Bottom line
Three things the tapes say, in the order they killed the theory.
There is almost no 84-month subprime paper. Exeter's longest loan is 80 months, Santander's 75, Bridgecrest's 72, and neither of the two big subprime shelves has written a single 84-month loan in any year since 2019. The 80,913 loans at 84 months or longer in the whole record are 80% new cars, sitting at GM Financial, Ally, Ford and Hyundai. The one lender writing 84-month used paper at scale is AmeriCredit, GM's subprime arm.
The term absorbed nothing. Hold the amount and rate where they were and swap only the term from 2020 to 2024: the median sub-600 borrower saved $3 a month at Exeter, nothing at Santander, $4 to $5 at AmeriCredit. The rate move over the same window cost the same borrower $17 to $19, $38 to $41 and $72. The median payment rose 27% in four years. Nobody hid the squeeze in the term because the term was already at the ceiling. It was passed straight through.
What did move was the income line, at the one lender that files it. Exeter's payment-to-income box held, and tightened: a median of 10.1% in 2019, 11.6% at the 2022 peak, 10.0% in 2024. It held because the income underneath it rose 39% while the share of that income anyone verified fell from 88% to about 21%.
I · There is almost no 84-month subprime paper
Everyone talks about 84-month loans as a subprime problem. In the filings they are not. Maximum original term ever filed, by shelf: Exeter 80, Santander 75, CarMax 74, Bridgecrest 72. Zero loans at 84 months or longer in any vintage on either of the two largest subprime shelves.
| Shelf | Longest term filed | Share of loans at 84+ months | Loans at 84+, all years |
|---|---|---|---|
| Exeter | 80 | 0.00% | 0 |
| Santander | 75 | 0.00% | 0 |
| Bridgecrest | 72 | 0.00% | 0 |
| CarMax | 74 | 0.00% | 0 |
| Carvana | 84 | 0.00 to 0.74% | 192 |
| AmeriCredit (GM Financial subprime) | 88 | 5.2 to 35.4% | 28,879 |
| GM Financial prime | 88 | 1.4 to 18.3% | 21,089 |
| Ally | 92 | 3.2 to 13.2% | 15,857 |
| Hyundai | 84+ | rising | 7,863, all new cars |
| Ford | 86 | 2.9 to 13.7% | 7,031, all new cars |
AmeriCredit is the exception, and it is a captive. Its share of used loans at 84 months or longer went from 1.7% in 2020 to 15.0% in 2024, on a median origination score near 585. Ally wrote zero 84-month used loans in 2023, 2024 or 2025. Honda, Nissan, Ford and Toyota all shortened their average term between 2019 and 2024.
The cap is a choice, not a limit on the collateral. Exeter's new-car loans also stop at 80. If 84-month subprime used paper exists in volume, it sits on bank and credit union balance sheets where nobody files a tape.
II · The term did not absorb anything
The rate did about fifteen times the work the term did. The share of sub-600 borrowers paying more than $700 a month went from 17% to 37% at Santander, 8% to 21% at Exeter, 9% to 23% at AmeriCredit and 4% to 25% at CarMax.
Term creep is real, small, and mostly one lender
Within a fixed score band, Exeter's average term rose about 2.5 to 5 months from 2020 to 2024, monotonically in every band. Santander's did not move at all, though its share above 72 months rose from 10% to 16%. AmeriCredit's sub-540 band went up about two months. Carvana is the exception: its share above 72 months went from 7% in 2020 to 41% in 2025.
Pool all the lenders together and you get a spike in 2022 that belongs to none of them. The 2022 trusts in the public record happen to be the long-term shelves, and Santander contributed only 3,997 loans written in 2022. Read term within a lender or do not read it.
Does longer term lose more money?
Raw, longer loans look safer. Exeter's 2022 originations at 24 months on book: 21.7% charged off at 72 months or under, 21.6% at 73 to 78, 11.1% above 78. That is the confound the CFPB flagged in its 2017 study of six-year loans, and this data reproduces it exactly. The longest loans go to the best borrowers. Exeter's 79-plus bucket carries a median score of 641 against 581 and a rate of 18.7% against 25.3%. Longer term buys a lower rate at Exeter, Santander and AmeriCredit, not a higher one. Any model treating term as a risk add-on has the pricing sign backwards.
Control for lender, vintage, score band, loan-to-value, new against used and loan size, and a penalty appears: 73 to 78 month loans charge off about 0.3 to 0.4 points more at 24 months than loans at 72 or under, in 66 matched cells. Real, but small. The gap between the raw number and the controlled one is the finding: long-term paper sits on cars twice as expensive, $32,095 financed against $15,444 at an identical median score. Long paper is also extended more. Exeter 2023: 49% of 73 to 78 month loans had at least one extension against 39% of shorter ones, and the direction holds at every lender in every year.
