mortgagetape^™

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

Follow-on to our recapture-mismatch MSR piece. That article measured the tier gap using HMDA aggregates and cumulative SLLD prepay curves — an indirect proxy. Fannie publishes the same signal directly, monthly, at the servicer level, and it confirms the mechanism with cleaner data. This is the readout.

📌 The one-line finding.

Fannie’s Data Dynamics August 2026 BCPR file puts the 30-year fixed cohort baseline at 7.9% one-month CPR. But within that headline, the top-25 servicers spread from 72% to 145% of their own cohort baseline — a ~2x range in cohort-normalized prepay speed on the same product. Freedom Mortgage runs at 145%. Planet Home Lending at 72%. On identical-mix pools, MSR value diverges by hundreds of bps between the two ends of that spread — and Fannie publishes the number every month.

Fannie’s Data Dynamics Framework

Fannie Mae’s Bond Class Prepayment Rate (BCPR) file, part of Data Dynamics, gives one-month annualized voluntary CPR by servicer, by product family, updated monthly. The BCPR_PCT_COH field is the servicer’s speed divided by a matched-cohort baseline (WAC × WALA × FICO × geography). That normalization strips out book-mix effects — it’s not “Rocket’s book is faster because Rocket holds fresher paper” (which would be trivially true). It’s “Rocket’s book is faster than statistically identical books held by other servicers.”

That distinction is load-bearing. Composition-adjusted speed differences aren’t a cohort accident. They’re a servicing-behavior signal.

August 2026 FRM30 all-cohort BCPR: 7.91%. That’s the volume-weighted average speed across every servicer’s book, controlling for cohort mix. It’s the number a Fannie MBS index would report.

The spread hidden underneath it: the top-25 servicers by book size range from 7.83% to 16.29% actual BCPR, and 72% to 145% cohort-relative BCPR. Cohort-normalized. Same product family. Same activity month.

The tier decomposition

Here’s the top 25 FRM30 servicers by August 2026 UPB, ordered by cohort-relative BCPR:

Servicer	Book $bn	1-mo BCPR	vs Cohort
Freedom Mortgage	97	16.3%	145%
Rocket Mortgage	435	14.2%	133%
AmeriHome	81	13.1%	124%
NewRez	171	12.0%	108%
PennyMac Corp	267	12.2%	108%
Wells Fargo	30	10.9%	102%
Lakeview Loan Servicing	232	11.3%	97%
Union Home Mortgage	27	10.5%	97%
LoanDepot	27	10.7%	97%
Fairway Independent	37	10.1%	96%
Onity Mortgage	160	10.4%	95%
Guild Mortgage	66	10.1%	92%
CMG Mortgage	63	9.8%	89%
United Wholesale Mortgage	329	9.8%	89%
CrossCountry Mortgage	145	9.7%	87%
US Bank	62	9.4%	86%
Movement Mortgage	34	9.1%	86%
Guaranteed Rate	55	9.6%	85%
Nexus Nova	71	9.5%	85%
CitiMortgage	39	9.4%	83%
Truist Bank	54	9.0%	83%
JPMorgan Chase	132	9.1%	82%
Citizens Bank	43	9.0%	82%
Onslow Bay Financial	31	9.1%	81%
Planet Home Lending	33	7.8%	72%

Three clear bands emerge on the cohort-relative axis:

Above 120% — recapture engines. Freedom, Rocket, AmeriHome. These are the direct-to-consumer originators that operate on a call-center refi cadence: when a rate move creates a refi opportunity, they call their existing borrower first. The old loan prepays on their book (feeding the high BCPR you see here); the new loan boards on their book with the new coupon. From the borrower’s perspective the relationship is continuous; from the BCPR calculation’s perspective the old loan just prepaid at 145%-of-cohort speed.

Between 95% and 120% — origination-first books. PennyMac, NewRez, Wells Fargo, Fairway, Union Home Mortgage. These are servicers whose books look statistically ordinary relative to their cohorts. They originate; the borrower stays if the borrower stays; when a borrower wants a refi they may or may not go back to the same lender. The book prepays at roughly cohort speed — no measurable recapture premium, no measurable leakage discount.

