mortgagetape^™

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

This article was written in collaboration with Optimal Blue, an Official AI Partner of Mortgage Tape. Loan-level credit and prepay findings are computed on Fannie Mae Single-Family Performance Data and the Freddie Mac Single-Family Loan-Level Dataset. Optimal Blue’s Competitive Data License Plus (CDL+) supplies the rate-regime anchor — the trailing-30-day retail lock median segmented by FICO × current-LTV × occupancy — used to define par and to identify the loans that acquired in-the-money refi incentive after origination.

📌 Quick takeaways by role

• Capital markets & MBS spec-pool desks: Yesterday’s piece landed the PIW credit pay-up (~13% matched-cell DPD outperformance). The offset is the prepay behavior — measured on the same matched-cell framework, PIW purchase loans prepay only 3.5% faster than full-appraisal peers, but PIW refi loans prepay 17% faster. Purchase PIW is a clean pay-up; refi PIW is a two-sided trade whose net depends on pool coupon and OAS regime. Mark two coefficients, not one. Jump to MBS takeaways.
• MSR desks: PIW share is a recapture-conditioned asset — good for the recapture-active shops (Freedom, Rocket, AmeriHome per the Sept 16 BCPR piece) that capture the faster refi flow back into new production, worse for the aggregators absorbing leakage without a rebook motion. Jump to MSR takeaways.
• GSE prepay-model teams & secondary marketing: Standard S-curves don’t carry a PIW dummy; the matched-cell evidence supports adding one, with the coefficient concentrated on the rate-regime that acquired ITM incentive after origination (not the born-ITM cohort). Response-to-incentive is the mechanism, not base attentiveness. Jump to prepay-model takeaways.

Yesterday’s PIW performance piece landed the credit finding: on the 2020–2021 GSE vintages, waived loans hit ever-60+ DPD at ~87% of the full-appraisal rate after matching on the observable-risk stack — a ~13% credit outperformance robust to every cut, translating to roughly 10–15 bps of equivalent risk-cost differential for PIW-heavy pools. The sophisticated reader’s next question is the offset: do PIW loans also prepay faster? If they do, the credit pay-up MBS holders should be under-marking is partly offset by a prepay discount they should also be under-marking. Both marks need to land together for a spec-pool bid to be right.

This piece runs the offset on the same matched-cell framework. Two findings fall out. The answer is different for purchase and refi. And the credit outperformance is not survivorship — a preemptive rejoinder to the natural skeptic who suspects PIW borrowers are churning out fast enough that the “surviving PIW pool” is just the low-risk residual.

Scene 1: Purchase PIW is a clean pay-up

On 2020–2021 purchase originations matched by FICO × LTV × DTI × occupancy, PIW loans prepay 3.5% faster than full-appraisal peers over the first 30 months of aging (CPR-30 ratio 1.035). That’s a small acceleration — well below the 1.10–1.20 range the negative-selection intuition would predict if PIW borrowers were dramatically more attentive. The credit ratio on the same cohort is 0.865 (yesterday’s headline) and 0.831 among survivors — 17% fewer serious delinquencies among the loans that are still active.

The net for purchase is unambiguous. A 17% credit gain against a 3.5% prepay giveback lands with credit dominating on any plausible pool coupon and OAS assumption. Purchase PIW-heavy pools should command a positive net pay-up, and desks running composition-based pay-up regressions should add a purchase-PIW-share regressor with a positive coefficient. There is no coupon at which purchase PIW turns into a discount.

Scene 2: Refi PIW is a two-sided trade

Refi is the load-bearing story. Matched-cell CPR-30 ratio is 1.169 — PIW refi loans prepay 17% faster than full-appraisal refi peers with the same observable-risk profile. The credit ratio on the same cohort is 0.869 (yesterday) and 0.851 among survivors — 15% fewer serious delinquencies.

15% credit gain, 17% prepay giveback. The two effects roughly offset in magnitude but not in duration weight, and the net for the MBS holder depends on the pool coupon relative to par. On a low-coupon pool at 40 bp OAS the credit gain outruns the prepay giveback because slower prepay is what a low-coupon holder wants; on a high-coupon pool the prepay giveback dominates because a high-coupon holder is exposed to prepay optionality on the wrong side. Anchoring to a par rate of ~6.0% (the current 30-year fixed conforming rate node against which “premium” and “discount” are defined here), the crossover — the coupon at which refi PIW turns from a net pay-up into a net discount — sits somewhere in the 6.5–7.0% range on plausible OAS regimes with our measured coefficients. The right answer is not one coefficient. It’s a walkthrough.