The 84-month loss question stays open. Once you control for everything, one matched pair survives.
III · Where the easing went
Exeter is the only lender in the public record that files payment-to-income, the income verification level and the origination rate on every loan. 253,404 loans.
The PTI box held. Median 10.1% in 2019, a peak of 11.6% in 2022, 10.0% in 2024. Within every score band the median sits between 10% and 12%, and the 2024 book is the tightest in the series.
It held because the income side moved. Payment divided by PTI gives the income the lender used. The median was $4,028 a month in 2020 and $5,596 in 2024, up 39%. Part of that is mix, since the 620 to 659 band more than doubled in loan count over the same years. Inside the fixed sub-540 band it is still up 30%. The Atlanta Fed's wage tracker for the bottom quartile of earners compounds to about 25% over the same four years. Reported borrower income at Exeter grew about half again as fast as wages for comparable workers.
And the share of that income anyone verified fell off a cliff: 61% of loans filed as income-verified in 2019, 88% in 2020, 32% in 2021, then 21%, 21%, 23%, and 16% in 2025. The drop came in 2021, before rates moved. The term bought $3 a month. Verification went from nine loans in ten to one in five.
Two things that stop this being a fraud story
Stated incomes at Exeter are lower than verified incomes in the same score band, by 20% to 27% in 2022 and 2023. If dealers were inflating the number you would expect the reverse. Exeter's verified borrowers also score lower than its unverified ones, 558 against 591 in 2021. Verification looks targeted rather than random, and a high stated income may be one of the things that triggers it.
And stated-income loans do not perform worse. Within score band, loan-to-value band and vintage, 80 matched cells, no difference.
IV · What the lenders did about it
Every lender in the record faced the same squeeze. They did not all do the same thing.
Credit Acceptance cut what it advances dealers. Its advance rate on dealer loans went from 46.4% of the contract on 2022 assignments to 43.2% on 2025 assignments, and it stayed profitable through an 8.2-point miss on its 2022 collection forecast, 67.5% initially against 59.3% now.
Westlake put money into the structure. After Morningstar DBRS raised its base-case loss on the 2022-2 deal from 10.10% to 18.75%, Westlake made a $30 million capital contribution to that trust's reserve account at the June 2025 payment date.
Ally moved up-market. Its highest credit tier was 47% of retail auto originations in the second quarter of 2026 against 42% a year earlier.
Credit unions pulled back, then stopped. Their auto book fell $10.4 billion year over year to the first quarter of 2025, and was essentially flat a year later, down $0.3 billion to $479.6 billion, with new-car loans still shrinking and used growing.
America's Car-Mart charged off 27.6% of its book in the year to April 2026 and its 10-K carries a going-concern warning. Issue 16 covers what it restated.
Consumer Portfolio Services grew into the same pool. It purchased $757.7 million of contracts in the second quarter of 2026, up 75% on a year earlier and the most in its history. That paper has not seasoned.
Fitch's subprime 60-plus delinquency index hit 6.90% at the January 2026 reading, the highest in a series that starts in 1994, with unemployment at 4.4% in the month it covers. Prime moved from 0.39% to 0.43%. That is not a shock arriving. It is a squeeze that was priced through to the payment, in plain sight, for four years.
What I would watch
Not the term. The term is at the ceiling and has been for years. Watch the income line. If a lender's PTI box is holding while its median payment climbs 27%, ask what happened to the denominator, and ask who checked it.
V · What this does not show
Verification code 2 means stated, not unchecked. If Exeter moved income confirmation to a payroll or bank-data feed and kept filing those loans as code 2, the verification collapse is a filing artifact and most of Section III goes away. I have not been able to rule that out. Until I can, the claim is that income stopped being filed as verified, not that it stopped being checked.
One lender files PTI and verification. The income finding is Exeter's until a second shelf discloses the same fields.
Securitized pools are selected from the book, not the book. And vintage and trust are confounded, since each trust holds one or two origination years; the within-shelf cuts are the clean ones, and even those can move with pool selection.
Nothing here speaks to current originations. There is no 2026 vintage on tape and 2025 is thin, 1,455 Exeter loans. Whatever is being written this quarter is not in this data.
No buy-here-pay-here dealer files a tape, so none of this covers the operators holding their own paper.
