mortgagetape^™

Written by Mortgage Tape Team — a group of industry analysts leveraging our proprietary mortgage-domain language models to synthesize and decode housing data.

📌 Takeaways by role

• Secondary-marketing managers: Net lender margin on a 30-year conforming purchase lock ranges from $16,919 (Washington) to $4,392 (Arkansas) — a 3.85× gap between the states where lenders keep the most dollars and the one where they keep the fewest, over Jul 1 – Oct 10, 2025. In basis points of loan amount, the range is 359 bps (Ohio) down to 148 bps (Arkansas). If your pipeline skews toward the thin-margin states, your per-loan contribution is less than half what it is in the thick-margin ones — before any other operational advantage.
• Lock desks: Point-margin and dollar-margin don’t rank the same states. Points are highest in small-loan states (Ohio 3.59, New York 3.32, Pennsylvania 3.09, Indiana 3.06). Dollars are highest in high-cost states where big loans compound through (WA $16.9K, CA $15.8K, MA $14.6K). Four states — Arkansas, Louisiana, Wisconsin, Alabama — sit below $8K per lock, materially under the ~$9–13K per-lock dollar floor that the top-25 origination states cluster around, and warrant per-market review.
• Mortgage CFOs: Colorado is the standout anomaly. Avg loan size is $483K — identical to Utah’s $483K, yet Colorado lenders keep only $9,695 per lock vs Utah’s $13,711. In basis-point terms, Colorado runs at 194 bps of loan amount vs Utah’s 292 bps — a ~$4,000 per-lock, 98-bp shortfall on identically-sized loans. Not loan size, not channel mix, not LLPA burden. Structural competitive compression that deserves its own look.

What’s being measured

Each Optimal Blue CDL+ rate-lock stamps the lock with the lender’s margin (what the lender is pricing above TBA + guarantee fee + servicing) and the LLPA (loan-level price adjustment, which flows to the GSE). Both are expressed in points — hundredths of the loan amount. We also carry the base loan amount, the property state, the product, the loan purpose, and roughly two dozen borrower and loan attributes per lock.

For this analysis we filtered to a clean comparable: 30-year conforming purchase locks, Jul 1 – Oct 10, 2025, with the margin, LLPA, state, and base loan amount fields all populated. That gave 35,956 qualifying locks across 30 states with at least 300 locks each.

Two derived metrics do the work:

• Net lender margin, points = lender_margin. This is what the lender keeps on the lock after passing LLPA through to the GSE.
• Net lender margin, dollars = lender_margin × base_loan_amount ÷ 100.

Both are shown by state below. They rank the states differently, which is the first interesting finding.

Points and dollars don’t rank the same states

In points, the leaders are small-loan states: Ohio at 3.59, New York (surprisingly high given scale) at 3.32, Washington at 3.24, Pennsylvania at 3.09, Indiana at 3.06. The laggards are a mix: Colorado at 1.94, Louisiana at 2.18, Wisconsin at 2.31, Arkansas at 1.48.

Flip the axis to dollars and the ordering changes:

Now Washington leads at $16,919 per lock, followed by California at $15,846 and Massachusetts at $14,575 — all high-cost states where big loans compound even modest point-margins into real dollar outcomes. Ohio, which led the point ranking, falls to the middle of the dollar pack at $9,590 because the average lock size is only $277K.

The dollar numbers are the economic reality. Points are what the lock desk sees; dollars are what the branch runs on and what the shareholder gets. A loan originator in Columbus takes 3.59 points on a $277K loan and clears $9,600. The same originator in Seattle takes 3.24 points on a $533K loan and clears $16,900 — about 1.75× as many dollars for a slightly lower point concession.

This produces an interesting pattern across the top-25 states: points-margin and dollar-margin are nearly uncorrelated at the state level. Lenders appear to adjust points by state to clear a dollar economic floor — more points where loans are small, fewer points where loans are big. The data is consistent with the view that lenders target per-lock contribution, not per-lock percentage.