The rate-regime cut reveals why refi PIW prepays faster. Splitting the 2020–2021 refi cohort by origination-rate regime — “locked out” (<3.5% note rate, no incentive during 2022–2026), “middle” (3.5–5.5%, acquired incentive as rates fell in 2024–2026), and “born ITM” (≥5.5%) — the PIW/full-appraisal cumulative-prepay ratio moves:

Regime	Purchase	Refi
Locked-out (orig <3.5%)	0.959	1.058
Middle (acquired ITM 2024–26)	1.063	1.204
Born ITM (orig ≥5.5%)	1.089	1.053

The middle-regime refi cell is where the mechanism lives. When a PIW refi loan started at a rate that was fine at origination and acquired in-the-money status as par rates fell, PIW borrowers responded 20% faster than full-appraisal peers. Locked-out loans (no incentive) show a small 1.06 gap that’s essentially base attentiveness; born-ITM loans (incentive present from day one) show 1.05, because both groups know from the start they’ll refi as soon as they can. It’s the acquired incentive — the “rates dropped, should I move?” moment — where PIW borrowers pull ahead. That’s response-to-incentive, not base churn. The mechanism is exactly what you’d expect from a screen that selected on AVM confidence, DU/LP acceptance, and digital-flow comfort in the first place: those borrowers noticed the rate move and acted on it.

Interlude: is the credit outperformance actually survivorship?

The natural skeptic — reading Scene 2 — will ask whether the credit-piece finding was survivorship all along. If PIW borrowers churn faster (Scene 2 says they do), then at any given loan age the surviving PIW pool is disproportionately the low-risk residual, and the credit outperformance we measured yesterday could be a burnout artifact rather than a fundamental screen effect. It’s the right question to ask. The test is to rerun yesterday’s matched-cell DPD comparison on the subset of loans that were still active at loan-age 60 months — no prepay event, no credit event, still on the tape. If survivorship were driving the credit finding, this subset should show a matched ratio approaching 1.0 (the low-risk residual outperforming its own peers by less than the full population did).

Cut	Purchase ratio	Refi ratio
Full population (yesterday’s headline)	0.865	0.869
Survivorship-conditioned (active at 60mo)	0.831	0.851
Post-60mo DPD only (survivors’ subsequent trajectory)	0.792	0.796

The direction is the opposite of a burnout story. Among survivors, the PIW/full-appraisal DPD ratio is lower — 0.83 on purchase and 0.85 on refi — and the post-60mo trajectory is lower still, 0.79 and 0.80. The credit outperformance strengthens the further you get from origination, not the other way around. Burnout is not the mechanism. What the data supports is two independent PIW effects operating on the same borrower: (1) genuinely lower fundamental risk (the AVM-confidence / DU-LP / 80% LTV screen selected on unobservables that keep showing up in the tail), and (2) response-to-incentive prepay behavior (attentive borrowers refi when incentive appears). Both effects survive the matched-cell match. Neither is masking the other.

The practical consequence is that yesterday’s credit coefficient does not need to be discounted for survivorship. The 0.87 stands. And this piece’s prepay coefficients — 1.035 on purchase and 1.169 on refi — are additive, not offsetting through some hidden channel. When you mark a spec pool, you mark both.

The net-effect walkthrough

For a hypothetical 30-year fixed-rate agency spec pool at a 6.5% coupon with a 40 bp OAS, upgrading 30 percentage points of collateral share from full-appraisal to PIW moves the pool’s expected return via two channels. On the credit side, a 13% DPD reduction converts (at the ~20–25% DPD-to-default conversion typical of recent-vintage agency data) to roughly 2–4 bps of CDR reduction over expected life, translating to something in the range of 7–12 bps of equivalent risk-cost pickup. On the prepay side, a 3.5–17% CPR-30 acceleration converts to roughly 0.4–2 CPR-points over the observation window, which on a 6.5% coupon at 40 bp OAS translates to a 1–7 bp expected-return giveback, weighted heavily toward the refi share of the pool.

On a purchase-dominant PIW upgrade, the credit gain runs roughly 7–12 bps and the prepay giveback runs 1–2 bps. Net: +5 to +10 bps positive across every plausible coupon and OAS combination in the grid. Clean pay-up.

On a refi-dominant PIW upgrade the credit gain still runs 7–12 bps but the prepay giveback runs 5–7 bps. Net: 0 to +7 bps positive at coupons below par (5.5–6.5%), collapsing toward zero at 6.75% and crossing into net-discount territory above 7% at tighter OAS. That’s the coupon crossover the walkthrough forces desks to think about. Composition-based pay-up regressions treating refi PIW as coupon-invariant are systematically wrong at both ends of the coupon distribution — over-paying below par, under-charging above.

Operational takeaways: mark two coefficients, condition on coupon, watch the servicer

📊 For capital markets & MBS spec-pool desks: mark two coefficients, not one

The purchase-refi asymmetry is the operational punchline. Purchase PIW-share is a coupon-invariant positive: add a purchase-PIW regressor to the pay-up model with a positive coefficient scaled to ~5–10 bps at 30 pp share differential. Refi PIW-share is coupon-conditional: at low coupons it prints positive, near par it prints roughly zero, above 7% at tight OAS it prints negative. A single “PIW-share” regressor that pools purchase and refi is wrong on both. The 2020–2021 coefficients are the anchor; use them as a prior for 2023–2024 collateral and revisit as the newer vintages age in. The credit outperformance we measured yesterday is not survivorship (this piece’s Interlude confirms), so the credit coefficient does not need a burnout discount.

