themortgagellm^™

Response

Per the Fannie May 2023 grid redesign vs the April 2023 grid, the loan-weighted LLPA shifts on purchase originations by FICO band (Fannie data):

The asymmetric implication for MSR / CPR modeling. For each loan originated PRE-May-2023 in the 620-679 FICO band, the borrower paid 70-90 bps more in upfront LLPA than the post-May-2023 equivalent borrower paid. That LLPA-cost gap directly tightens the break-even rate-drop needed to justify a refinance, because new-loan closing costs (including current LLPAs) are now structurally lower for these segments.

Concrete math. For a 640-659 FICO borrower at 85-90% LTV on a $300K loan:

• Originated April 2023: paid 2.500% LLPA = $7,500 upfront
• Originating May 2023+: pays 1.500% LLPA = $4,500 upfront
• Spread: 100 bps = $3,000 lower closing cost on the new loan
• Translates to roughly 25-35 bps lower break-even rate-drop required to justify a refinance

In aggregate, pools holding this cohort were priced under a structural assumption that the borrower needed (say) a 75 bps rate drop to refi. The new reality is closer to 50 bps for the same economic break-even. The CPR vega on these pools is therefore higher than the original pricing assumed.

What our desk should do:

1. Identify pre-May-2023 exposure: pull pool composition by FICO band x origination year for any spec pool or MSR strip with concentration in 2021-2022-Apr-2023 origination dates and material 620-679 FICO weight.
2. Re-mark CPR assumptions: for the 620-679 FICO portion of those pools, apply an upward adjustment to CPR vega — the rate-down sensitivity should be tighter than the original underwriting assumed. Empirically the data hasn’t shown this yet because the lock-in regime suppresses ALL prepayments, but it will materialize when rates fall.
3. Update forward MSR valuations: the same logic applies to MSR strips — the implied present value of servicing income on LLPA-cut cohorts is lower than the pre-2023 valuation assumed, because the cash flows are likely to shorten faster on the next rate cycle.
4. Consider hedge ratio adjustments: if CPR vega is higher than originally modeled, the hedge ratio (short interest rate exposure to offset prepay sensitivity) should be larger for these specific cohorts.

The honest framing for risk committee. The empirical signal hasn’t materialized yet — current CPR is suppressed by the lock-in regime, so we can’t demonstrate the LLPA-driven CPR step-up in real data. But the structural risk is REAL, and a desk that hasn’t re-thought CPR assumptions on LLPA-cut cohorts will be surprised when rates fall. This is an “adjust the model now while it’s quiet” conversation, not a “wait for the data” conversation.

Reference: Fannie LLPA Matrix vs the May 1, 2023 redesign, base grid cells for FICO 620-779 across all purchase LTV buckets. Comparable Freddie Exhibit 19 cells are byte-identical post-May-2023 (FHFA-aligned framework).