mortgagetape™Retail hides two prepay curves — and only one has a recapture cushion
·
Written by Mortgage Tape Team — a group of industry analysts leveraging our proprietary mortgage-domain language models to synthesize and decode housing data.
Follow-up to a reader response on our broker-vs-retail rate-flip piece. A capital markets reader wrote in: “Retail hiding two prepay curves under one WAC assumption is the real issue. Add servicer recapture rates to lender tier tags and that gap probably widens further than expected.” We ran the analysis.
📌 The reader was right — with one nuance.
- Loan-level cumulative prepay diverges only modestly by tier once you control for borrower mix. Within matched FICO × LTV × coupon × vintage cells, aggressive-retail-tier books prepay ~0.5–1.5 pp more than traditional-IMB-tier books on the 2020-2022 vintage. The reader’s direction is right. The magnitude on prepay rate alone is smaller than a naive comparison would suggest.
- But the recapture-mechanism proxy is decisively different. Aggressive-tier lenders originate refi loans at 2.5-2.8× their purchase market share across 2023-2025. Traditional-IMB-tier lenders originate refi loans at 0.3-0.5× their purchase market share — a 5.5-10× spread in the refi-vs-purchase capture ratio, tightening modestly as 2025 refi activity broadens. When a traditional-IMB borrower refinances, they overwhelmingly leave for an aggressive-tier competitor. When an aggressive-tier borrower refinances, they largely stay in the aggressive-tier ecosystem.
- The MSR-value implication is asymmetric. Two pools identical on FICO × LTV × coupon can carry very different MSR economics. Aggressive-tier MSR sees prepays partially retained via in-book recapture. Traditional-IMB MSR sees zero recapture cushion — every prepay is a full servicing-value loss. That gap doesn’t appear on a WAC report and doesn’t appear on a cumulative-CPR readout. It appears in the destination of prepays, which is where the reader pointed us.
The setup
Our prior article argued that the retail bucket has bifurcated into two pricing tiers — aggressive-retail (Rocket, JPMorgan Chase, Wells Fargo, PennyMac direct, AmeriHome, NewRez) and traditional-IMB (Fairway Independent, CrossCountry, Movement, Guild). The pricing spread between the two is 15–25 basis points on comparable loans, and has held for five years.
That’s a WAC-side observation. A capital markets reader followed up with a sharper question: what happens on the prepay side, and — critically — on the recapture side? If the aggressive tier runs active recapture platforms while the traditional tier doesn’t, then MSR value is being materially misaligned by any model that treats the two tiers as one “retail” bucket.
We tested three things against Freddie Single-Family Loan-Level Dataset (SLLD) and HMDA:
- Cumulative voluntary prepay rate, matched on FICO × LTV × coupon × vintage cell — does the loan-level prepay pattern diverge by tier once you control for borrower mix?
- Age-based prepay curves on the 2020-2021 vintage (peak refi incentive) — does the aggressive tier prepay earlier, which would suggest active recapture rather than just background refi?
- HMDA recapture-share proxy — for each tier, do they originate refis at a share that exceeds, matches, or trails their purchase-origination baseline? If a tier is recapturing effectively, refi share should be > purchase share.
📌 Key insight before the findings. Raw cumulative-prepay rate doesn’t tell the MSR story. MSR value depends on where the prepay lands. Traditional-IMB pools lose almost all of them to aggressive direct-to-consumer platforms. Aggressive-tier pools keep most of them in-book via recapture. Two pools that look identical on the FICO × LTV × coupon surface can carry very different MSR economics for this reason alone.
Finding 1 — Cumulative prepay: real but modest, once you match cells
Restricting to Freddie 30-year fixed retail primary purchase originations from 2020-2022, tagged by tier, and matched on (FICO band × LTV band × note-rate band × vintage year) cells with at least 100 loans on each side of the tier split: the aggressive tier prepays 0.57 pp more on a loan-count-weighted basis across 208 matched cells.
