# LendRisk Analytics · full text edition Built 2026-10-11. Index: https://lendriskanalytics.com/llms.txt --- title: "What the tape said: entry, not exit" url: https://lendriskanalytics.com/insights/entry-not-exit.html publisher: LendRisk Analytics series: What the Tape Said issue: 16 published: 2026-10-11 kind: Tape read description: "A borrower scored 661 to 780 reaches 60 days past due within two years 1.1% of the time at Ford and 29.3% of the time at Santander. Once any borrower is 60 days down, almost every lender charges off at the same rate. Entry into distress is where subprime risk lives; exit is close to a constant. Counted from 907,334 loans in the public ABS-EE tapes." html: https://lendriskanalytics.com/insights/entry-not-exit.html --- # What the tape said: entry, not exit [← All insights](https://lendriskanalytics.com/articles.html) What the Tape Said · Issue 16 Tape read · 14 min read Subprime auto · Entry versus exit · ABS-EE loan level # Entry, not exit. Loan-level filings through July 2026 · Public record through October 11, 2026 · LendRisk Analytics Ask whether subprime auto lending is risky and the honest answer is that the question is missing a subject. For the borrower pool the losses are enormous and almost entirely expected. For the senior bondholder they barely exist. For the lending firm they are survivable right up until the week they are not, and what kills the firm is rarely the borrowers. I went back to the loan tapes to see which part of that the filings support, and two pieces of the standard story did not survive. 27.9x Gap between the best and worst lender in how often a 661-780 borrower reaches 60 days down by month 24 1.5x The same gap for what happens after that borrower is 60 days down 24.00% Pool principal lost on Exeter's 2022 trust, with every class A note paid in full 1,844 bps Distance between reported and extension-adjusted 60+ delinquency, subprime tier, July 2026 ## Bottom line Forty-four public auto securitizations file a loan-by-loan tape every month under Form ABS-EE. That is 907,334 loans across 43 trusts and fourteen lenders, prime captives and deep subprime shelves side by side, each loan carrying its score at origination, its days past due, its extensions and its charge-off. Nothing in this paper is a published ratio. Every figure is counted from those individual records. Counting them that way says the risk in subprime auto is concentrated in one place, and it is not where the index points. Borrowers with identical credit scores reach serious delinquency at wildly different rates depending on who lent to them. Once they get there, almost every lender's outcome converges. Entry is where the dispersion lives; exit is close to a constant. ## The read A credit score is supposed to be the thing that tells you how a borrower behaves. In these tapes it is a weak predictor of anything once you know the lender's name. A borrower scored between 661 and 780 , which every bureau and every rating agency calls prime, goes 60 days past due within two years about 1.1% of the time at Ford and 29.3% of the time at Santander. Same band, same country, same two-year window, twenty-eight times the failure rate. Then the convergence. Among loans that did reach 60 days down, the share charged off within the next twelve months runs 68.1% at Exeter, 68.9% at Santander, and 67.8% at World Omni, which is a Toyota-affiliated prime captive. Three lenders at opposite ends of the credit spectrum land within 1.1 points of each other. Whatever separates a prime book from a subprime one, it stops operating at the moment a borrower is two payments behind. What this changes If exit odds are near constant, then loss severity is mostly fixed and the whole business reduces to controlling how many loans arrive at that door. That is an underwriting and servicing problem, not a borrower-quality problem, and it is why two lenders can hold the same credit band and write books that behave nothing alike. ## I · Entry and exit Here are the two distributions on one scale. The left column is how often a prime-band borrower reaches 60 days past due by month 24 , one dot per lender. The right column is what share of loans that got there were charged off within a year. Fig 1 · The spread collapses once the borrower is already late Counts of individual loans. Left: ever 60+ by month 24 on book, score band 661-780 at origination, cells with at least 1,000 loans at risk. Right: charged off within 12 months of the first 60+, all bands, among loans whose first 60+ happened on tape. 43 public trusts, 907,334 loans, ABS-EE tapes through July 2026. Left: 131,348 loans at risk in the 661-780 band. Right: 150,688 loans whose first 60+ happened on tape. Charge-off is absorbing here, so a loan that dies stays in the denominator after it leaves the tape. The left column spans twenty-eight to one. The right column spans one and a half to one, and the ordering inside it does not sort by tier: CarMax, a prime shelf, sits at the bottom with 56.1% , and World Omni, also prime, sits at 67.8% , above AmeriCredit's subprime book at 57.5% . Bridgecrest is the one genuine outlier at 86.1% . One caution A borrower holding a 700 score who ends up at Exeter is a borrower the captives declined, for reasons the tape cannot see: payment-to-income, down payment, time in file, a recent derogatory. Some unknown part of that 27.9x is selection rather than lender behaviour. The tape carries no income and no debt-to-income field, so I cannot size it. What the tape does establish is that the score alone, which is what most of the public commentary leans on, carries far less information than its prominence