higher rate has to find you a better loan elsewhere to come out ahead.

What HMDA cannot tell you

  • No credit score and no first-time-buyer flag. Both channels' medians mix borrowers; a lower rate can reflect a stronger credit file. We show CLTV and loan size next to rate for that reason, and the within-lender table helps, but neither removes the problem.
  • Rate at origination, not the quote. HMDA has what the borrower closed on, not the rate sheet or the offers a borrower turned down.
  • Broker fees are not a field. Borrower-paid compensation sits inside origination charges; lender-paid compensation is not reported in HMDA and is only visible through rate and lender credits.
  • Applications that never reached the lender. A broker's own turn-downs and pre-qualifications are not in the data, which affects the denial comparison.
  • Two states and one product. Texas and Florida, conventional 30-year home purchases, 2025. FHA, VA, refinancing and other states can differ, and this is not 2026 pricing.
  • Code 2 is not only brokers. The FIG wording is “Not submitted directly to your institution”.

What a reader can do with this

  1. Ask for two Loan Estimates on the same day, one from a lender directly and one from a broker, with the same rate lock period, and compare the origination charges, discount points and lender credits before you compare rates.
  2. Ask the broker how it is paid on this loan, by you or by the lender, in dollars. Regulation Z does not let a broker paid directly by the borrower also take pay from the lender.
  3. Look up the license numbers of the broker company and the individual originator in NMLS Consumer Access before you hand over documents.
  4. Price points the same way. Our mortgage points calculator shows how many months a point takes to pay back, which is the test for any rate-versus-cost trade, broker or not.
  5. Treat the medians as a map. The $2,135 gap in median total loan costs is a distribution, not a quote; the borrowers who came out well in either channel compared more than one offer.

FAQ

Sources, read November 15, 2026: CFPB HMDA Data Browser, 2025 loan-level files for Texas and Florida (conventional loans, originated and denied actions; filtered by our script to first-lien, owner-occupied, site-built, one-to-four-unit home purchases, with originated loans limited to 360-month terms) and 2025 HMDA aggregations by lender; FFIEC 2025 HMDA Filing Instructions Guide (version dated December 2024: data fields 93 and 94 and paragraphs 4(a)(18), 4(a)(19) and 4(a)(24)); CFPB announcement "2025 HMDA Data on Mortgage Lending Now Available" (March 31, 2026); eCFR text of 12 CFR 1003.4 and 12 CFR 1026.36 (version dated November 10, 2026). Medians, shares, denial rates and the worked example are our arithmetic. This is analysis of public documents, not investment, legal or tax advice.

Affiliate Disclosure: Some links are affiliate links. We may earn a commission at no extra cost to you. This does not affect our ratings. Learn more.