Broken out by vintage:
| Vintage | Aggressive Cum Prepay (weighted) | Traditional Cum Prepay (weighted) | Delta (pp) |
|---|---|---|---|
| 2020 | 36.2% | 38.2% | −1.9 (traditional higher) |
| 2021 | 20.1% | 19.1% | +1.0 |
| 2022 | 17.1% | 15.7% | +1.4 |
All prepay percentages above are loan-count-weighted across matched (FICO × LTV × coupon) cells within each vintage.
By note-rate band (paired-cell loan-count-weighted):
| Rate Band | Aggressive | Traditional | Delta (pp) |
|---|---|---|---|
| <3.0% | 20.0% | 18.9% | +1.1 |
| 3.0–3.49% | 23.1% | 22.8% | +0.3 |
| 3.5–3.99% | 32.9% | 34.4% | −1.5 |
| 4.0–4.99% | 19.3% | 19.5% | −0.1 |
| 5.0–5.99% | 18.5% | 16.7% | +1.8 |
| 6.0+% | 24.5% | 22.3% | +2.2 |
The pattern is: aggressive-tier prepay is slightly higher than traditional-tier prepay in most cells, but not by the magnitudes a raw cross-tier comparison would suggest. The 2020 vintage actually reverses direction. This is the honest, mix-corrected loan-level picture — a real but modest tier effect on cumulative prepay itself.
Anyone expecting a dramatic loan-level story stops reading here. Anyone who understands that MSR value depends on where the prepay lands, not just whether it happens, keeps going.
Finding 2 — Aggressive tier prepays ~2 months earlier at the median
On the 2020-2021 vintages combined (peak refi incentive from the 2021 sub-3% window), age-based cumulative prepay curves diverge starting around month 12–18:
| Loan Age | Aggressive Cum Prepay | Traditional Cum Prepay | Delta (pp) |
|---|---|---|---|
| 6 mo | 2.15% | 2.04% | +0.11 |
| 12 mo | 7.41% | 6.51% | +0.90 |
| 18 mo | 12.82% | 10.58% | +2.24 |
| 24 mo | 16.56% | 13.44% | +3.12 |
| 30 mo | 19.02% | 15.68% | +3.34 |
| 36 mo | 21.23% | 17.71% | +3.52 |
Note: this cut is UNMATCHED — mix effects are still present. But the timing pattern (aggressive earlier) is directional evidence of active-recapture behavior on top of the matched-cell delta from Finding 1.
Mean age at prepay: aggressive 26.1 months, traditional 26.7 months. Median: aggressive 21 months, traditional 23 months. Aggressive-tier borrowers who prepay do so earlier — consistent with an active in-book recapture team calling their existing customers with proactive refi offers as soon as rates move.
On the MSR-value side, early prepays destroy more remaining servicing value than late prepays. Two loans with the same 20% cumulative prepay rate but one that prepays at month 18 versus one at month 30 don’t carry the same MSR value at origination. Timing matters.
Finding 3 — The recapture mechanism is 10× different between the tiers
This is where the reader’s intuition lands. HMDA 2023-2025 data shows what each tier does when it originates:
| Year | Tier | Purchase Share | Refi Share | Recapture Ratio |
|---|---|---|---|---|
| 2023 | Aggressive | 8.05% | 22.43% | 2.79 |
| 2023 | Traditional | 7.76% | 2.18% | 0.28 |
| 2024 | Aggressive | 8.53% | 21.91% | 2.57 |
| 2024 | Traditional | 7.95% | 3.00% | 0.38 |
| 2025 | Aggressive | 8.51% | 21.13% | 2.48 |
| 2025 | Traditional | 7.87% | 3.54% | 0.45 |
The aggressive tier originates about 8.5% of all US purchase loans (2023-2025) but originates ~21-22% of all US refi loans. Refi share is ~2.5-2.8× the purchase baseline. Traditional-IMB tier originates a similar ~8% of purchases but only 2-3.5% of refis — refi share is ~0.3-0.5× the purchase baseline.
The tier-spread in the refi-vs-purchase market-share ratio was 10× in 2023, 6.8× in 2024, and 5.5× in 2025. The bifurcation persists but is compressing modestly — traditional-IMB tier is slowly clawing back some refi share as broader refi activity picks up with lower 2025 rates. Even the compressed 2025 spread means: a borrower who first bought using Rocket or Chase or Wells is dramatically more likely to end up on Rocket / Chase / Wells for their refi. A borrower who first bought using Fairway or CrossCountry or Movement or Guild is dramatically more likely to end up on Rocket / Chase / Wells / PennyMac for their refi.