The anomaly: Colorado

Across the top-25 origination states, there’s a soft “dollar floor” of roughly $9,000 per net lender margin per lock. Below that, you find a handful of states that sit significantly below their peers of similar loan size:

Arkansas, Louisiana, Wisconsin, and Alabama sit below $8,000 per lock on avg loan sizes of $307K / $278K / $333K / $307K — small enough markets that the thin economics are plausibly a mix of local lender concentration and FICO/LTV composition.

Colorado is different. Colorado’s average 30-year conforming purchase loan is $483K — within $5K of Washington’s $533K at the top of the loan-size table and literally identical to Utah’s $483K. Its locks are not small. But Colorado lenders keep only $9,695 per lock, versus $13,711 (Utah), $14,575 (Massachusetts), $15,846 (California), and $16,919 (Washington).

The gap converts cleanly to basis points. Colorado lenders are earning 194 bps on loan amount vs Utah’s 292 bps — a 98-bp shortfall on identically-sized conforming purchase loans. On a $483K Colorado lock, that’s ~$4,000 of margin that neighboring lenders of similar borrower profile are keeping and Colorado lenders aren’t.

Loan size can’t explain it: Colorado’s are big. Points-vs-dollars tradeoff can’t explain it: lenders are keeping fewer points and fewer dollars than UT/MA/CA/WA. LLPA burden doesn’t explain it: Colorado’s average LLPA is –0.43 points (a credit), roughly in line with the national distribution. Retail-vs-broker mix doesn’t explain it either: Colorado is 91.5% retail — lower than most peers but not low enough to drive a 100-bp gap.

What cross-source data says: one signal, two nulls

The honest question is whether we can attribute the Colorado compression to a specific structural mechanism using data outside Optimal Blue. Mortgage Tape has HMDA (every US origination), GSE acquisition tapes (Fannie + Freddie), Ginnie Mae (llmon2 loan-level), and FFIEC Call Reports in one join-ready place. We tested three mechanisms:

Test 1 — Bank-portfolio origination share by state (HMDA). Hypothesis: states with more bank-portfolio originations give borrowers an escape valve from GSE-grid pricing (banks fund from deposits, not TBA, and don’t price to the agency grid), so markets with low portfolio share should compress harder. Result: null. Pearson r = –0.27 across 30 states, and if anything the sign is inverted (thin-margin AR 43%, LA 46%, WI 50% all have higher portfolio shares than thick-margin WA 32%, CA 34%, AZ 24%). Portfolio share is driven by local bank density, which doesn’t map cleanly onto margin.

Test 2 — Servicer concentration by state (GSE pool details + Ginnie llmon2, HHI metric on recent-vintage servicing book). Hypothesis: states dominated by a few national aggregators (Lakeview, Freedom, PennyMac, Mr Cooper) should have thinner margins because those aggregators set a tight exit-trade price expectation. Result: null. Pearson r = +0.19. If anything, thick-margin states (CA HHI 1,174; NY 1,038; MA 1,032) have more concentrated servicing books than thin-margin ones (AR 762, OH 758, MO 764). The exit channel doesn’t predict the primary spread at state aggregation.

Test 3 — GSE borrower note rate by state (Fannie SFP + Freddie SLLD, 2024–25 vintage acquisitions). Hypothesis: if lenders are compressing margin, borrowers in those states should get a thinner rate — the direct symptom of the compression showing up in a different data source. Result: partial. The national correlation is weak (r = +0.16), but Colorado specifically has the 2nd-lowest GSE borrower note rate in the country at 6.67%, below only Utah (6.63%) and above Idaho (6.65%) and Iowa (6.65%). The national mean is 6.78%. That is exactly what compression looks like: Colorado lenders can’t hold rate or points, so the borrower captures both the rate concession and the thin lender margin.