Two inputs are rebuilt, not read. Scheduled payment is the median positive value over each loan's life, because first-period filings are unreliable and Carvana files zero there. The rate for shelves that do not file one is solved from amount, payment and term, and runs about 0.7 points high against Exeter's filed rate.
The wage comparison is approximate. The Atlanta Fed tracker is median growth for individuals observed a year apart, a job-holder measure. Compounding its December readings gives about 25% for the bottom quartile from 2020 to 2024. A different wage series would give a different number; none would get close to 39%.
Proof: every figure, traced
Computed means counted from the Exhibit 102 loan tapes on file with the SEC, construction stated. Verified means read at the named external document. Derived means arithmetic on figures above it.
| Claim as stated | Source and construction | Status |
|---|---|---|
| 3,077,939 loans, 48 trusts, 16 lenders, originations 2014 to 2025, loan-months January 2021 to July 2026 | SEC Form ABS-EE Exhibit 102 filings for the 48 public auto trusts in the panel; fields per 17 CFR 229.1125 Item 3 (originalLoanTerm, originalLoanAmount, obligorCreditScore, paymentToIncomePercentage, obligorIncomeVerificationLevelCode, originalInterestRatePercentage, paymentExtendedNumber) | Computed |
| Maximum filed term: Exeter 80, Santander 75, CarMax 74, Bridgecrest 72, Carvana 84, AmeriCredit 88, GM Financial 88, Ally 92, Ford 86; zero 84-plus loans at Exeter and Santander in every vintage | Tape, max of originalLoanTerm by shelf and vintage | Computed |
| 80,913 loans at 84 months or longer; 80.4% new vehicles; AmeriCredit 28,879, GM Financial 21,089, Ally 15,857, Hyundai 7,863, Ford 7,031, Carvana 192; AmeriCredit used 84-plus share 1.7% (2020) to 15.0% (2024); Ally zero 84-month used loans 2023 to 2025 | Tape, vehicleNewUsedCode by shelf | Computed |
| Median sub-600 payment $438 to $560 (Exeter), $494 to $625 (Santander), $438 to $559 (AmeriCredit); term effect $3 / $0 / $4 to $5; rate effect $17 to $19 / $38 to $41 / $72 | Tape, shift-share holding two of amount, rate and term at the observed medians | Computed |
| "About fifteen times": rate effects of $18, $40 and $72 against term effects of $3, $0 and $4.5 | Arithmetic on the row above; the Santander ratio is undefined and the figure is the Exeter and AmeriCredit average | Derived |
| Share of sub-600 borrowers paying over $700: Santander 16.6 to 37.0%, Exeter 7.6 to 21.4%, AmeriCredit 9.1 to 22.8%, CarMax 4.4 to 24.9% | Tape, 2020 against 2024 vintages | Computed |
| Within-band mean term: Exeter sub-540 70.89 to 73.43, 620 to 659 70.55 to 75.72; Santander sub-540 70.65 to 70.72, share above 72 months 10.08 to 15.89%; AmeriCredit sub-540 72.41 to 74.18; Carvana share above 72 months 7.31% (2020), 34.92% (2024), 40.73% (2025); Santander 2022 originations in panel 3,997 | Tape, scored loans only | Computed |
| Exeter 2022, CNL at 24 months by term: 21.70% at 72 or under, 21.60% at 73 to 78, 11.13% above 78; 79-plus median score 641 against 581, rate 18.67% against 25.32% | Tape, gross cumulative charge-off at month 24, dollar-weighted, charge-off absorbing, 60% observation gate | Computed |
| Fully controlled term penalty: 66 paired cells, mean +0.31pp, median +0.42pp, longer worse in 58%; partially controlled 171 pairs mean −0.16pp; long-term paper $32,095 financed against $15,444 at identical median score 611 | Tape, controls for shelf, vintage, score band, LTV band, new/used, financed-amount band | Computed |
| Extensions: Exeter 2023 49.24% of 73 to 78 month loans against 39.24% of 72 or under; same direction at Exeter 2022, AmeriCredit 2022 and 2024, Santander 2023 | Tape, positive paymentExtendedNumber in any period | Computed |
| Exeter 253,404 loans; PTI medians 10.13 / 10.99 / 11.46 / 11.56 / 11.33 / 9.97 for 2019 to 2024; p90 16.62 in 2024 | Tape, Exeter static fields | Computed |
| Implied income median $4,028 (2020) to $5,596 (2024), +39.0%; +30% inside the sub-540 band; 620 to 659 band 6,145 to 13,783 loans | Tape, scheduled payment over filed PTI | Computed |
| Income-verified share 61.2 / 87.9 / 31.6 / 20.8 / 20.6 / 22.8 / 16.1% for 2019 to 2025 | Tape, obligorIncomeVerificationLevelCode in (3, 4) | Computed |