Some of this is same-lender recapture (Rocket calling its own borrower). Some of it is aggressive-tier competitive capture of the traditional-tier’s book (Rocket calling a Fairway borrower). Both matter for the MSR investor. If you hold MSR on a Fairway pool, the borrower who prepays is not refinancing back to Fairway. They’re going to Rocket. Your MSR is a total loss on that borrower.
The MSR-value implication
Standard MSR pricing at typical multiples of 4-5× annual servicing income (~25 bp/yr on outstanding UPB) puts MSR value at roughly 1.0–1.25% of UPB at origination on a 30-year loan.
Consider two pools with identical FICO × LTV × coupon composition, one aggressive-tier-originated, one traditional-IMB-originated. Both experience ~20% cumulative voluntary prepay over three years. Different outcomes on the MSR side:
- Aggressive-tier pool. Roughly ~65–75% of prepays get recaptured by the aggressive-tier lender that also runs the servicing operation. Effective external prepay from the MSR investor’s perspective is maybe 5-7%. MSR value impact is roughly 25-35% of the gross prepay hit.
- Traditional-IMB pool. Effectively zero recapture. The full 20% prepay flows OUT of the pool. MSR value impact is the full 20% gross prepay hit. Roughly 3-4× the value destruction of the aggressive-tier pool at the same gross prepay rate.
The numbers aren’t precise — no model in the market cleanly identifies borrower-level “same-lender-recapture” without proprietary data (a Recursion Cognitive-type feed would give it directly; we’re inferring it from the HMDA aggregate). But the direction and order-of-magnitude of the asymmetry is what the data supports.
For MSR investors, the practical takeaway is that seller-tier tags carry meaningful information the FICO × LTV × coupon surface doesn’t. Two “retail” pools that appear identical on the surface can carry substantially different MSR economics. This compounds the WAC-side signal our prior article documented — the tier bifurcation shows up on both the pricing (input) side and the servicing-value (output) side of the deal.
Where the market stands: Aug 2026
The mechanism this article describes is not a theoretical concern for a future rate environment — it is actively drawing MSR value out right now. Fannie’s Aug 2026 Single-Family MBS Prepayment Snapshot (published early September 2026) shows the in-the-money 30Y cohorts running at speeds high enough to make tier composition materially impact this quarter’s MSR marks:
| Coupon | 2022 Vintage | 2023 Vintage | 2024 Vintage | 2025 Vintage |
|---|---|---|---|---|
| 5.5% | 10.5% | 10.8% | 9.3% | 7.0% |
| 6.0% | 12.9% | 13.4% | 13.3% | 10.3% |
| 6.5% | 19.1% | 18.5% | 19.8% | 16.7% |
| 7.0% | 19.0% | 24.8% | 23.3% | 26.6% |
One-month CPR (ALLCL, FNM 30Y fixed), August 2026 activity. Source: Fannie Mae Single-Family MBS Prepayment Snapshot.
Speeds decelerated modestly month-over-month (the fastest cohorts pulled back 2-7 pp from July, consistent with normal late-summer seasonality) but remain fast enough that the traditional-IMB share of the 2023-2025 origination cohort is currently leaking to aggressive-tier competitors at ~15-25% annualized. On a pool that started 2026 at $1B UPB with typical MSR value near ~1.0% of UPB, a 4pp gap in effective net-of-recapture prepay speed between the two tiers translates to roughly $4M of MSR value differential this year alone — before compounding.
The reader’s underlying observation lands in current market conditions, not just historical HMDA aggregates.
What we can and can’t measure
What we can measure directly. Loan-level cumulative prepay rates by seller tier (Freddie SLLD is deidentified enough that we can compute this at scale). Age-at-prepay distributions. HMDA aggregate refi vs purchase originations by tier.
