The question
In Current, then gone, loans that got an extension went on to charge off at 2.4 times the rate of loans that never did, odds ratio 2.6 after controls. The raw gap on these three shelves is close to double.
| Lender | Share of loans ever extended | Charge-off rate, ever extended | Charge-off rate, never extended |
|---|---|---|---|
| Exeter (EART) | 40% | 40.5% | 20.0% |
| AmeriCredit (AMCAR) | 36% | 23.6% | 12.8% |
| Santander Drive (SDART) | 18% | 23.9% | 13.7% |
The easy read is that lenders extend to hide bad loans. The extension resets the delinquency clock, so a loan that is falling behind shows up as current again.
The harder question is whether it also pays. If an extension buys more cash than a repossession would, it is more than cosmetics. It is a decision with a number attached.
What the extension bought
I took every loan that was extended and later charged off. That is 80,714 loans carrying $1.72 billion in balance at the moment of their first extension. These are the failures. If extensions only delay the inevitable, this is where it should show.
From the first extension to charge-off took a median of 12 months. In that time the borrowers paid 31.9 cents per dollar of balance. When the car was finally repossessed and sold, the lender recovered another 33.7 cents. Total: 65.6 cents.
The alternative is repossessing at the first extension. I cannot see the cars, so I priced that repo at what the same lender recovered on comparable loans it did repossess without extending: same lender, same origination year, same vehicle value band, same loan age. That rate already includes the time it takes to sell the car. It comes to 52.3 cents.
The extension wins by 13.3 points of balance. On a $15,000 balance, that is about $2,000 more cash. It won on 67% of loans.
Add back the extended loans that recovered and paid off, and the gap widens. Across 129,641 extended loans that have fully resolved, the lender collected 83.3 cents against a repo value of 53.9. That is 29.5 points, on 79% of loans.
Eight of twelve, checked
The rule of thumb holds up for the average extended loan. In the 12 months after an extension, borrowers made a mean of 7.4 payments, and 58% made eight or more.
The loans that eventually failed did worse: 5.3 payments out of 12, and only 30% made eight.
Five payments still beat the repo. The repo only returns about half the balance. Five payments plus the eventual recovery get the lender to about two thirds. The borrower who is going to default anyway is still worth more paying than parked at auction.
Where the edge shrinks
The gain is the same sign everywhere. It is a different size.
By vintage. Loans written in 2020 to 2022 gained 15 to 16 points from extending. 2023 gained 8.6. 2024 gained 1.1, and Santander's 2024 loans went negative, at minus 7.4.
Read the recent years with care. More than half of extended loans are still open and are left out. The ones that have already failed in 2023 and 2024 books failed fast, and fast failures pay less. So the recent gains are probably understated. A shrinking edge on the newest books is still worth watching.
By lender. AmeriCredit gained the most at 17.9 points, then Exeter at 12.0 and Santander at 10.6.
By vehicle value. The gain is widest on the most expensive cars. Cash collected after the extension runs 30 to 34 cents whatever the car. Recovery rises with value, from 23.5 cents under $10,000 to 40.6 cents above $30,000. The gain runs from 10.3 points on the cheapest band to 15.3 on the most expensive.
When it was granted
Extensions go out early. In the month before the first extension, 61% of extended loans were less than 30 days past due. Another 25% were 30 to 59 days late, 10% were 60 to 89, and 3% were 90 or more.
The loans that went on to charge off were further behind, though not by as much as you would expect. Half were still under 30 days late when the lender stepped in. 29% were 30 to 59, 16% were 60 to 89, and 5% were 90 or more.
Then the extension resets the clock, and the loan reads current. A borrower who was 45 days behind last month shows up as on time this month, and may stay that way for a year before the charge-off.
Two things true at once
The extension is rational. It collects more cash than the repo, even on the loans that die. A lender that refused to extend would leave money on the table.
The extension also blinds whoever is reading the delinquency number: the investor, the warehouse bank, the rating agency. A loan that will charge off next year reads current today. In Current, then gone, Exeter's reported delinquency fell while same-age charge-offs rose 61%. This is the other half of that story. The lender was being paid to wait.
Method and limits
Data. SEC ABS-EE loan-level filings for 14 Exeter (EART), Santander Drive (SDART) and AmeriCredit (AMCAR) trusts, 923,232 loans, filings through July 2026. Prime shelves and captives were left out on purpose.
Extension. The first filing month where the extension count is above zero. Balance at extension is the beginning balance that month.
Charge-off and recovery. A charge-off is any charged-off principal above zero. Recovery is the recovered amount, which on these shelves already includes repossession proceeds.
What the extension returned. All payments from the extension month through the last filing, plus recovery.
The repo comparison. Balance at extension times the recovery rate on never-extended loans that charged off, matched on lender, origination year, vehicle value band and loan age. Matched cells with fewer than 200 loans fall back to the lender and vintage rate. These rates already include the time to sell the car, so no separate holding cost is subtracted. This is a repo priced at what the same lender got on comparable loans. The tape does not show the cars.