suggests. ## II · Same score, different lender The full grid, read across a row for one lender over the credit spectrum and down a column for one score band across lenders. Fig 2 · The column moves more than the row Ever 60+ days past due by month 24 on book, by score band at origination. Shade runs with the rate. A cell appears only where at least 1,000 loans are at risk and the outcome is known for 60% of the cohort. Subprime shelves are named in breach-ink. A dot means the lender has too little paper in that band to divide. Counts of loans, never balances. | Lender | 300-540 | 541-600 | 601-660 | 661-780 | 781-900 | |---|---|---|---|---|---| | Ford | · | · | · | 1.1% | 0.1% | | Hyundai | · | · | · | 1.6% | 0.1% | | Ally | · | · | · | 2.8% | 0.5% | | GM Financial | · | · | · | 2.9% | 0.4% | | Carvana | · | 10.6% | 7.0% | 3.0% | 0.7% | | CarMax | · | 23.3% | 12.4% | 3.1% | 0.4% | | World Omni | · | · | 14.7% | 4.2% | 0.7% | | AmeriCredit | 26.4% | 19.8% | 13.3% | 19.4% | · | | Exeter | 42.3% | 36.4% | 29.8% | 24.1% | · | | Santander | 41.7% | 35.7% | 35.0% | 29.3% | 8.7% | The same grid as numbers. Values at or above 19% in breach-ink. Read Santander's row from left to right. Its deep subprime band runs 41.7% and its prime band runs 29.3% , a difference of twelve points across four hundred points of credit score. Now read the 661-780 column from top to bottom: 1.1% to 29.3% , twenty-eight points, across lenders who all call that paper prime. The column moves more than the row. ## III · One lender's improvement, read as the market's The standard account of the cycle says the 2022 and 2023 books were written at peak used-car prices to borrowers flattered by stimulus-era credit models, that those cohorts are the problem, and that they are amortising away. Consumer Portfolio Services gave the market the cleanest version of it on its Q2 2026 call, disclosing recovery rates by vintage of 22% for 2022 , 25% for 2023 , 37.5% for 2024 and 47.1% for 2025 . On recoveries, our own tapes agreed; that was Issue 15 . On delinquency the tapes disagree, and the disagreement only shows up once you stop pooling lenders. Fig 3 · The same question asked inside three lenders separately Ever 60+ days past due by month on book, one line per origination year, score band held at 541-600 throughout so neither the lender nor the credit band moves inside a panel. EART 2021-1 through 2025-1, SDART 2021-1 through 2025-1, AMCAR 2021-1 through 2024-1. Each curve stops at the first month its at-risk count falls more than 5%, the point where a pool amortises fast enough that the survivors stop being representative. At month 18 on book, Exeter's cohorts run 32.5% for 2021 , 32.3% for 2022 , 28.2% for 2023 and 25.4% for 2024 . That is the published story exactly: a peak in the pandemic-era books and steady improvement since. AmeriCredit's run 7.6% for 2020 , 13.5% for 2021 , 13.9% for 2022 and 17.9% for 2024 . Its worst cohort is its newest, and the line is still rising. Santander sits between them and essentially flat, 29.4% for 2023 against 28.3% for 2024 . Pooling those three lenders into one curve, which is what an index does, produces a shape that belongs to none of them. In these trusts the 2022 cohort is 69% Exeter with no Santander paper in it at all, and the 2023 cohort is 65% Santander with no AmeriCredit. A year can look better or worse purely because a different lender wrote more of it. How this was caught I ran the pooled version first and it reproduced the published story convincingly. Holding the lender fixed is what broke it. Any vintage comparison drawn across lenders whose composition shifts year to year is measuring the composition. ## IV · A fifth of the pool is current on an extension An extension moves a past-due loan back to current on the tape without the borrower catching up. Add back every loan returned to current that way in the trailing six months and subprime 60+ delinquency goes from 8.10% to 26.55% , a gap of 1,844 basis points on the July 2026 tapes. Prime runs 0.73% against 3.91% , a gap of 318 . The window is a stated convention, not a filed figure, and the magnitude scales with it: three months gives roughly 1,057 basis points on the subprime tier, twelve months roughly 3,105 . The ordering across lenders holds at every window. Fig 4 · What a current-or-not reading does not see Subprime tier, balance-weighted across every subprime deal on tape. The dark line is 60+ delinquency as reported. The lighter line adds back loans held current by an extension granted in the trailing six months. The bars along the bottom are how many deals carry each month. 10 subprime deals and $4.68 billion of collateral at July 2026, 6 deals at the left edge. Extension events read from paymentExtendedNumber, which Item 3(j)(2) of 17 CFR 229.1125 defines as the months a loan was extended during the reporting period. The within-year sawtooth is pool age rather than a credit cycle. Read it July to July, which holds the seasonal position fixed, and the band has widened for three years: 1,098 basis points in 2024 , 1,344 in 2025 , 1,844 in 2026 , with the share of the pool touched by an extension going 11.85% to 14.84% to 21.00% . Those three readings rest on 10 , 12 and 10 deals, so the comparison is not resting on a changing pool count, but the month-to-month shape of the line is. The timing is the part worth keeping. The adjusted measure peaked in January 2026 at 27.86% and the extension stock peaked a month later at 23.09% , while reported delinquency kept climbing until June. Reported delinquency rising through the first half of 2026 is partly the extension stock unwinding and handing loans back. Whether