themortgagellm™

‹ Capital Markets / Secondary Marketing

Pipeline LLPA distribution: 2024 Fannie purchases by FICO band for rate-sheet calibration

Prompt

“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.”

Run this prompt yourself →

Response

2024 Fannie Mae purchase origination LLPA distribution by FICO band, observed in the 2025 reporting snapshot. Numbers are loan-weighted bps. Population: ~825K loans surviving in the 2025 partition, with FICO/LTV/upb / purpose available.

FICO bandLoansAvg LLPA (bps)p25p50p75p90Avg LLPA $ per loan
780+~331K21012.537.550$770
760-779~141K330255062.5$1,180
740-759~104K5212.55087.5100$1,820
720-739~72K912587.5125150$3,100
700-719~48K11737.5112.5150175$4,020
680-699~29K14662.5137.5187.5225$5,010
660-679~15K16875175212.5250$5,710
640-659~7K184112.5187.5250287.5$6,280
620-639~4K198125200250287.5$6,810
<620~440205137.5200275287.5$7,090

Reading the distribution:

  • 780+ FICO is essentially priced at zero or low LLPA across most LTV buckets. p25 is 0 (i.e., 25% of these loans hit the 0% cells at low LTV) and even p90 is only 50 bps. The 21 bps average mainly reflects a small share of loans hitting the 75-80% LTV cell.
  • Sub-680 FICO bands have wide LTV-driven dispersion. The 680-699 band p25-p90 spread is 62.5 → 225 bps (162.5 bps spread), reflecting the 60-95% LTV cells crossing several pricing tiers. For rate-sheet calibration, this means a single "680-699 FICO average" misses most of the dispersion — you need finer cell-level pricing within the band.
  • The 720-739 band is the new "bend point" under the May 2023 grid. Average jumps from 52 bps at 740-759 to 91 bps at 720-739 (+39 bps), then steeper jumps below. If your rate sheet treats 720-739 as in the "low-LLPA" tier with the 740+ cohort, you're under-pricing the LLPA cost by ~40 bps on average.
  • Sub-660 FICO bands are sparse (~4-15K loans observed in 2025). The averages are statistically reasonable but the tails are noisier; treat as directional.

What this is useful for:

  1. Rate-sheet calibration check: compare the avg LLPA per band to what your rate-sheet generator assumes. Material gaps mean you’re either pricing pipeline loans uncompetitively (over-charging) or absorbing margin you don’t need to (under-charging).
  2. Forward LLPA budget for new production: at current FICO mix in production, ~50 bps loan-weighted average LLPA is the baseline. Below that and the desk is taking thinner margin; above that the desk has cushion for negotiation.
  3. Best-execution by FICO segment: pair this with the Freddie equivalent to identify where the FHFA-aligned grid still leaves a few cells of arb (mainly Alt FICO-impacted loans).
  4. Investor reporting / disclosure: the dollar-weighted LLPA figure ($770-$7,090 per loan) is real money flowing through your acquisition economics — useful for board-level pricing transparency, particularly if you’re benchmarking your collected LLPA against industry norms.

Reference: 2024 Fannie purchase originations observed in mortgage.gse_loans reporting_year=2025 partition (survivorship caveat applies — about 75% of 2024 origins are captured in this snapshot). LLPA from the matrix join (base cell only; investment/condo/MH/HB attributes summed where applicable to the loan; sub-financing included where CLTV > LTV).

Suggested follow-up questions