| Stated incomes 20 to 27% below verified in the same score band, 2022 and 2023; verified median score 558 against 591 unverified, 2021; stated-income loans no worse, 80 paired cells, mean −0.17pp | Tape | Computed |
| Bottom-quartile wage growth about 25%, 2020 to 2024 | Atlanta Fed Wage Growth Tracker, 1st to 25th wage percentile, 12-month moving average, via FRED; December readings 5.5, 7.6, 5.9, 4.3 for 2021 to 2024, compounded to 25.4% | Derived |
| CFPB, November 2017: six-year-plus loans defaulted at about 8% against 4% for shorter; longer-term borrowers had lower scores (674 against 713) and larger loans | CFPB Quarterly Consumer Credit Trends, Growth in Longer-Term Auto Loans | Verified |
| Credit Acceptance: 2022 advance rate 47.4% overall, 46.4% dealer loans; 2025 dealer loans 43.2%; 2022 forecast 67.5% initial against 59.3% current, variance −8.2% | Credit Acceptance Q4 2025 earnings release, 8-K, forecasted collection and advance rate tables | Verified |
| Westlake 2022-2: base-case loss revised to 18.75% from 10.10%; $30,000,000 capital contribution to the reserve account at the June 2025 payment date | Morningstar DBRS, rating actions on six Westlake Automobile Receivables Trust transactions, July 7, 2025 | Verified |
| Ally: 47% of retail auto origination volume in the highest credit quality tier, Q2 2026, against 42% in Q2 2025 | Ally Q2 2026 press release; Ally Q2 2025 press release | Verified |
| Credit union auto loans down $10.4 billion (2.1%) year over year to $481.4 billion at Q1 2025; down $0.3 billion (0.1%) to $479.6 billion at Q1 2026, new −$3.6 billion, used +$3.3 billion | NCUA Quarterly Data Summary, 2025 Q1; 2026 Q1 | Verified |
| Car-Mart net charge-offs 27.6% of average finance receivables, fiscal 2026; substantial doubt about ability to continue as a going concern | America's Car-Mart Form 10-K, fiscal year ended April 30, 2026 | Verified |
| CPS purchased $757.7 million of new contracts in Q2 2026, up 75% from $433.0 million | CPS Q2 2026 earnings release, 8-K Exhibit 99.1 | Verified |
| Fitch subprime 60-plus delinquency index 6.90% at the January 2026 index date, a record for a series beginning in 1994; prime 0.43% from 0.39% | Fitch Ratings and Black Book, Vehicle Depreciation Report 2026 | Verified |
| Unemployment 4.4% in December 2025, the collection period behind the January 2026 index date | BLS Employment Situation, December 2025, released January 9, 2026 | Verified |
Loan-level data. SEC Form ABS-EE, Exhibit 102, for 48 public auto trusts across Santander, Exeter, AmeriCredit, GM Financial, Carvana, CarMax, Bridgecrest, Ally, Ford, Honda, Hyundai, World Omni, Toyota, Nissan and Volkswagen shelves. Exeter static fields (payment-to-income, income verification level, origination rate) from the same filings. Field definitions per 17 CFR 229.1125, Item 3.
Method. Scheduled payment rebuilt as the median positive value over each loan's life. Implied rate solved from amount, payment and term by the annuity formula and validated against Exeter's filed rate. Charge-off treated as an absorbing state; 24-month cells suppressed where genuine observation covers under 60% of the cell. Duplicate filed periods (19 of 48 trusts) deduplicated on loan and period after field-by-field confirmation that they are exact copies. Band cuts require a filed score of 300 or higher.
External documents. CFPB (November 2017). Credit Acceptance Q4 2025 release. Morningstar DBRS (July 7, 2025). Ally Q2 2025 and Q2 2026 releases. NCUA quarterly data summaries, Q1 2025 and Q1 2026. America's Car-Mart FY2026 10-K. CPS Q2 2026 release. Fitch and Black Book Vehicle Depreciation Report 2026. BLS Employment Situation, January 9, 2026. Federal Reserve Bank of Atlanta Wage Growth Tracker via FRED.
Companion issues. Issue 16 on entry against exit from distress. Issue 15 on recovery by vintage. Issue 13 on extension-adjusted delinquency.
Where this brief reasons beyond what the filings show, it is labeled as an inference. Point-in-time reading of filings through the July 2026 reporting period and the public record through October 18, 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.