an extension is a cure or a deferral is visible six months later. Across the tapes, loans back at 60 days down within six months of an extension run 40.4% at Exeter and 39.8% at Bridgecrest, against 18.0% at AmeriCredit, 9.1% at World Omni and 4.0% at Honda. Exeter extends about four loans in ten and four in ten of those come straight back. ## V · What the bondholder sees None of the above has reached a senior noteholder. | Trust | Pool loss to date | Class A notes | Delinquency | Trigger | |---|---|---|---|---| | Exeter 2022-1 | 24.00% | all retired | 14.40% | 40.0% | | Exeter 2023-1 | 22.81% | all retired | 12.04% | 40.0% | | Exeter 2024-1 | 17.80% | all retired | 10.48% | 40.0% | | Santander 2023-6 | 9.54% | all retired | 9.96% | 24.0% | | Santander 2024-1 | 9.60% | all retired | 9.70% | 24.0% | Cumulative net loss and delinquency as each servicer reports them in the deal's latest Form 10-D exhibit, July 2026 period. A trust whose class A balances have all gone to zero has paid those noteholders in full. Nothing here is a rating, and excess spread and turbo features are not modelled. Exeter's 2022 trust has written off twenty-four cents of every dollar of original pool principal and every class A and class B holder was repaid at par. Its delinquency trigger sits at 40% and the deal has never been within twenty-five points of it. That is the answer to whether the asset is dangerous, and the answer is that it was priced, structured and tranched so that it would not be, for the people at the top of the stack. ## VI · Where it does break The failures of the last three years did not come through the collateral. American Car Center told staff it was closing in February 2023 , the day after pulling a $222 million bond sale. Tricolor filed Chapter 7 in September 2025 ; federal prosecutors in the Southern District of New York allege it had pledged about $2.2 billion of collateral against roughly $1.4 billion of real loans. PrimaLend, which lent to buy-here-pay-here dealers rather than to borrowers, filed Chapter 11 in October 2025 after vehicle values and delinquencies moved against the collateral behind its dealer lines. Each arrived through the liability side. Our own perimeter makes the same point by omission. Tricolor never filed a single ABS-EE loan tape, because every one of its deals was privately placed. Neither did Westlake, GLS, Flagship, First Investors or Consumer Portfolio Services, whose vintage recovery disclosure I quoted above and cannot check against collateral. The lender whose book could not be examined is the one that turned out not to have the book. The listed operator that got closest to trouble on credit alone is America's Car-Mart, the only public pure-play deep subprime buy-here-pay-here lender. Its FY 2026 10-K reports net charge-offs of 27.6% of average finance receivables, an allowance at 25.15% , and a provision running 40.8% of sales. It has filed for forty-six quarters and is still operating. In July 2025 it filed a non-reliance 8-K because it had omitted the ASC 310-10-50-42 disclosures covering loan modifications to borrowers in financial difficulty, declared a material weakness, and restated. What it now discloses is that 30.3% of its portfolio was modified at least once during FY 2026 and that roughly half of its contracts are modified at some point over their life. The symmetry The one listed subprime lender forced to restate was restating about how much of its book it was extending. That is the same measurement the tapes make directly in section IV, arrived at from the opposite direction. ## What I would watch Not the delinquency headline. It mixes a growing share of smaller and newer issuers into the same number, it moves with tax-refund season, and it tells you nothing about severity. Watch entry rates inside a single lender and a single band, which is the only cut where a change means a change in behaviour. Watch the extension stock, because distress parked there is distress that has not been priced yet. And watch whether any long-tenured issuer loses warehouse or term ABS access, because that is the mechanism that has ended companies, and it leaves no trace in a delinquency index until after the fact. ## Caveats Score bands are the score at origination as the issuer filed it. Issuers do not all file the same kind of score and some file internal tier codes in the same field, so unscored loans are excluded from any weighted average rather than counted as zero. A loan that paid off or charged off before a pool's cutoff never appears on any tape, and a tape ends at the clean-up call, which over-represents survivors at long seasoning; the curves here stop where that begins to bite. Extension-adjusted delinquency is a modelling choice with a stated window and is not a filed number. The CPS vintage recoveries come from a call transcript, not a filed statement, and one transcript service renders the 22% figure as a 2020 -to- 2022 vintage rather than 2022 alone. The Tricolor criminal allegations are allegations until adjudicated. Figures for American Car Center, Tricolor and PrimaLend are drawn from the companies' own filings and from contemporaneous reporting, not from loan tapes, because none of those three filed any. --- title: "What the tape said: the bottom moved more" url: https://lendriskanalytics.com/insights/set-at-signing.html publisher: LendRisk Analytics series: What the Tape Said issue: 15 published: 2026-10-05 kind: Tape read description: "Used cars are worth a third more than before the pandemic and subprime lenders recover eleven points less on them. The averages hide why. Fifty thousand repossessions from the public loan