themortgagellm™What the investor LLPA hike actually did
·
Written by The Mortgage LLM Team—a group of industry analysts leveraging our proprietary mortgage-domain language models to synthesize and decode housing data.
📌 Executive takeaways by role
- Capital markets, MBS investors & MSR owners: The 2022–23 investor LLPA hike worked as a credit filter, not just a fee grab. The post-hike agency investor book is measurably better collateral — p10 FICO up 10 points, mean LTV down 2.7pp, state mix rotated from CA/AZ/CO/WA to TX/OH/NC/TN — and in matched FICO/LTV cells the post-cohort experiences 22–25% of the pre-hike early-stress rate. Pool composition improved; MSR strip and spec pool pricing should reflect it. Jump to capital-markets takeaways.
- Secondary marketing & correspondent aggregators: The volume story most of the industry blames on the LLPA is almost entirely rate shock — agency investor originations fell 59% count from Q4 2021 to Q4 2024, and owner-occupied fell 59% over the same window. The LLPA didn’t collapse the channel. It selected. Routing decisions for investor files should be based on which counterparty prices the improved risk correctly, not on volume-loss narratives. Jump to secondary-marketing takeaways.
- Credit risk & product executives: ~4pp of the investor purchase market shifted from agency execution to private-label / bank-portfolio channels between 2021 and 2024. It is a real substitution — but a modest one, and the risk that moved was disproportionately the marginal risk the LLPA was designed to price out. If your portfolio is absorbing that residual, price for it: the borrowers you’re picking up are the ones the agencies decided they didn’t want at the current fee. Jump to risk-takeaway.
Between April 2022 and May 2023, the FHFA and the GSEs roughly doubled the top-of-grid LLPA on investment-property purchase loans. The highest-LTV, mid-credit cell moved from about 175 bps upfront to about 400 bps across two announcements — the April 2022 fee bump (effective May 2022) and the May 2023 aligned pricing framework. Three years later, we have enough loan-level data to answer the question the industry hasn’t answered publicly: what did that actually accomplish?
The short version: the hike worked as a selection filter, not a rent grab, and the volume drop most of the industry attributes to it isn’t the LLPA’s fault. Both findings are visible directly in the loan-level performance data.
The volume story: rate shock, not LLPA
Start with the number everyone reaches for first. Fannie’s investor purchase originations averaged ~9,500 loans a month through Q4 2021, dropped to about 4,950/month by Q4 2022, and settled at ~3,900/month through Q4 2024 — a 59% count decline in three years.
But the LLPA didn’t cause most of that. Owner-occupied purchase originations fell 59% count and 53% UPB over the identical window. The investor-to-owner-occupied ratio held nearly flat: 6.61% in Q4 2021, 6.65% in Q4 2024. Normalize investor volume against the ambient purchase market and the entire raw drop is accounted for by rate shock — the mortgage rate moving from the 3s into the 6s and 7s crushed purchase volume across every occupancy category. The LLPA hike is a second-order effect on top of that, not the driver of the headline decline. Any post-2022 volume narrative that reads the raw agency investor drop as an LLPA verdict is reading rate shock instead.
What the LLPA actually did: the surviving book
Even after the rate-shock volume drop, the composition of what came through changed. The 2018–2021 pre-hike cohort against the 2023–2025 post-hike cohort, cohort-averaged:
| Attribute | Pre 2018-21 | Post 2023-25 | Shift |
|---|---|---|---|
| Mean FICO | 765.9 | 771.1 | +5.2 |
| Median FICO | 775 | 780 | +5 |
| p10 FICO | 711 | 721 | +10 |
| Mean original LTV | 73.5% | 70.8% | −2.7 pp |
| p90 LTV | 80.0% | 80.0% | unchanged |
| Mean DTI | 37.5 | 35.5 | −2.0 |
| p90 DTI | 47 | 48 | +1 |
Read the p10 row — the tenth-percentile FICO, effectively the credit floor of the cohort (90% of borrowers came in above it). The hike’s clearest single fingerprint isn’t a compression at the top of the credit distribution — the p90 LTV is unchanged, the p90 DTI actually ticked up a point — it’s a lifting of the floor. The marginal-credit investor borrower who used to pay 175 bps for agency execution didn’t accept 400 bps. Mean FICO rose 5 points; the tenth-percentile borrower’s FICO rose twice that much. Same story on LTV: mean and p90 tell you less than the fact that lower-quality loans were selectively priced out. The LLPA didn’t punish everyone equally; it filtered the bottom of the applicant pool.