The combined cross-source reading: state-level structural metrics don’t attribute cross-state margin variance in any clean way. Two of three structural lenses are nulls. The one that fires is a symptom (borrower rate), not a cause. The mechanism almost certainly lives below state aggregation — Denver and Boulder specifically, where the Front Range is Zillow’s, Rocket’s, and Veterans United’s natural TAM; where a dense cluster of mid-tier IMBs and active credit unions compete for the same conforming-purchase deal; where consumer rate-shopping intensity is empirically among the highest in the country. State averages wash all of that out.

That’s a legitimate finding, not a cop-out. The cross-source data told us where the answer is NOT (not bank portfolio share, not servicer concentration), which focuses the follow-on question onto MSA / county / lender-specific cuts. That’s exactly what a Primary/Secondary Spread Explorer reference tool with MSA drill-down would answer — on the roadmap.

What this means in practice

For a lender running a national pipeline, the implication is immediate: a basis-point of margin is not a basis-point of margin everywhere. The economic value of winning a Washington lock is roughly 1.76× winning an Ohio lock in dollar terms, and nearly 4× winning an Arkansas lock. Pipeline mix shift is a strategic lever — a 10-point reallocation of volume from Arkansas to Washington is worth roughly $1,250 per lock, before any operational effect, before any pull-through differential.

For MSR traders and capital-markets desks, the implication is that lenders in thin-margin states are structurally more fragile. When rates rise and lock volume falls, the Arkansas or Colorado lender has less margin cushion to absorb the drop. Pull-through pressure, exit-channel aggression, and MSR-release behavior should all skew differently in these states versus the thick-margin peers.

For M&A analysts, the geographic mix of a target’s book is a legitimate normalization input. A $500M-UPB IMB whose pipeline is 60% Colorado and Arkansas is a different animal from a $500M-UPB IMB whose pipeline is 60% Washington and California — and the per-loan P&L is the fastest way to see that difference without diligencing the hedge book.

What we can’t claim

This is one cohort of data on one product (30-year conforming purchase), filtered to a single lock-pricing source (Optimal Blue CDL+). Several caveats are worth flagging:

• Sample bias. OB CDL+ represents roughly 35% of US origination lock volume, skewed toward mid-tier independent mortgage banks. Large-bank retail and some wholesale channels are under-represented. The headline pattern is likely directionally correct across the industry, but the exact bp numbers in each state will differ for lenders outside the OB user base.
• One cohort window. Jul–Oct 2025 was a specific rate environment. Historically, state-level margin dispersion tightens in refi booms and widens in purchase-dominated markets. The 2026 picture may look different.
• Attribution runs out at state aggregation. Three cross-source structural tests above (bank-portfolio share, servicer concentration, GSE borrower rate) produced two nulls and one state-specific signal. The mechanism exists but it operates below state level. The next data pass is MSA / county cuts, which this piece did not run.

How this was built

Every number above was computed on Mortgage Tape’s full-stack data platform. The core margin cut sits on ~20 million Optimal Blue CDL+ lock records (2011–present, loan-level) joined to our vintage-encoded LLPA matrix (every Fannie and Freddie LLPA grid revision 2017–present). The cross-source attribution tests brought in 220 million+ HMDA LAR records (2007–2025 for the portfolio-share cut; 2025 vintage drawn from the CFPB Quarterly LAR release, since the full-year annual LAR won’t publish until March 2026), 100M+ GSE acquisitions (Fannie SFP + Freddie SLLD origination tapes, state × note rate), and the Ginnie llmon2 + GSE pool detail servicing book (state × servicer concentration). Four data sources, four queries, one join-ready data layer, no manual data engineering.

This is the kind of question that cross-source joined data in one place makes answerable in a minute. Any secondary-marketing analyst with the right stack can run it. If you’d like the state-by-state table or an MSA-level cut, sign in and ask the chat directly — the data is already there.

---

Analysis by Mortgage Tape — a housing-finance intelligence platform. Natural-language access to HMDA, GSE, Ginnie Mae, Optimal Blue, FFIEC, NAIC, and non-agency CMBS data, plus 15+ calibrated loan-level scoring models.