Payment count. A month counts as paid if the borrower paid at least half the scheduled payment. Measured only on loans with 12 months observed after the extension.
Censoring. 52% of extended loans are still open and are excluded. Recent vintages lean toward fast failures, so their gains are likely understated.
Scope. Three public subprime shelves, not the whole market. The extension field is a count reported by the servicer, and definitions can differ by issuer.
Proof: every figure, traced
Computed means counted from the Exhibit 102 loan tapes on file with the SEC, construction stated above. Derived means arithmetic on computed figures.
| Claim as stated | Construction | Status |
|---|---|---|
| 923,232 loans, 14 trusts: EART 254,235; SDART 423,746; AMCAR 245,251 | All EART, SDART and AMCAR panels in the store, one row per loan | Computed |
| Ever extended 40% / 36% / 18%; charge-off rate ever extended 40.5 / 23.6 / 23.9%; never extended 20.0 / 12.8 / 13.7% | Extension = paymentExtendedNumber above zero in any period; charge-off = chargedoffPrincipalAmount above zero | Computed |
| 80,714 extended-then-charged-off loans; $1.717 billion balance at first extension; median 12 months (mean 14.7) from first extension to charge-off | Beginning balance in the first extension month | Computed |
| Realized 65.6¢ = 31.9¢ payments after extension + 33.7¢ recovery; counterfactual 52.3¢; gain 13.3¢; extension won on 67.2% of loans | Dollar-weighted; counterfactual = balance × matched recovery rate (lender, vintage, vehicle value band, months-on-book band; cells under 200 fall back to lender-vintage) | Computed |
| About $2,000 on a $15,000 balance | 0.133 × 15,000 = 1,995 | Derived |
| Resolved extended loans 129,641 (48,927 paid off, 80,714 charged off): realized 83.3¢, counterfactual 53.9¢, gain 29.5, won on 79.4% | Same construction, charged-off and paid-off loans together | Computed |
| Payments in next 12: all extended 194,017 loans, mean 7.4, 58.0% at 8+; extended-then-charged-off 68,329 loans, mean 5.3, 30.3% at 8+ | Paid month = cash at least half the scheduled payment; loans with 12 observed months | Computed |
| Gain by vintage: 2020 +15.8, 2021 +14.9, 2022 +16.1, 2023 +8.6, 2024 +1.1; Santander 2024 −7.4; 2015 to 2019 +35 to +40 on 235 to 1,725 loans | Tables 3c and 3e of the working output | Computed |
| Gain by lender: AmeriCredit +17.9 (21,081 loans), Exeter +12.0 (41,019), Santander +10.6 (18,614) | Table 3b | Computed |
| Gain by vehicle value band +10.3 / +13.6 / +12.6 / +12.7 / +15.3; recovery 23.5 / 26.1 / 29.4 / 35.5 / 40.6¢; cash after extension 34.0 / 33.9 / 32.1 / 32.1 / 30.1¢ | Table 3d | Computed |
| Days past due the month before first extension: all extended 61.0 / 25.2 / 10.4 / 2.8% (0 to 29 / 30 to 59 / 60 to 89 / 90+); later charged off 49.5 / 28.9 / 15.6 / 5.1%; extension month itself 96.0% at 0 to 29 | Table 7; the same-month figure is post-reset and is not used as borrower status | Computed |
| 51.7% of extended loans still open and excluded | Section 6 of the working output | Computed |
| Extended loans charged off at 2.4 times the rate of never-extended loans (19.65% against 8.31%), odds ratio 2.60 after controls; Exeter same-age charge-offs rose 61% (11.16% to 17.93% at month 24) while reported 60+ delinquency fell | Issue 18, Sections I and II | Computed |
Loan-level data. SEC Form ABS-EE, Exhibit 102, monthly asset-level files for Exeter Automobile Receivables Trusts, Santander Drive Auto Receivables Trusts and AmeriCredit Automobile Receivables Trusts, 14 trusts, filings through the July 2026 reporting period. Fields per 17 CFR 229.1125 Item 3: paymentExtendedNumber (3(j)(2)), reportingPeriodBeginningLoanBalanceAmount, actualInterestCollectedAmount and actualPrincipalCollectedAmount, chargedoffPrincipalAmount, recoveredAmount, currentDelinquencyStatus, vehicleValueAmount, originationDate.
Reproduction. Script and full output tables are kept with the working files for this series; every table referenced in the proof column is in that output.
Companion issues. Issue 18, the delinquency-loss decoupling this piece answers. Issue 13, the extension-adjusted delinquency construction. Issue 15, recovery by vintage.
Where this brief reasons beyond what the filings show, it says so. Point-in-time reading of filings through the July 2026 reporting period and the public record through October 31, 2026. LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any institution mentioned. This is not investment, legal or accounting advice.