Where the survivors are, geographically
The state mix moved with the credit shift. The 15 largest investor markets by acquisition volume rearranged themselves between the pre and post cohorts in a specific direction:
| Rank ↓, shift | Pre share | Post share | Read |
|---|---|---|---|
| CA | 11.68% | 7.12% | −4.6 pp — the biggest single-state drop |
| AZ | 3.59% | 2.64% | −1.0 pp; falls from rank 5 to rank 11 |
| CO | 3.09% | out of top-15 | Dropped from the leaderboard |
| WA | 2.86% | out of top-15 | Dropped from the leaderboard |
| TX | 10.60% | 13.03% | +2.4 pp — new #1 |
| NC | 4.05% | 5.28% | +1.2 pp |
| OH | 3.00% | 4.23% | +1.2 pp |
| TN | 2.85% | 3.36% | +0.5 pp |
| IN | not top-15 | 2.32% | New to the leaderboard |
The pattern is coherent: out of the volatile-HPI, high-cost, appreciation-play markets (California, Arizona, Colorado, Washington), into the mid-cost, cash-flow-friendly Sun Belt and Midwest (Texas, Ohio, Tennessee, North Carolina, Indiana). The LLPA didn’t just filter borrowers by credit and leverage; the composition shift is also a rotation away from the markets where investor economics depend on appreciation continuing and toward markets where investor economics work on the rent roll. That is exactly the kind of selection a well-calibrated risk fee should produce.
Do those loans actually perform better?
The load-bearing test. In two matched FICO/LTV cells, we compared cumulative 60+ DPD by loan-age 24 months across the pre and post cohorts — apples to apples, aging-fair:
- 720–739 FICO / 70–75 LTV: pre cohort 60+ DPD 4.53% at 24 months, post cohort 1.11%. Ratio 0.246.
- 760+ FICO / 60–70 LTV: pre cohort 1.44%, post cohort 0.32%. Ratio 0.224.
In both cells, the post-hike cohort is experiencing roughly 22–25% of the pre-hike cohort’s early-stress rate. That is well beyond the 0.80 threshold we set as the “selection worked” bar in the preflight, and it is not a marginal signal.
Two honest caveats to name. The pre cohort spans 2018–2021 and picks up the COVID-era forbearance DPD spike; even applying a generous 1.5× discount to the pre baseline, the ratio only widens to roughly 0.33 — still a two-thirds reduction. And only about 40–48% of the post cohort has actually reached 24 months of observation in our latest data snapshot, so the long tail could still develop stress. But the loans already at 24 months are performing dramatically better than any reasonable pre-hike benchmark, and the direction is unambiguous.
Modification rates would be the cleaner stress metric, but the post cohort is currently too young for modifications to have crystallized — the aging-fair mod-rate comparison at 24 months produces post-cohort mod counts of zero and one loan in each cell, statistically meaningless. 60+ DPD is what the data can actually answer today.
The substitution story, sized honestly
If the hike selected for better credit inside the agency book, some of the marginal borrowers who used to pay 175 bps for agency execution had to go somewhere at 400 bps. HMDA tells us where — and how much.
Of all originated investor purchase applications reported to HMDA, the agency share of loan count dropped from 25.9% in 2021 to 21.9% in 2024 — a 4 percentage-point decline. UPB share tells the same story more mildly (down 0.6 pp) because the loans that shifted skew smaller. Meanwhile, the total investor purchase market (all channels) contracted 25% between 2021 and 2024 — nowhere near the 60% agency decline. The gap between those two — the 25% total-market decline and the 60% agency-only decline — is exactly what the substitution to non-agency looks like: a modest but real ~4pp share drift into private-label securitization and bank-portfolio channels.
That is a smaller substitution than the industry-standard risk-shift narrative implies. Real, but not dominant. And critically, the borrowers who moved were disproportionately the marginal-credit, high-LTV, coastal-volatile-HPI profiles the LLPA was designed to price out. If the risk shifted, it shifted into the channels that were willing to take it at the new premium — which is how a pricing signal is supposed to work.