tapes, split by the year each contract was written: the cohorts that make up nearly all of what is being repossessed are recovering less than a year ago, and the index only looks flat because newer paper is replacing older. Every figure computed from the raw filings." html: https://lendriskanalytics.com/insights/set-at-signing.html --- # What the tape said: the bottom moved more [← All insights](https://lendriskanalytics.com/articles.html) What the Tape Said · Issue 15 Tape read · 15 min read Subprime auto · Recoveries by vintage · ABS-EE loan level # The bottom moved more. Loan-level filings through July 2026 · Public record through October 5, 2026 · LendRisk Analytics Over the past decade the value of a used car went up and the amount a subprime lender gets back when it repossesses one went down. Both at once. The Manheim index sits about a third above its 2019 level; Fitch's subprime recovery index closed 2025 at 32.64% against a pre-pandemic average of 43.73%. The usual explanation stops at the auction. This one goes to the contract. Recovery is a car's sale price divided by the balance still owed on it, and the balance was set the day the loan was written. Contracts written in 2022 and 2023 were priced at the top of the market, with the shortfall on the trade-in rolled into the new note. Bigger balances on the same cars. The bottom of the ratio moved more than the top, and no index built on an average across vintages can show you that. The loan tapes can. 50,865 Repossessions with a reported sale price, Santander and Exeter public trusts, 2021 to July 2026 24.7% Exeter recovery on 2022-vintage loans sold in the past twelve months 47.3% Exeter recovery on 2024-vintage loans sold in the same twelve months 50.0% Santander pooled recovery in 2026, flat on 2025, while 98% of the paper inside it recovered less ## Bottom line Take every public auto ABS trust that reports a sale price on each repossessed vehicle. That is Santander's five SDART trusts and Exeter's five EART trusts in the public record, 50,865 repossessions between them with proceeds attached, plus Carvana's prime shelf as a control. Divide each sale price by the loan balance just before the loss. Group by the year the contract was written. Then do the thing the indices do not: hold the sale date fixed and compare vintages against each other in the same auction months. | Repossessed cars sold July 2025 to June 2026, by year the contract was signed | Exeter, cars sold | Exeter recovery | Santander, cars sold | Santander recovery | |---|---|---|---|---| | Written 2020 | 322 | 29.7% | 518 | 48.6% | | Written 2021 | 1,159 | 22.5% | 414 | 40.1% | | Written 2022 | 1,609 | 24.7% | 253 | 34.6% | | Written 2023 | 2,807 | 32.8% | 4,282 | 46.5% | | Written 2024 | 4,084 | 47.3% | 3,412 | 57.0% | Counts are repossessed vehicles sold at auction in the twelve months shown with a sale price on the tape, not loans originated; they cover only the loans inside each lender's public trusts, and older vintages are small because most of that paper has paid off or charged off years earlier. Recovery is sale proceeds (net of repossession expenses, as filed) over the loan's balance before loss, dollar-weighted within each cell. Same twelve months of auctions in every row, so the market is held constant and only the contract differs. Every figure in this brief is computed from the Exhibit 102 loan tapes each trust files monthly with the SEC. Same auction floor, same twelve months. A 2024 Exeter contract came back with 47 cents on the dollar and a 2022 contract with 25. Santander, a stronger book, shows the same shape at a higher level: 57 against 35. These are not averages of different markets. They are different balances meeting the same bids. The read Recovery is reported as one number because that is how an index works, and the one number is lying by omission. Santander's pooled recovery on everything it sold was 51% in 2024, 50% in 2025 and 50% in 2026, which reads as a floor. Underneath, **the 2022, 2023 and 2024 contracts, which were 98% of what it repossessed in 2026, each recovered less in 2026 than in 2025, and less in 2025 than in 2024.** The only cohort that rose was 2020 paper, one car in a hundred. The pooled line held because 2024 originations, which recover best, went from nothing to nearly half of what was being repossessed. The index measured the mix. The vintage measures the loan. ## I · What a recovery index is averaging Start with the arithmetic, because the whole brief rests on it. Recovery rate is proceeds over balance. Proceeds are set by the auction on the day the car sells. Balance is set by three things that were decided at signing: the price the dealer put on the car, how much of that price plus rolled-in negative equity plus add-ons was financed, and how fast the term lets it amortize. A contract priced at the 2021 peak carries a balance the 2026 auction cannot reach. A contract priced after the index gave back a fifth does not. Put both through the same sale and the first one recovers less, not because the market is weak but because its denominator is large. A recovery index mixes those two contracts into one figure, weighted by whatever happens to be getting repossessed that month. When the mix shifts toward younger paper, the index can hold flat or rise while every cohort inside it deteriorates. That is not a hypothetical. It is Santander's last three years. Chart 1 · Santander: the pooled line holds while the paper underneath it falls Recovery by year of sale, 2022 to 2026. The dark line is all repossessions pooled, the way an index reports. The thin lines are the same repossessions split by the year the contract