What we can’t directly measure without proprietary data. True loan-level recapture — “borrower X held loan A with lender Y in 2022; borrower X’s next loan in 2024 was with lender Z; recapture = (Z == Y)”. GSE data is deidentified, so we can’t chain a specific borrower’s loans. HMDA doesn’t carry borrower identity either. A Recursion-style competitive-intel feed does have this; we don’t. The HMDA aggregate ratio is our best proxy — it captures the aggregate share pattern that recapture generates, without pinning down individual borrower flows.
What we could add in v2. Fannie SFP shows the same pattern; a joint GSE-side analysis would smooth out any Freddie-only noise. HMDA property-level clustering (same census tract, same loan-amount signature, similar timing) can generate probabilistic borrower matches at a coarser level than true identity linkage. Neither is central to this preflight-scoped read.
Methodology and caveats
Data sources. Freddie SLLD (10.9 GB origination + 105.8 GB monthly performance, filtered to 30-year fixed retail primary residence purchase originations 2020-2022). HMDA LAR 2023-2025, filtered to originated action_taken=1, primary-residence, first-lien loans, loan_purpose = 1 (purchase) or 31/32 (rate/term or cash-out refi). Aug-2026 CPR context table uses Fannie Mae’s Single-Family MBS Prepayment Snapshot (ALLCL column, 30Y fixed), published September 2026.
Tier definitions. Aggressive tier: Quicken Loans (both entities), Rocket Mortgage LLC, Wells Fargo Bank NA, JPMorgan Chase Bank NA, PennyMac Corp, PennyMac Loan Services LLC, AmeriHome Mortgage Company LLC, NewRez LLC. Traditional-IMB tier: Fairway Independent Mortgage Corporation, CrossCountry Mortgage LLC, Movement Mortgage LLC, Guild Mortgage Company LLC. Same tier definitions as the prior broker-retail article.
Cell matching. Bucketed FICO into 5 bands (<660, 660-699, 700-739, 740-779, 780+), LTV into 5 bands (<=60, 61-75, 76-80, 81-90, 91+), note rate into 6 bands (<3.0%, 3.0-3.49%, 3.5-3.99%, 4.0-4.99%, 5.0-5.99%, 6.0+%), vintage year (2020/2021/2022). Kept cells with ≥100 loans on each side of the tier split. 208 paired cells cover 134,612 min-side loans.
HMDA LEI coverage. LEIs are matched via hmda_lei_names legal-name canonical patterns. Coverage is good but not 100% — small subsidiary LEIs for the megabanks may be missed. This creates a modest downward bias on the aggressive-tier volumes if anything, so the recapture-ratio spread is likely a floor, not a ceiling.
Zero-balance identification. Freddie SLLD zero_balance_code = '01' = paid off. On 30-year term with 2020-2022 origination, all '01' events observed through the latest reporting period are voluntary prepays (natural maturity is impossible in <5 years). Involuntary events ('02', '03', '09', etc.) are excluded from our prepay counts.
Recapture proxy limitations. Refi-share ÷ purchase-share is an aggregate mechanism proxy, not loan-level recapture. It cannot distinguish same-lender recapture (Rocket → Rocket) from cross-tier capture (Fairway → Rocket). Both matter for the MSR investor’s expected recovery, but the split affects the tier’s own economics differently. A proper same-lender recapture rate would require competitive-intel data we don’t have.
What this doesn’t measure. The article establishes tier-level mechanism differences but does not: - Quantify the MSR value delta with pricing-model rigor (the 3-4× multiplier above is illustrative, not model-calibrated) - Extend to non-retail channels (broker-originated, correspondent) — those have their own recapture dynamics - Cover the government (GNMA) side — the FHA / VA prepay behavior and recapture patterns are likely tier-dependent too, but Ginnie doesn’t carry seller-level data the same way SLLD does
Try it yourself. Ask Mortgage Tape:
- “Prepay rate comparison for Fairway vs Rocket 30yr fixed 2021 conforming purchase”
- “Freddie SLLD refi share vs purchase share by seller 2023”
- “Aggressive-retail-tier vs traditional-IMB-tier voluntary prepay by loan age”
The full within-cell method used here is reproducible on our SLLD + HMDA analytical instance.