The Defect You Already Shipped
A loan that misses its first scheduled payment is the most expensive two words in secondary marketing, because early payment default does not mean the borrower hit a rough patch in year three. It means the loan was wrong at the closing table and nobody caught it. The borrower who cannot make payment one almost never could have made it, and a file that was underwritten to an income that was not real, or a debt that was not disclosed, or an occupancy that was a fiction, is a file that your pre-funding quality control already looked at and passed.
That is the part production leaders skip over. Early payment default is not a servicing problem that starts after the loan boards. It is an origination defect with a delayed alarm, and the alarm goes off in a different building than the one where the mistake was made. By the time the missed payment shows up, the loan has been sold, the reps and warranties are live, and the people who could have stopped it have moved on to the next pipeline. The signal that would have been worth the most, the one that says the decision you made ninety days ago was the wrong one, arrives at the one desk that cannot do anything with it.
Early payoff is the same shape of problem wearing different clothes. A loan that pays off in month three or four did not necessarily do anything wrong, but it cost you anyway, because the premium you were paid on the sale gets recaptured under the investor's early-payoff clause, and the loan that funded, closed, and sold at a gain turns into a loan that funded, closed, and clawed the gain back. Both events share a property that makes them worth building around: they are post-funding facts that indict a pre-funding decision, and in most shops the fact and the decision never get introduced to each other.
Why the Sample Was Always Going to Miss It
The reason early payment default keeps surprising people is arithmetic. Fannie Mae's post-closing quality control framework is built on a selection process that pulls a sample, historically around ten percent of fundings, for a full re-review. A ten percent sample is a reasonable way to estimate a defect rate across a book. It is a terrible way to catch the specific loan that is going to default on payment one, because that loan has a ninety percent chance of not being in the sample at all. The defect that hurts you most is the one your sampling methodology is statistically designed to let through.
We wrote at length about what changes when QC runs at 100% coverage instead of a 10% sample, and early payment default is the cleanest argument for the shift. When every file gets the same pre-funding review, the EPD loan was reviewed too, which means an EPD is no longer evidence that your sample was unlucky. It is evidence that your review missed something on a file it actually read, and that is a far more useful thing to know, because it points at the review rather than at the dice.
The pre-funding QC review is where this has to land, under Fannie's D1-2 prefunding process. The question an EPD forces is not "was this file in our sample" but "which control, applied to this file before the wire, should have flagged what later caused the default, and why did it not." That question only has an answer if the file was reviewed in the first place and if the review's reasoning was recorded in a form you can go back and read.
What an Early Default Is Actually Telling You
Walk back a dozen early payment defaults and they do not scatter randomly across the defect taxonomy. They cluster, and the clusters tell you where your origination process is thin.
Income is the most common. A borrower whose qualifying income was overstated, through a misread of a self-employed cash-flow analysis, a bonus treated as base, a one-time payment annualized, cannot carry the payment the overstatement qualified them for, and the default is the income error becoming visible. Undisclosed debt is the second, where a borrower took on an obligation between application and closing that the file never saw, so the debt-to-income ratio the approval relied on was fiction by the time the loan funded. We built undisclosed-debt monitoring specifically because this is where a quiet pre-closing change turns a sound approval into an EPD.
Occupancy fraud is the third cluster and the ugliest, because an investment property disguised as a primary residence was mispriced, misunderstood, and underwritten to the wrong standard, and the first missed payment is often the moment the fiction collapses. Each of these clusters is a correction to a specific upstream control, and that is the entire value of treating EPD as a feedback signal rather than a loss to be absorbed. An EPD that gets charged to a reserve and forgotten taught you nothing. An EPD that gets traced to the control that should have caught it, and that correction applied to every future file, is the one thing that makes next quarter's EPD rate lower.