Below 90% — aggregators absorbing the leakage. Chase, US Bank, Truist, Citizens, Planet Home Lending, Onslow Bay Financial. These portfolios are inherited from thousands of correspondent sellers and warehoused for the long haul. When their borrowers want to refi, they call Rocket or Freedom or their local broker — not the servicing bank on the statement. The refi-eligible borrowers leak out; the surviving book is composed of sticky borrowers (low WAC, refi-hesitant, high friction). Result: cohort-normalized speeds ~15-30% below the cohort baseline.

The counterintuitive point: the sub-100% servicers aren’t “worse” at servicing. They’re the recipients of the leakage the 120%+ servicers create. Same borrower book, different servicing philosophy, opposite BCPR outcomes.

The UMW datapoint

One row worth pausing on: United Wholesale Mortgage at 89% of cohort. UMW is a wholesale lender — it doesn’t originate direct-to-consumer, and it’s not a bank aggregator. Its book comes entirely from independent broker submissions. On the recapture spectrum, wholesale sits in a third position: the broker who submitted the ORIGINAL loan through UMW is the one calling the borrower for a refi, and that broker may or may not submit the refi through UMW again.

The 89% ratio suggests brokers, in aggregate, move borrowers around competitively — some to UMW again, some to Rocket, some to another wholesaler entirely. UWM’s book leaks in the same way an aggregator’s does, just through a different channel. The market has been describing wholesale-vs-retail as if channel were the tier axis; BCPR data shows the axis is actually recapture-active vs recapture-passive, and wholesale sits closer to the passive end than commonly assumed.

One nuance worth stating explicitly: UWM’s 89% measures servicer-book retention on the Fannie pool — whether the loan stays alive on UWM’s servicing tape. It does NOT measure whether the same borrower’s next loan was ultimately submitted through a UWM broker (which may have gone to another wholesaler, or to a retail lender the broker also works with). The 89% is a lower bound on channel loyalty; broker-level attribution would need a competitive-intel data feed BCPR doesn’t provide.

Rocket accelerating, Freedom softening — a month-over-month readout

The August 2026 vs June 2026 shift is worth noting. Speeds decelerated in absolute terms (June was a peak refi month; late-summer seasonality kicked in), but the tier ordering shifted:

Servicer	Jun 2026 cohort%	Aug 2026 cohort%	Δ
Rocket Mortgage	122%	133%	+11
AmeriHome	111%	124%	+13
Freedom Mortgage	158%	145%	−13
PennyMac	114%	108%	−6
NewRez	114%	108%	−6
CitiMortgage	85%	83%	−2
Planet Home Lending	73%	72%	−1

Rocket and AmeriHome are converting more of their refi opportunities than Freedom is right now. The tier ordering flipped: Freedom used to be the market’s dominant recapture machine, and it still is at the absolute level (145% is comfortably the highest), but Rocket and AmeriHome are closing the gap. On a two-month view this could be noise; on the same trajectory for another quarter it’s a meaningful competitive shift among the aggressive tier.

MSR value implication

Two pools that look identical on FICO × LTV × coupon carry substantially different MSR economics depending on who services them. At a 5x multiple on ~25 bps of annual servicing income, MSR value at origination on a 30-year loan is around 1.0-1.25% of UPB. What happens under different tier assignments:

• Freedom-tier pool (145% cohort): Effective prepay rate 45% above cohort. On a $1B book at the current 8% cohort baseline, that’s roughly $12M/year of “extra” prepay-driven servicing loss vs the average book. Over a 3-year MSR horizon, ~$36M of value differential — before compounding.

• Planet-Home-tier pool (72% cohort): Effective prepay rate 28% below cohort. Same $1B book: roughly $7M/year of “saved” servicing revenue vs the average book. Over 3 years, ~$21M of preserved value.

• Total spread across the extremes: ~$57M per $1B UPB over 3 years, purely from servicer tier. That’s roughly 6% of the MSR asset’s own dollar value.

