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.

📌 Quick takeaways by role

• Capital markets & MBS investors: After matching on the observable-risk stack (FICO × origination-LTV × DTI × occupancy), 2020–2021 PIW loans hit ever-60+ DPD at ~13% below their full-appraisal peers — 0.865 on purchase, 0.869 on refi. Robust across 79 purchase cells and 109 refi cells. Match on current LTV and the purchase ratio drops to 0.697. PIW-heavy pools carry a defensible pay-up. Jump to capital-markets takeaways.
• GSE credit-risk & policy: The waiver-eligibility screens (AVM confidence, DU/LP gates, the pre-2025 80% LTV cap) are picking, not hiding. Every hypothesis where PIWs would mask latent risk — geographic HPA masking, LTV cushion inflation, forbearance drag — fails on the data. The pre-expansion program earned its calibration; the February 2025 high-LTV expansion inherits a defensible base, subject to its own aging test in 2027–2028. Jump to credit-policy takeaways.
• Secondary-marketing desks & originators: Pricing models that treat PIW and non-PIW loans identically after matching on the observable risk stack are systematically over-charging for PIW files. The gap is small at loan level (~10–15 bps of equivalent risk-cost differential) but structural — a defensible PIW dummy in the pricing engine captures it. Jump to pricing takeaways.

The industry debate on Property Inspection Waivers has run on the same two-sided intuition for a decade. Positive-selection side: the GSEs only grant a waiver on loans their AVM and underwriting systems have already flagged as low-risk, so waived loans should outperform — the pre-screened subset of an already-screened book. Negative-selection side: the appraisal was doing work — catching condition defects, comp-selection games, and outright fraud that no automated model sees — so waiving it leaves a valuation-uncertainty premium that will show up in the tail. Both stories are told at every trade conference. Neither has been adjudicated with public loan-level evidence.

The 2020–2021 vintages are the right cohort to settle it. Waivers reached peak penetration in this window: 65% of Fannie’s 2020–2021 rate-term refi originations and 67% of Freddie’s. Those loans are now aged five to six years, past the sharp early-default window, deep enough to carry a delinquency signal. 17.15 million loans, 6.94 million waived, 10.21 million with a full appraisal. This is the matched-cell test.

The raw gap is huge — most of it is selection

The unmatched comparison points sharply toward outperformance. PIW-flagged purchase loans in the 2020–2021 book hit ever-60+ DPD at 43% of the full-appraisal rate; on refi, at 66%. A two-to-one advantage on purchase would be extraordinary — and any reader with matched-cell instincts will immediately discount it. Waiver eligibility is not random: it requires an AVM confidence score above a threshold, a DU or LP acceptance in the low-risk lane, and (through January 2025) an LTV at or below 80%. Every one of those gates points at the low-risk end of the observable distribution. Of course the waived book performs better in raw comparison — it’s the pre-filtered book.

The real question is what happens after you match: same FICO band, same LTV band, same DTI band, same occupancy. Restricted to cells with at least 500 loans in each arm:

Loan purpose	Cells (n ≥ 500 each)	Ratio ever-60+	Ratio ever-90+	Ratio ever-180+	Raw unmatched 60+
Purchase	79	0.865	0.845	0.830	0.430
Rate-term refi	109	0.869	0.856	0.844	0.656

Two things fall out. Roughly half of the visible PIW outperformance is composition — the gap between the raw ratio (0.43 / 0.66) and the matched ratio (0.87 / 0.87) is exactly what the eligibility screen is doing on the observable-risk stack. But half doesn’t survive matching, and that residual is the signal. PIW loans in the 2020–2021 book still hit ever-60+ DPD at ~87% of the full-appraisal rate after matching — a 13% relative outperformance that widens to 0.83 on ever-180+. Whatever the screen picks up beyond FICO × LTV × DTI × occupancy shows up more in the serious-default tail than in early delinquency, the pattern you’d expect from a screen filtering on borrower and property attributes rather than seasonal cash-flow risk.