Where the Reps and Warranties Actually Bite
The reason this is not merely a cost-accounting exercise is that the loan is sold under representations the lender has to stand behind, and an early default is one of the signals that invites the investor to test those representations. Fannie Mae's loan repurchase and make-whole framework is the governing text, and a loan that defaults early is a loan more likely to be pulled into a review that can end in a repurchase demand, a make-whole payment, or an alternative remedy.
The ground here has shifted in a way worth getting right. The GSEs floated an automatic repurchase trigger tied to early defaults and then backed away from making it automatic, keeping the emphasis on QC review of loans that default early rather than a hard trigger. That is not relief. It means the repurchase conversation is a conversation, and a conversation you win or lose on the quality of your file and the defensibility of your decisions. A lender who can show the investor exactly how a loan was underwritten, which conditions were cleared against which evidence, and why the decision was reasonable on the facts known at closing, is in a different negotiating position than a lender reconstructing a file from memory. The reps and warranties behind every investor sale are only as defensible as the record you kept of the decision.
Closing the Loop, Not Just the File
The operational failure in most shops is not that they ignore EPDs. It is that they resolve them one at a time. An EPD comes in, someone pulls the file, finds the income was overstated, books the loss or fights the repurchase, closes the ticket, and moves on. The file is handled. The loop is open, because nothing changed about the control that let the overstatement through, so the next file with the same pattern defaults the same way, and the one after that.
Closing the loop means the EPD does not terminate at the file. It terminates at the decision. When a loan defaults early, the question we force is not only "what was wrong with this loan" but "what, in how we review loans, let this one through, and what has to change so the next one does not." That is a different kind of work, because it treats the origination process as the thing under review rather than the individual loan, and it only works if the original decision was recorded in enough detail to re-examine.
This is where the architecture matters more than the intent. An agent that holds the loan's history keeps the reasoning that produced each decision, so when an EPD surfaces, we are not guessing at what the underwriter saw. We are reading the actual evidence the decision was made on, the actual income calculation, the actual condition clearances, each cited to the document that satisfied it. The Evidence Trail is what turns an early default from a loss into a lesson, because a lesson requires a record of the decision you are trying to learn from, and most manual processes do not keep one at the grain this needs.
The Early-Payoff Side Nobody Instruments
Early payoff gets less attention than early default because it does not feel like a defect. The loan performed. The borrower refinanced or sold and paid you back. But the investor's early-payoff clause recaptures the premium on a loan that pays off inside the window, usually the first few months, and a lender running thin margins can give back real money to EPOs without ever building a view of where they come from.
The useful question is whether your EPOs are random or patterned, because patterned EPOs are a product and pricing signal. If a cluster of early payoffs traces to a particular lead source, a particular loan officer, or a particular rate-and-term profile, you may be originating loans that were always going to refinance quickly, paying a premium to acquire them, and watching the premium get recaptured on schedule. That is not a compliance problem. It is a margin leak, and it is invisible unless something is watching the payoff events and tying them back to the origination characteristics that predict them. We instrument this the same way we instrument EPD, because the mechanism is identical: a post-funding event that only teaches you something if it gets connected to the pre-funding facts that produced it.
What I Tell Secondary Marketing
The instinct when EPDs and EPOs climb is to tighten everything, add overlays, pull a bigger sample, slow the pipeline down until the numbers improve. That works, and it works by originating fewer loans, which is a cure that looks a lot like the disease. The better move is narrower. Find the specific control that each cluster of early defaults traces back to, fix that control, and leave the rest of the pipeline alone, because the problem was never the whole process. It was two or three thin spots that the sample was never going to find and that nobody routed the evidence back to.
At Sei we build the origination desks so the decision and its evidence survive to be re-examined when the loan's performance comes back to question it, because an early default or an early payoff is only worth what you learn from it, and you can only learn from a decision you kept the record of. The loan that missed payment one was clean when you shipped it. The only thing that keeps the next one from missing payment one too is knowing precisely why this one was not as clean as it looked, and having the proof in hand when the investor asks.
Pranay Shetty
CEO & Co-Founder