💼 For MSR desks: PIW share is a recapture-conditioned asset

A PIW-heavy servicing book carries both the credit outperformance (positive for expected servicing life) and the 17% refi-side prepay acceleration (negative for expected servicing life). The two effects are not symmetric across servicers — they interact with recapture discipline. Freedom, Rocket, and AmeriHome (per the Sept 16 BCPR piece running 124–145% of cohort baseline BCPR) capture some fraction of the faster PIW refi flow back into new production, so the prepay giveback partly reappears as new-origination revenue rather than as pure strip loss. Recapture-passive aggregators (Planet Home, Chase, Truist at 72–83% of cohort BCPR) absorb the same flow as leakage. On a servicing acquisition, PIW share should be marked conditional on the acquirer’s recapture behavior, not on a book-average assumption.

⚙️ For GSE prepay-model teams & secondary marketing: add a PIW dummy, weight it toward acquired-ITM regime

Standard S-curves don’t carry a PIW dummy. The matched-cell evidence supports adding one, with a purpose-conditional coefficient (small on purchase, larger on refi) and a rate-regime-conditional weight (largest on loans that acquired in-the-money status after origination — the “middle-regime refi 1.20 ratio” cell). Response-to-incentive is the mechanism, not base attentiveness, so the PIW dummy should activate on the incentive-response term of the S-curve, not on the base hazard. Pricing engines should carry the same two coefficients — credit-positive on purchase, credit-and-prepay-two-sided on refi — with the prepay coefficient wired into the front of the S-curve rather than a flat premium/discount adder.

What this doesn’t measure

Full monthly ITM hazard. The rate-regime cut in Scene 2 uses origination-note-rate regime as a proxy for “did this loan acquire in-the-money incentive?” rather than the fully-specified monthly hazard join against OB CDL+ retail lock median (which was queued as a follow-up to contain scan cost). The full monthly join would sharpen the response-to-incentive attribution but the direction and rough magnitude hold on the proxy — the proxy identifies the middle-regime refi cell cleanly.

Servicer-tier cross-cut. The Sept 16 BCPR piece flagged a 72–145% cohort-normalized tier spread across the top-25 servicers. Whether PIW share compounds or attenuates that tier gap — do recapture-active servicers do even better on PIW pools, or does the PIW attentiveness effect already saturate what recapture would otherwise capture — is a natural follow-up piece rather than a robustness cut here.

Loss severity conditional on default. Same open question as yesterday’s credit piece. Sample supports it, join is deferred, separate follow-up.

Post-February-2025 high-LTV waivers. 6–12 months old; too early for a matched-cell prepay test with any observation depth. The 2020–2021 finding says the eligibility screens picked well on the pre-expansion book; whether the expanded screens produce the same purchase-refi asymmetry is a 2027 question.

Methodology

Loan-level performance and origination data sourced from Fannie Mae Single-Family Performance Data and Freddie Mac Single-Family Loan-Level Dataset, consolidated via Mortgage Tape’s analytics instance. Cohort: 2020–2021 origination vintages for the primary matched-cell results (17.4M loans through both GSEs; 3.7M additional 2022–2024 loans included for the by-vintage robustness cut), first-lien fixed-rate, primary residence and second home. Waiver classification uses the property_valuation_method field on each tape; Fannie “Waiver Only” and “Waiver + Property Data” pooled into a single PIW class; “Not Available” / “Other” / “Unknown” cells dropped. Matching cells: FICO band × origination-LTV band × DTI band × occupancy; cells retained where n ≥ 500 in both PIW and full-appraisal arms; composition-adjusted aggregate ratio computed as the weighted average of within-cell PIW-over-full-appraisal ratios, weighted by full-appraisal cell size. Prepay outcome: voluntary payoff at zero-balance-code 01, cumulative to loan-age 30 months (CPR-30 metric). Rate-regime cut uses origination-note-rate bucket as a proxy for the fully-specified monthly-hazard join against Optimal Blue CDL+ retail lock median (deferred to a follow-up); market rate anchor is the OB CDL+ trailing-30-day retail lock median segmented by FICO × current-LTV × occupancy. Survivorship-conditioned matched-DPD restricts to loans active at loan-age 60 months (no prepay, no credit event); Fannie’s loan_age field caps at 60 for still-active loans so the Fannie side of that cut restricts to 2020 vintage plus 2020–2021 Freddie SLLD, with 1.08M–1.64M survivors per arm. Cash-out refi reported separately in the underlying data and not folded into the primary aggregate. Informational, not advice.

Try it yourself. Ask Mortgage Tape: “what’s the PIW-over-full-appraisal CPR-30 ratio on 2021-vintage refi originations in the 700–739 FICO × 75–80 LTV cell, split by origination-note-rate regime?” — the underlying loan-level data is queryable directly in the platform.