The effect is not concentrated in a single corner of the credit surface. Broken out by FICO × origination-LTV cell, PIW-over-full-appraisal ratios sit below 1.0 in every cell with sufficient sample:

FICO band	≤60 LTV	60–70 LTV	70–75 LTV	75–80 LTV
620–659	0.892	0.766	0.804	0.794
660–699	0.901	0.823	0.780	0.809
700–739	0.900	0.818	0.845	0.849
740–779	0.874	0.818	0.796	0.861
780+	0.860	0.848	0.861	0.856

Twenty cells, every one of them below 1.0, cell-level range 0.77–0.90. The pattern is uniform enough that no single FICO × LTV combination is carrying the aggregate — the eligibility screen adds signal across the whole grid, not in one corner.

The purchase-refi symmetry also matters. Negative-selection intuition — the appraisal catches valuation games PIWs miss — would predict a divergence, because purchase transactions carry both a buyer and a seller with incentive around inflated value while refis carry only the borrower. The data shows no divergence. Purchase (0.865) and refi (0.869) sit within 0.4 percentage points of each other. If purchase PIWs were hiding valuation inflation refi PIWs weren’t, purchase ratios would sit above refi ratios. They don’t.

What would refute the finding, and what actually happens

A skeptic has three natural rejoinders. Each has a corresponding cut, and each points the same direction as the headline.

“HPA is masking it.” The 2020–2021 book has ridden 30–65% cumulative HPA on state-level FHFA HPI through Q2 2026. If PIW loans were disproportionately inflated at origination, that inflation could hide behind HPA in strong markets and surface in weak ones. The mechanism would show up as the PIW-to-full-appraisal ratio climbing toward 1.0 in the weakest-HPA states. It doesn’t. Across all 50 states with n_PIW ≥ 2,000, the raw ratio ranges 0.37 (IL) to 0.55 (LA) — every state below 1.0, no inversions. DC at 11% HPA (0.42), Louisiana at 23% (0.55), and Oregon at 31% (0.48) are statistically indistinguishable from Maine, Vermont, and New Jersey at 62–65% HPA. If PIW outperformance were latent-HPA-masking, the scatter would tilt. It’s flat.

“You’re matching on origination LTV — the current-LTV picture would be different.” Rebucketing on current LTV (origination property value grown by state HPI, divided into current UPB) and rerunning the match:

Loan purpose	Ratio ever-60+ (orig LTV)	Ratio ever-60+ (current LTV)
Purchase	0.865	0.697
Rate-term refi	0.869	0.872

Refi is unchanged. Purchase drops to 0.697 — PIW purchase loans outperform their same-current-equity peers by 30%, not 13%. Pre-2025 policy pinned purchase PIWs at ≤80% origination LTV; after 30–60% HPA they cluster in the 50–60% current-LTV band, dominated by high-quality peers on the appraised side. Matching on current LTV compares like-to-like at today’s equity, and the PIW book looks substantially better under that comparison, not worse.

“DPD is a soft signal — show me actual credit events.” Restricting to loans that reached a credit-event zero-balance code (foreclosure sale, short sale, deed-in-lieu, third-party sale):

Loan purpose	PIW event rate	Full-appraisal event rate	Ratio
Purchase	0.033%	0.132%	0.25
Rate-term refi	0.026%	0.047%	0.55
Cash-out refi	0.044%	0.070%	0.63

At the credit-event incidence layer PIWs are 25–63% as likely to hit resolution as full-appraisal peers. Whatever the eligibility screen picks up survives all the way to realized-loss outcomes. Loss severity conditional on default remains the open question — sample is thin at that severity and the join is deferred.

What the eligibility screens are doing

The picture is not “PIWs are safer because they hide risk.” It’s the opposite. The AVM confidence gate encodes property-market density and comp quality the observable-risk stack doesn’t. The DU/LP acceptance lane encodes credit-model borrower quality FICO doesn’t fully capture. The pre-February-2025 80% LTV cap selects against the high-leverage end where valuation manipulation would have highest payoff. Each of those filters operates before the observable-risk stack we match on — so the residual outperformance is what those upstream screens contribute. The waiver adds signal on top of the risk-stack, not a shortcut beneath it.