Operational takeaways
📊 For capital markets, MBS investors & MSR owners: pool composition improved
The post-2023 investor purchase book is different collateral. p10 FICO is 10 points higher, mean LTV is 2.7pp lower, and the state concentration rotated out of volatile-HPI markets. In matched cells the early-stress rate is a quarter of the pre-hike baseline. That translates directly to pool pricing: post-hike investor pools deserve tighter spec pay-ups on state concentration than the pre-hike book, MSR strip valuations should reflect the lower expected mod/DPD probability, and any pricing model that carries a fixed “investor risk premium” hardcoded to pre-2022 experience is over-discounting the post-2023 collateral. The pricing dispersion between vintages is bigger than the LLPA differential itself.
⚙️ For secondary marketing & correspondent aggregators: rethink the volume narrative
The dominant industry narrative — that the LLPA hike killed investor lending — is wrong on the facts. Agency investor volume dropped 59% between Q4 2021 and Q4 2024; owner-occupied dropped 59% in the same window. The rate shock did the volume damage. What the LLPA did was select — and the surviving book is better collateral than the pre-hike book was. That reframes the correspondent-aggregator routing question: files that clear the post-hike LLPA are, on the data, meaningfully lower-risk than the same cells looked in 2020. Price them accordingly rather than applying a pre-hike overlay to a post-hike risk profile.
💼 For credit risk & product executives: the residual moved, and it moved marginal
About 4pp of the investor purchase market shifted from agency to non-agency between 2021 and 2024 — into private-label securitization, bank portfolios, and non-QM DSCR channels invisible to loan-level GSE data. The borrowers who moved skew toward the marginal-credit, high-LTV, volatile-HPI profiles the LLPA was designed to price out. If your book is picking that residual up, understand what you are buying: it is the risk the agencies decided they didn’t want at the new fee, not a diversified investor pool. Price for the composition, not for the historical spread.
What we know, and what we don’t
The investor LLPA hike accomplished what it was designed to do: it filtered the agency book toward higher-credit, lower-leverage, more cash-flow-oriented investor borrowers, and the surviving book is performing at a fraction of the pre-hike stress rate. The volume drop most commentary blames on the LLPA is rate shock, not LLPA. The substitution to non-agency channels is real but modest — 4pp of share drift, disproportionately at the margin the fee was meant to move.
What we cannot see: the loans that migrated to bank portfolio and non-QM securitization sit outside GSE loan-level disclosure, so we cannot directly measure whether their performance validates the LLPA’s implied credit-cost differential. The story of what happened to those loans — and whether the non-agency channels are pricing that residual correctly — is the next piece. The 2024 and 2025 vintages will also age into their peak-stress window over the next 24 months, and the current 22–25% ratio may widen or tighten as they do. The next annual data refresh will let us test both.
For now, the answer to did the investor LLPA hike work? is: yes, and mostly not through the mechanism the industry credits or blames it for.
Methodology. Loan-level performance and origination data sourced from Fannie Mae Single-Family Performance Data; Freddie Mac STACR reference pools are included where covered but represent a small share of the investor purchase-loan universe in the available disclosure window. Application-level substitution data from HMDA Snapshot LAR (Federal Financial Institutions Examination Council). All consolidated via The Mortgage LLM’s analytics instance. LLPA fee levels read from the published Fannie Mae purchase-grid LLPA matrices in force at each origination window; investor grid cell fees referenced against the April 6, 2022 and May 1, 2023 FHFA announcements. Investment-property universe defined as owner-occupancy code “I” on the GSE loan-level data and occupancy_type 3 on HMDA. Purchase originations only. Cohorts: pre-hike 2018–2021 origination vintages, mid-hike 2022 vintage (transition, reported separately in tables), post-hike 2023–2025 vintages observed through the most recent disclosure snapshot. Matched-cell performance test compares cumulative 60+ days-past-due at loan age 24 months across cohorts within two FICO × LTV cells; modification rate at 24 months was too sparse in the post cohort (n = 0 to 1 modified) to power a stable ratio and is not reported. Pre-cohort 60+ DPD baseline includes the 2020–2021 COVID-era forbearance spike; a 1.5× discount applied to the pre baseline as a robustness check widens the matched-cell ratio only to about 0.33 without changing the direction of the finding. Approximately 40–48% of the post-hike cohort had reached 24 months of observation at the snapshot date; the long-tail performance of the remaining post-hike loans is still developing. Owner-occupied purchase volume used as a rate-shock control in the volume analysis. Informational, not advice.