was written. Cells with fewer than 40 sales are not plotted. Santander SDART 2021-1, 2022-1, 2023-6, 2024-1 and 2025-1, 30,035 repossessions with proceeds. Pooled: 69.6 / 57.6 / 51.3 / 50.4 / 50.0. By vintage, 2024 to 2026: 2020 paper 52.1, 49.1, 52.4; 2021 paper 44.5, 40.2; 2022 paper 44.9, 37.1, 34.9; 2023 paper 55.6, 49.6, 45.5; 2024 paper 58.6, 55.6. Share of repossessed balance written in 2024: zero in 2024, 25% in 2025, 47% in 2026. Y axis 4.4 pixels per point, linear. Read the two 2026 endpoints. Pooled, 50.0%, a tenth of a point below 2025. By vintage, the 2023 paper went from 49.6 to 45.5 and the 2024 paper from 58.6 to 55.6, and together those two vintages were 96% of what Santander repossessed in the first seven months of 2026. The line that looks like stability is a composition effect. Had the mix stayed where it was in 2025, the pooled figure would have fallen two to three points instead of none. **Inference** Fitch's subprime recovery index rose from 32.64% at the end of 2025 to 39.5% in June 2026 and 38.0% in July. Some of that is tax season, which Fitch says. Some of it, on this evidence, is the 2022 and 2023 cohorts amortizing out of the index while 2024 and 2025 paper comes in. That second part is not a recovery in anything. It is a change of subject. The same mechanism explains why CPS could report its recovery rate rising from 28.5% to 33.3% in two quarters while telling investors the gain came from older vintages leaving the book. ## II · Same lender, same auction, different year on the contract The cleanest test strips the market out entirely. Take only cars sold in one window, July 2025 through June 2026, so every row faces the same bids, and split by the year the contract was signed. That is the table at the top. Chart 2 is the same thing for both subprime shelves side by side. Chart 2 · Recovery by origination year, cars sold in the same twelve months July 2025 through June 2026 sales only. Exeter in red, Santander in dark. Counts beneath each bar. The auction market is the same in every column; only the contract date changes. Same construction as the opening table. Santander's 2022 column rests on 253 sales because most of its 2022 originations sit in trusts that are not in the public record; the within-trust check in the Proof section gives the same ordering at 34.7% for SDART 2025-1 alone. Y axis 3 pixels per point, linear. Two features of the shape matter. First, 2022 is the floor at Exeter and near it at Santander, with 2021 close behind. Those are the contracts priced at or just after the Manheim peak of 257.7. Second, the climb from 2022 to 2024 is steep and monotonic at both lenders: eight points a year at Santander, eleven at Exeter. Those are the years the index gave back a fifth of its value and underwriting tightened, and both of those shrink the denominator. Carvana's prime trusts, the control, show a muted version of the same thing in the same window: 2021 paper 35.5%, 2022 paper 37.3%, 2024 paper 47.2%. Prime borrowers put more down and roll less in, so the vintage gap is narrower. It does not disappear. ## III · Price paid, or amount advanced? Two things make a 2022 balance large. The car was expensive, and the lender financed more than the car was worth, because the trade-in was underwater and the shortfall was rolled forward. The loan tape separates them. Each loan carries the vehicle's value at origination and the original amount financed, so the loan-to-value at signing is a direct field, not an estimate. Cut each vintage by that ratio. Chart 3 · Exeter: recovery by origination year and loan-to-value at signing Dollar-weighted recovery, all sale dates, cells with at least 40 repossessions. Read down a column to see the vintage effect at a fixed advance rate. Read across a row to see the advance-rate effect inside one vintage. | Written | LTV ≤ 100% | 100 to 110 | 110 to 120 | 120 to 130 | Over 130% | |---|---|---|---|---|---| | 2020 | 66.2% | 61.6% | 57.5% | 55.5% | 53.8% | | 2021 | 43.4% | 37.7% | 38.0% | 36.4% | 32.5% | | 2022 | 39.3% | 30.2% | 28.9% | 24.9% | 24.1% | | 2023 | 46.1% | 38.2% | 34.4% | 32.3% | 27.1% | | 2024 | 61.5% | 52.7% | 50.2% | 45.3% | 40.2% | EART 2021-1 through 2025-1, 20,830 repossessions with proceeds. Santander's grid has the same two gradients: 2022 runs 50.9 / 47.4 / 39.9 / 38.5 / 32.2 across the LTV buckets and 2024 runs 69.8 / 60.3 / 56.8 / 54.6 / 48.1. Loan-to-value is original amount financed over the vehicle value the servicer reported at origination, both fields on every loan. Both gradients are there, and neither explains the other away. Across any row, more financed against the car means less recovered: the far-right column runs 10 to 20 points below the far-left in every vintage. That is the negative-equity effect, and it is real. But look down the left-hand column, where every loan was financed at or under the car's value and the advance rate cannot be the difference. A 2022 contract at that LTV recovered 39%. A 2024 contract at the same LTV recovered 62%. Twenty-two points, with the amount advanced held constant. That is the price paid for the car, and it is the larger of the two effects. **Inference** The tape settles the question Issue 15 left open three days ago. The vintage effect is mostly the price of the car at signing, with advance rate as a second, independent drag. One consequence for a lender reading its own severities: a 2022 book will underperform a 2024 book at the same auction by something like 20 points of recovery, and no amount of remarketing skill closes that. It was closed, or not, at the desk. ## IV · Where the 2020 paper fits One more column deserves a sentence, because it is the mirror image. Contracts written in 2020 recover best of any cohort at both lenders: 69% at Santander and 58% at Exeter across all sale dates. Those cars were priced before the run-up and much of the paper was repossessed and sold into the 2021 and 2022 peak. The same arithmetic that punishes a 2022 contract rewarded a 2020 one. A recovery index that happened to be heavy in 2020 paper in 2022 and heavy in 2022 paper in 2025 would show a collapse that was partly the market and partly the calendar. ## V · What others have found No rating agency, regulator or vendor has published a recovery split by origination year and loan-to-value at signing, which is why the tape was worth running. Everything around that result has been published, and it all points the same way. 1 **Fitch names the vintages and the two drags.