themortgagellm™Use Case · Capital Markets / Secondary Marketing
Forward-looking pricing and risk decisions for the secondary marketing desk: cell-by-cell LLPA best-execution between Fannie and Freddie, cohort-level P&L attribution (LLPA collected vs realized credit losses), May 2023 LLPA redesign cohort exposure and the implied prepayment risk on existing pools, and pipeline LLPA distribution for rate-sheet calibration. Built on the same vintage-aware LLPA matrix used by the Origination tools — but the questions are at the pool / pipeline / cohort level, not the individual loan.
Sample prompts
Best-execution: where do Fannie and Freddie LLPA grids actually diverge today?
“On a $400,000 conventional purchase loan — FICO 720, LTV 80%, owner-occupied single-family, fixed-rate — do Fannie and Freddie price the upfront LLPA the same today? Show me where the post-May-2023 grids diverge across the broader FICO × LTV cell space.”
For this specific loan profile, Fannie and Freddie price the LLPA IDENTICALLY at 0.875% (87.5 bps = $3,500) — FHFA's aligned-pricing framework eliminates base-grid divergence post-May-2023. Across the full purchase base grid (81 cells), Fannie and Freddie now match cell-for-cell. The remaining divergence is in attribute rows: Freddie's Alt FICO adder (+25 bps for non-standard tri-bureau scoring) has no Fannie equivalent, and Freddie's caps framework differs slightly in labeling though not in net effect…
Cohort attribution: did the 2018-2020 LLPA collected on Fannie purchases earn its risk premium?
“For Fannie Mae purchase originations from 2018-2020, compare the upfront LLPA collected versus the realized credit losses we've actually incurred on that cohort. Was the GSE — and by extension, our desk's prior pricing decisions on these loans — adequately compensated for the credit risk taken?”
Yes — by ~40-75 bps in every year of the 2018-2020 Fannie purchase cohort. Loan-weighted average LLPA was 55-78 bps; estimated realized credit losses (LGD-by-zero-balance-code) were 0.8-2.9 bps. Spread of +55-75 bps over realized cost means the LLPA collected covered ~20-27× the actual losses incurred. Useful as a benchmark for forward LLPA targets and for sizing the structural risk-pricing premium the GSEs actually earn…
MSR-CPR exposure: which Fannie segments got 50+ bps LLPA cuts in May 2023?
“The May 2023 LLPA redesign cut fees for certain borrower segments. Which Fannie purchase loan segments saw the largest LLPA reductions, and how should our desk think about the implied CPR / MSR-pricing risk on pools holding those cohorts when rates eventually fall?”
The largest May 2023 LLPA cuts hit the 620-679 FICO purchase cohort: 640-659 received -91.5 bps relief, 620-639 got -72.7 bps, 660-679 got -71.4 bps. Pools originated pre-May-2023 with concentrated exposure to these bands hold loans paying ~75-90 bps higher in LLPA than the same borrower today — which lowers the break-even rate-drop needed to justify refinancing. Our prepayment models priced under pre-2023 LLPA assumptions likely understate CPR for these cohorts if rates drop materially below current levels…
Pipeline LLPA distribution: 2024 Fannie purchases by FICO band for rate-sheet calibration
“Show the LLPA distribution paid on Fannie purchase loans originated in 2024, broken out by FICO band. Include the mean and p25/p50/p75/p90 within each band and the dollar-weighted average to benchmark against our current rate-sheet assumptions.”
For 2024 Fannie purchase originations (Fannie data, observed in 2025), loan-weighted average LLPA was ~50 bps overall — but the band-level distribution shows the spread that matters for rate-sheet calibration. The 740+ FICO bands cluster at 25-40 bps on the cells they hit; sub-680 bands cluster at 150-200 bps and have wider intra-band dispersion driven by LTV mix. Useful for sanity-checking whether the desk's rate-sheet LLPA inputs match observed acquisition economics…