The numbers are illustrative — a proper MSR pricing model would layer in discount curves, servicing cost structures, and prepay-in-lieu-of-default effects — but the directional implication is real. On the same cohort composition, the servicer tag is a stronger predictor of MSR runoff than any single FICO or LTV band difference within that cohort.

For MSR bidders: cohort-relative BCPR is a public, monthly, servicer-level number Fannie has been publishing for years. Using it as a first-order filter on servicer-transfer bids is a free arbitrage.

📌 MSR Desk Rule.

Relying solely on FICO × LTV × WAC leaves money on the table. Servicer-level BCPR is a public, monthly metric that directly exposes underlying runoff risk — and it’s already priced by the desks that use it. Pools passed to a 145%-tier servicer are worth materially less than pools passed to a 72%-tier servicer at the same coupon and credit stack. Servicer identity is a first-order pricing feature, not a secondary adjustment.

What this validates (and what it doesn’t)

Our earlier recapture-mismatch MSR article used HMDA refi-share ratios and cumulative Freddie SLLD prepay curves to argue the same tier bifurcation exists. That analysis was directionally right and dimensionally correct, but indirect — it inferred the recapture mechanism from aggregate origination shares rather than measuring it.

Fannie’s BCPR data measures it directly. The mechanism is now confirmed by two independent data sources (HMDA aggregates + Fannie servicer BCPR), through two different lenses (annual refi-share ratio + monthly cohort-normalized prepay speed), producing the same tier ordering and roughly the same magnitude of spread. That convergence is what makes the argument robust rather than one-source-based.

What this article does NOT do: chain individual borrowers across the tier boundary. BCPR gives servicer-level aggregates, not borrower-level flows. When a Truist loan prepays and the borrower’s next loan is with Rocket, BCPR shows the effect (Truist’s speed sub-100%, Rocket’s speed 133%) but doesn’t confirm the specific handoff. Only proprietary competitive-intel feeds do that. The public data is enough for the tier signal; it isn’t enough for lender-of-record MSR chain attribution.

Methodology and caveats

Data source. Fannie Data Dynamics BCPR file, product_family = FRM30, entity_type = ‘Servicer’, bcpr_term = 1, act_date = 2026-08-01. All figures are volume-weighted across the servicer’s book. Cohort-relative BCPR uses Fannie’s own BCPR_PCT_COH field — a matched-cohort baseline computed by Fannie using WAC, WALA, FICO, and geography. This is public disclosure data, refreshed monthly.

Filter. Restricted to servicers with more than $5B UPB in FRM30 pools for readability. That threshold captures roughly 87% of the FRM30 servicing book by UPB and produces 25 rows. Extending the ranking to smaller servicers preserves the same tier ordering; small-book noise dominates below ~$1B.

What BCPR measures. One-month annualized voluntary CPR (payoffs from refi, sale, or curtailment). Excludes involuntary payoffs (default, repurchase). The value published on the Aug 2026 file reflects loan payoffs observed during the July 2026 payment cycle.

Definition of cohort. Fannie’s BCPR_PCT_COH field normalizes each servicer’s speed by a cohort baseline they compute internally — typically matched on WAC bucket, WALA bucket, FICO bucket, and geography. We use Fannie’s baseline as published; we don’t re-compute it. Any composition drift within our observation window is captured in Fannie’s denominator, not in our numerator.

What this doesn’t measure. Involuntary prepay (defaults, buyouts, foreclosures — a separate 2-4% of runoff for GSE product), curtailment (which is voluntary but not refi-driven), or servicer transfers within the reporting period. All three are handled cleanly by Fannie’s underlying methodology; our aggregations inherit those definitions.

---

Try it yourself. Ask Mortgage Tape:

• “Compare Rocket vs Chase Fannie servicer BCPR by month for 2026”
• “Which servicers have the highest cohort-relative BCPR on FRM30 pools?”
• “MSR value differential between a 145% and a 72% cohort-relative servicer on a $500M book”

The BCPR feed is loaded monthly through August 2026. Freddie’s parallel servicer-level data is also present for cross-agency comparisons.