** Its 2026 depreciation report calls 2022 and 2023 the cohorts where "performance weakness was most acute," originated "during a period of market expansion and comparatively looser underwriting standards" and "later exposed to rising debt servicing burdens, elevated cost of living pressures, and declining used vehicle values." It records lenders tightening from late 2022 "through higher minimum FICO thresholds, lower loan to value caps, and enhanced income verification." And it says why the auction did not rescue subprime: recoveries "did not meaningfully benefit from tariff related used vehicle demand in early 2025, as collateral pools typically consist of older, higher mileage vehicles with limited secondary market appeal." Prime did benefit, with a trailing recovery rate of 62.27% and a May 2025 peak of 72.08%. That is the market effect, present where the collateral is young and absent where it is not. 2 **S&P puts 2022 at a near record on losses.** In May 2025 it reported subprime 2023 and 2024 vintages "reporting lower cumulative net losses than 2022's near-record-high levels," with the 2023 vintage at 7.30% at month seventeen against 7.99% for 2022 and 8.00% for 2008. Loss and recovery are two sides of the same ratio; a vintage that loses like 2008 is a vintage that recovers badly. 3 **The CFPB measured the rolled-in balance directly.** Its June 2024 report on negative equity, from its auto finance data pilot, found 11.6% of vehicle loans between 2018 and 2022 carried negative equity from a prior loan; that those accounts averaged a loan-to-value of 119.3% against 88.9% with a positive trade-in and 101.6% with none; and that borrowers who financed negative equity "were more than twice as likely to have their account assigned to repossession within two years" than those with positive equity. That is the right-hand column of Chart 3, measured on a different dataset. 4 **TransUnion dated the climb in advance rates.** Average originating loan-to-value on used vehicles reached 125 in the first quarter of 2023, from 110 a year earlier and 112 in early 2020. Those are the contracts that sit in the 2023 row of the grid. 5 **One source runs the other way, and it should.** Fitch attributes part of the rise in its delinquency index to "composition effects as stronger pre-pandemic vintages amortized and were replaced by weaker post pandemic cohorts." That is a mix effect with the opposite sign to the one in Chart 1, and both are right. Older paper is better credit, so its departure raises the delinquency index. Older paper is also the worst collateral balance, so its departure raises the recovery index. The same roll-off makes delinquency look worse and recoveries look better at the same time. Neither index is reporting a change in any loan. ## VI · What this does not show 1 **Two subprime servicers, not the industry.** Santander and Exeter are the only subprime shelves in the public record that report a sale price on each repossessed car. AmeriCredit flags repossessions but reports proceeds on almost none; GM Financial reports none. Both are covered in the Proof section on the one ratio they do file, recoveries over charged-off principal, and the vintage ordering is the same, but it is a different measure and is not mixed with the one above. Westlake, CPS, DriveTime, Flagship and Credit Acceptance issue under Rule 144A with no public loan tape at all. 2 **Exeter's public trusts went nearly dark on repossessions in 2023 and 2024.** They reported 725 and 373 repossession flags in those years, and proceeds on 896 and 428 loans, against 6,172 flags in 2022 and 25,370 in 2025. This is the reporting break Bill Ploog documented in Exeter's flags, visible in the proceeds field too, followed in 2025 by a catch-up in which 25,370 loans were flagged and proceeds reported on 10,121. Exeter's 2023 and 2024 sale-year figures are excluded from any pooled-versus-vintage comparison, which for Exeter uses 2025 and 2026 only. The same-window table begins in July 2025 and is not affected, but a reader should know that Exeter's 2025 proceeds cover well under half of the repossessions it flagged that year. 3 **Each trust is a selected pool.** A 2022 loan in SDART 2023-6 and a 2022 loan in SDART 2025-1 passed different eligibility screens at different times. The within-trust rows in the Proof section show the ordering holds inside single trusts, which is the fair test; the pooled vintage figures mix selection effects and are presented with that caveat. 4 **Levels are not comparable to Fitch's or CPS's.** Fitch measures recoveries against gross losses. CPS measures auction proceeds net of expenses against balance at time of sale. This brief measures filed proceeds, which the regulation defines as net of repossession fees and expenses, against the largest reported balance in the three months before the proceeds appear, which catches the balance before any charge-off stub. The three definitions move together but sit at different levels. Only direction is compared across sources. 