After Closing, a File Can Still Be Corrected — Within Limits
For the borrower, closing is the end of the process. For the loan, it is the point at which the file passes to a series of parties who did not originate it, cannot ask the borrower anything, and are each looking for something different.
The asymmetry that governs everything downstream is this: scrutiny of the file increases sharply after closing, while the cost and difficulty of adding to it rise at the same time.
That is not the same as saying the file is frozen, and the stronger claim — often made, including in an earlier version of this article — is wrong. Fannie Mae's remedies framework provides for correcting and remedying origination defects after delivery, including through additional documentation covering the underwriting period, and post-closing quality control carries its own reverification obligations. [1][2] A defect found after delivery can, in defined circumstances, be cured.
What genuinely changes is the terms on which the file can be added to:
- Additions happen inside a remedy process with its own deadlines and evidentiary standards,
not as ordinary file assembly.
- Later evidence can establish a fact about the underwriting period. It cannot retroactively
establish that the lender held that evidence when it decided. Those are different questions and an investor asks both.
- The original record must be preserved as delivered, with later additions timestamped and
distinguishable. A correction that overwrites history destroys the evidence it was meant to supply.
- Cooperation is no longer assumed. The borrower has moved on and the employer has no continuing
obligation to respond, so obtaining evidence is slower and sometimes unsuccessful. That is a constraint on effort, not a rule against asking.
The chain the loan actually travels
Delivery and purchase. The loan is delivered to an investor or aggregator against a set of representations and warranties — assertions about how the loan was originated, underwritten and documented. The purchase price reflects those assertions being true.
Securitisation. Loans are pooled and the pool is sold. At this point the loan's performance is separated from its origination, and the file becomes the only record of how it was made.
Servicing. Servicing rights may be retained or sold, and may transfer more than once over the loan's life. Each transfer moves the file to a party with no relationship to the origination and no access to anyone who participated in it.
Review. At any point in this chain, an investor may examine the file — routinely, on a sample, or because the loan defaulted early. Early payment default is a common trigger, which means a disproportionate share of file reviews happen on loans already performing badly.
What a repurchase demand actually is
A demand is not a claim that the loan defaulted. It is a claim that a representation about the loan was untrue at the time of sale.
This distinction is the one most often missed. A loan can perform perfectly and still be repurchased for a defect in how it was documented. A loan can default catastrophically with no demand, because nothing was misrepresented — the borrower simply lost their job.
The financial consequence: an estimated average cost of $32,288 per repurchase demand. [1] The loan returns at par while the collateral and borrower position have moved, generally adversely.
The current picture, and a divergence worth noting
The Milliman Mortgage Repurchase Index eased for both GSEs in Q4 2025 — Fannie Mae from 0.181% to 0.173%, and Freddie Mac from 0.277% to 0.260%. [3] Earlier in the year the index had risen for both, from 0.178% to 0.185% and from 0.301% to 0.309% respectively.
Read that carefully, because the index is easily misread. It is a modelled estimate of lifetime repurchase risk on newly acquired loans — not a count of loans actually bought back. A movement in the index is a movement in a projection. It is a useful leading signal and it is not evidence of what happened.
Two things in that data are worth more attention than the direction of travel.
The two GSEs differ by roughly 50%. Freddie's index has run consistently above Fannie's. Milliman sets out several possible explanations for the gap — including differences in acquisition mix and in review practice — without settling on one. An earlier version of this article stated the explanation more definitively than the source supports.
Actual buyback activity has not moved in step. Seller buybacks declined in Q1 2025 alongside an increase in withdrawn claims, and in Q2 repurchases of Fannie loans fell sharply while Freddie buybacks rose. [4] Withdrawn claims matter, but not for the reason usually given. A withdrawal is not proof that no defect existed: demands are also withdrawn because the seller corrected the file, supplied missing documentation, or agreed an alternative resolution. What a withdrawal does show is that the outcome turned on what the seller could produce.
The improving backdrop is at least partly rate-driven. Prevailing 30-year rates fell from 6.55% in Q3 2025 to 6.23% in Q4 and 6.11% by Q1 2026, reducing the share of high-DTI originations. [3] A defect rate improved by market conditions is not the same as an origination process improved by controls, and it reverses when rates do.
Why rebuttal capability is the underrated variable
The rise in withdrawn claims points at something that rarely features in discussions of repurchase risk.
When a demand arrives, the seller's position depends on what can be shown about a file reviewed months or years earlier: which documents were in it, what values were read from them, which requirements were applied, and what the standard was at the time. A defect that never existed is still expensive if it cannot be demonstrated not to have existed.
This is a records problem, and it is the one part of repurchase exposure a lender controls completely after closing. The loan cannot be re-underwritten. The borrower cannot be re-asked. What can be improved is the quality of the evidence retained about the review that was performed.
Three properties determine whether a rebuttal is straightforward or a reconstruction exercise:
Whether the review's coverage is known. If the loan was in a sample, that is a fact worth recording. If it was not reviewed, that is also worth recording, and knowing which is which matters when a demand arrives.
Whether the standard applied can be identified. Requirements change. A rebuttal argues that the file met the standard in force at origination, which requires knowing what that standard was and being able to show it — not reasoning backward from today's rule set.
Whether values are traceable to source documents. A demand alleging an income miscalculation is answered by showing which documents were read, what was read from each, and how the figure was derived. If that derivation exists only as a number in a system, the answer takes weeks.
The practical consequence
Post-closing quality control is often positioned as loss prevention: find defects before the investor does. That is real, and it is the smaller half.
The larger half is that post-closing review is the last opportunity to create a durable record of a file that will be examined by parties who were not there, using standards that will have moved, at a time when nobody involved remembers it. The file can still be corrected after delivery; what cannot be reconstructed later is the reasoning nobody wrote down at the time.
What survives that is not institutional memory. It is whatever was written down, and how specifically it was written down.
Sources
- Fannie Mae Selling Guide, D2-1-04, Identifying and Remedying Origination Defects Under the Remedies Framework. Sets out permitted corrections and the conditions under which additional documentation covering the underwriting period may resolve a defect.
- Fannie Mae Selling Guide, D1-3-03, Lender Post-Closing Quality Control Review — Data Integrity. Post-closing reverification duties, subject to the stated scope and exceptions.
- Milliman Mortgage Repurchase Index, 2025 Q4. The index is a modelled estimate of lifetime repurchase risk on newly acquired loans — not an observed buyback rate. Milliman offers several possible explanations for the Fannie Mae / Freddie Mac difference and does not settle on one.
- Inside Mortgage Finance reporting on seller buyback volumes and withdrawn claims, 2025. A withdrawn demand is not proof that the underlying defect was disproved — corrections, indemnification and other resolutions also end in withdrawal.