Operational takeaways: pricing, calibration, and where the pay-up lives

📊 For capital markets & MBS investors: PIW-heavy pools carry a defensible pay-up

The 0.87 matched-cell ratio translates to roughly 10–15 bps of equivalent risk-cost differential — small at loan level, structural at pool level. Composition-based pay-up regressions on TBA vs spec should add a PIW-share regressor for the 2020–2021 collateral bucket; the coefficient is defensible on 17.15M loans. Pre-2022 collateral has small PIW share and is locked from refi optionality regardless. 2023–2024 vintages have PIW share but not enough aging yet — use the 2020–2021 coefficient as a prior and revisit in 2027.

⚙️ For GSE credit-risk & policy leaders: the pre-expansion program earned its calibration

Every hypothesis where the pre-2025 waiver program leaks risk fails on the 2020–2021 book: matched outperformance, purchase-refi symmetry, robustness to LTV definition, geographic uniformity, credit-event incidence reproducing the signal. The eligibility screens (AVM confidence, DU/LP gate, 80% LTV cap) are doing selection work above the observable-risk stack. The February 2025 high-LTV expansion inherits a defensible base — subject to the aging test its own book will support in 2027–2028. That test matters because the expansion stretched two of the three screens (LTV cap and AVM confidence at higher LTVs); DU/LP applies unchanged.

💼 For secondary-marketing & originator pricing: PIW should be a small positive in the risk model

Pricing engines that treat PIW and non-PIW identically after risk-stack match are systematically over-charging PIW files by ~10–15 bps equivalent. Add a PIW dummy in the credit-risk model, calibrated to the 2020–2021 coefficient. Two caveats: on cash-out refi the matched ratio is 0.905 (same direction, smaller margin), so dummy strength should vary by loan purpose; and the finding is on GSE tapes — private-label / bank-portfolio AVM programs are a separate research question.

What this doesn’t measure

Loss severity conditional on default. DPD and credit-event incidence both point to PIW outperformance, but they measure whether a loan defaults, not how much is lost when it does. That analysis is a natural follow-up.

Post-February 2025 high-LTV waivers. Those originations are 6–12 months old, well short of the aging required to see the signal. The 2020–2021 finding says the pre-expansion screens picked well; whether the expanded screens also pick is a 2027–2028 question.

Selection outside the observable-risk stack. Our matching controls for FICO, LTV, DTI, occupancy, and loan purpose. AVM confidence scores and DU/LP model outputs — the screens the GSEs actually use — aren’t on the public tape. The finding is that after matching on what a typical secondary-marketing risk model uses, PIWs still outperform.

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, first-lien fixed-rate, primary residence and second home (investor properties out of scope per the different eligibility rules). 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” (3.8% of Freddie) dropped from the cell math. Matching cells: FICO band (620–659, 660–699, 700–739, 740–779, 780+) × origination-LTV band (≤60, 60–70, 70–75, 75–80) × DTI band (<36, 36–42, 43+) × occupancy. Cells retained only where n ≥ 500 in both the PIW and full-appraisal arm. Composition-adjusted aggregate ratio computed as the weighted average of within-cell PIW-over-full-appraisal DPD ratios, weighted by full-appraisal cell size. State-level HPA computed from FHFA HPI 2021Q1 to 2026Q2. Current-LTV rematch approximates state-level HPA (MSA-level rematch is a follow-up). Credit-event incidence identified via GSE zero-balance codes for foreclosure sale, short sale, deed-in-lieu, and third-party sale. Loss severity conditional on default is not reported in this piece; the sample supports the analysis but the join was deferred. Cash-out refi is reported separately and not folded into the primary aggregate. Informational, not advice.

Try it yourself. Ask Mortgage Tape: “what’s the PIW-to-full-appraisal ever-60+ DPD ratio for 2021-vintage primary-residence rate-term refis in the 700–739 FICO × 75–80 LTV cell?” — the underlying loan-level data is queryable directly in the platform.