5 **Santander's 2022 vintage is thin in the public record.** 529 repossessions in total and 253 in the same-window table, because Santander's 2022 originations were largely securitized into trusts this brief does not hold. The direction matches Exeter's at ten times the count, and the within-trust figure for SDART 2025-1 is 34.7%, but the Santander 2022 number should be read as indicative. 6 **Nothing here is a loss-severity forecast.** Recovery is one input to severity. The other is how much of the balance is owed at default relative to origination, which these cuts hold fixed by construction. A full severity curve by vintage is the next piece, and it needs the same tape. Falsifiable The reading fails if, in the January 2027 tapes, the 2024 vintage at either lender recovers within five points of the 2022 vintage on cars sold in the same quarter. It also fails if Santander's within-vintage recoveries turn up in the second half of 2026 while its pooled figure stays flat, which would mean the pooled line was telling the truth after all. **Both checks run on the next six monthly filings, with the construction given in full below.** ## Proof: every figure, traced Computed means the figure was calculated from the Exhibit 102 loan tapes on file with the SEC and the construction is stated. Verified means read at the named external document. Derived means arithmetic on figures above it. | Claim as stated | Source and construction | Status | |---|---|---| | Data: 48 public auto ABS trusts, monthly Exhibit 102 tapes January 2021 to July 2026; this brief uses SDART 2021-1, 2022-1, 2023-6, 2024-1, 2025-1; EART 2021-1 through 2025-1; CRVNA 2021-P1, 2022-P1, 2024-P2, 2025-P1 | SEC EDGAR Form ABS-EE filings by each issuing entity; fields per 17 CFR 229.1125 Item 3: repossessedProceedsAmount (3(k)(1), "net of repossession fees and expenses"), originationDate, originalLoanAmount, vehicleValueAmount (3(d)(6), "the value of the vehicle at the time of origination"), reportingPeriodBeginningLoanBalanceAmount | Verified | | Recovery construction: proceeds summed per loan; denominator is the largest beginning balance in the three reporting periods ending at the first proceeds period; rows above 150% dropped (4 at Santander, 6 at Exeter); group figures dollar-weighted | Stated method. Robustness: using instead the balance in the proceeds period, or the balance when the repossessed flag first appears, changes vintage levels by one to three points and the ordering not at all; see the robustness row | Computed | | Counts: Santander 30,035 and Exeter 20,830 repossessions with proceeds (50,865); Carvana P 2,024 | Tape, as above | Computed | | Same-window table: Exeter 29.7 / 22.5 / 24.7 / 32.8 / 47.3% and Santander 48.6 / 40.1 / 34.6 / 46.5 / 57.0% for 2020 to 2024 vintages, sales July 2025 to June 2026, with the counts shown | Tape, sale period filtered to 2025Q3 through 2026Q2 | Computed | | Santander pooled recovery by sale year 69.6 / 57.6 / 51.3 / 50.4 / 50.0 (2022 to 2026) | Tape, all vintages, by year of first proceeds | Computed | | Santander by vintage and sale year: 2020 paper 77.3, 64.7, 52.1, 49.1, 52.4; 2021 paper 62.3, 54.7, 44.5, 40.2; 2022 paper 44.9, 37.1, 34.9; 2023 paper 55.6, 49.6, 45.5; 2024 paper 58.6, 55.6 | Tape, cells of 40 or more sales | Computed | | Santander repossessed balance by vintage, 2026: 2023 paper 49%, 2024 paper 47%, 2020 to 2022 paper 4% combined; 2024 paper was 0% in 2024 and 25% in 2025 | Tape, balance-weighted shares | Computed | | "Two to three points" counterfactual: 2026 vintage recoveries at 2025 mix weights give 47.5% against 50.0% pooled | Sum of 2026 vintage rates times 2025 balance shares (2021 paper held at its 2025 rate, having no 2026 cell) | Derived | | Exeter pooled 36.8% in 2025 and 37.0% in 2026, with every vintage lower in 2026: 2021 paper 24.6 to 22.0; 2022 paper 26.0 to 24.5; 2023 paper 35.5 to 31.5; 2024 paper 49.7 to 45.6 | Tape; 2023 and 2024 sale years excluded for Exeter (527 and 257 proceeds events against thousands of repossession flags) | Computed | | Exeter vintage by LTV grid, all 25 cells as printed; Santander 2022 row 50.9 / 47.4 / 39.9 / 38.5 / 32.2 and 2024 row 69.8 / 60.3 / 56.8 / 54.6 / 48.1 | Tape; LTV = originalLoanAmount / vehicleValueAmount; buckets at 100, 110, 120, 130% | Computed | | Within one trust: SDART 2025-1 recovers 34.7% on 2022 paper, 43.2% on 2023, 57.1% on 2024; EART 2025-1 recovers 47.5% on 2024 paper against EART 2023-1 at 25.8% on 2022 paper | Tape, per-trust cells of 50 or more | Computed | | 2020 vintage best at both lenders: 69.4% Santander (7,113 sales), 58.2% Exeter (5,446) | Tape, all sale dates | Computed | | Carvana P same window: 2021 paper 35.5%, 2022 paper 37.3%, 2024 paper 47.2%; LTV at signing median 1.00 to 1.04 | Tape | Computed | | Share of loans written above 120% LTV: Santander 2022 vintage 55% against 38 to 42% in other years; Exeter 2024 vintage 55%, its highest | Tape. Stated because it cuts both ways: Santander's 2022 cohort does carry more rolled-in balance, while Exeter's best-recovering vintage carries the most | Computed | | Robustness: under three denominators (balance in the proceeds period; largest balance in the trailing three periods, the published basis; balance when the repossessed flag first appears) Santander's 2022 vintage reads 39.3 / 39.1 / 39.4% and its 2024 vintage 57.6 / 57.1 / 57.1%; Exeter's 2022 vintage 29.2 / 29.1 / 29.1% and 2024 vintage 47.9 / 47.5 / 47.6%. On the alternative ratio, recoveries over charged-off principal, 2022 is the worst vintage at all four servicers: AmeriCredit 42.0% (2020 paper 60.2%), GM Financial 54.3% (73.3%), Santander 34.1% (59.1%), Exeter 31.8% (44.4%) | Tape; three denominators and the charge-off ratio computed on the same loan sets | Computed | | Fitch: 2022–2023 vintages "most acute"; underwriting tightened via "lower loan to value caps"; subprime recoveries "did not meaningfully benefit from tariff related used vehicle demand in early 2025, as collateral pools typically consist of older, higher mileage vehicles"; prime TTM recovery 62.27%, peak 72.08% May 2025; delinquency "composition effects as stronger pre-pandemic vintages amortized" | Fitch and Black Book, Vehicle Depreciation Report 2026 , text extracted from the PDF | Verified | | S&P: 2023 vintage CNL 7.30% at month 17 against 7.99% (2022) and 8.00% (2008); 2023 and 2024 vintages "lower cumulative net losses than 2022's near-record-high levels" | Auto Remarketing, May 15, 2025 , reporting S&P Global Ratings | Verified | | CFPB: 11.6% of 2018–2022 vehicle loans carried negative equity; mean LTV 119.3% with negative equity against 88.9% (positive trade-in) and 101.6% (no trade-in); "more than twice as likely to have their account assigned to repossession within two years" | CFPB, Negative Equity in Auto Lending, June 2024 , text extracted from the PDF | Verified | | TransUnion: used-vehicle originating LTV 125 in Q1 2023, 110 in Q1 2022, 112 in Q1 2020 | TransUnion, June 20, 2023 | Verified | | Fitch subprime recovery index 32.64% end-2025, 43.73% pre-pandemic average; 39.5% June and 38.0% July 2026; tax-season improvement "expected to be short lived" | Fitch and Black Book, Vehicle Depreciation Report 2026 ; Auto Remarketing, August 24, 2026 | Verified | | Manheim 257.7 peak at end-2021; 206.2 mid-September 2026; December 2023 "about 33% higher than at the end of 2019" | Cox Automotive releases: December 2024 , mid-September 2026 , December 2023 | Verified | | CPS recovery rate 28.5% Q4 2025 to 33.3% Q2 2026; vintage recoveries 22 / 25 / 37.5 / 47.1% for 2022 to 2025 as spoken by COO Mike Lavin; older vintages to "flush out" | CPS 8-K Exhibit 99.1, Q2 2026 ; Q2 2026 call transcript | Verified | | Exeter "ceased reporting nearly all repossessions on form ABS-EE exhibit 102" from November 2022; this brief's own counts by year of first flag / first proceeds: 2021 2,420 / 1,743; 2022 6,172 / 5,846; 2023 725 / 896; 2024 373 / 428; 2025 25,370 / 10,121; 2026 (to July) 7,353 / 7,109 | Bill Ploog, CUCollector, September 20, 2025 ; tape counts computed here | Verified | Reproduction: the panel is built by the tape pipeline described in Issue 13; the analysis script for this brief groups proceeds events by trust and loan, applies the denominator rule above, and emits every table printed here. Earlier drafts of this analysis used CarMax as a second prime control; its trusts report proceeds on a post-charge-off stub balance in most cases, 5,514 of 6,524 events failed the 150% screen, and it was dropped rather than patched. **Sources & notes** **Loan-level data.** SEC Form ABS-EE, Exhibit 102 asset-level files, filed monthly by Santander Drive Auto Receivables Trusts 2021-1, 2022-1, 2023-6, 2024-1 and 2025-1; Exeter Automobile Receivables Trusts 2021-1 through 2025-1; Carvana Auto Receivables Trusts 2021-P1, 2022-P1, 2024-P2 and 2025-P1; and, for the alternative ratio only, AmeriCredit Automobile Receivables Trusts 2021-1 through 2024-1 and GM Financial Consumer Automobile Receivables Trusts 2021-1 through 2025-1. Field definitions per 17 CFR 229.1125, Item 3. **Indices and issuers.** Fitch Ratings and Black Book, Vehicle Depreciation Report 2026. Fitch mid-year commentary via Auto Remarketing, August 24, 2026. S&P Global Ratings vintage commentary via Auto Remarketing, May 15, 2025. CFPB, Negative Equity in Auto Lending (June 2024). TransUnion, June 20, 2023. Cox Automotive Manheim Used Vehicle Value Index releases. Consumer Portfolio Services Q2 2026 release and call. Bill Ploog, CUCollector, September 20, 2025, on the Exeter reporting break. **Revision note.** This is the second version of Issue 15. The first, published October 2, 2026 as "Set at signing," made the vintage argument from Fitch, Cox and one issuer's spoken figures and said in its Limits that the mechanism was inferred and that the direct test belonged in the tape. This version is that test. The external figures carried over were re-read at source; nothing from the first version was retained on trust. 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 5, 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. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html) LR LendRisk Analytics Independent market research Related [Issue 13 Current, on tape.](https://lendriskanalytics.com/insights/current-on-tape.html) [Issue 8 Current, on paper.](https://lendriskanalytics.com/insights/current-on-paper.html) --- title: "What the tape said: modified, and not paying" url: https://lendriskanalytics.com/insights/modified-and-not-paying.html publisher: LendRisk Analytics series: What the Tape Said issue: 14 published: 2026-09-15 kind: Research brief description: "Every federally insured credit union files one line for loans it has modified for borrowers in trouble, and since 2024 a second line for the modified loans that are already late again. The first has more than doubled since the definition changed. One dollar in four on it is not paying. What the line says about next year's charge-offs, where it says nothing, and who leaves it blank. Every figure rebuilt from the raw NCUA archives; inferences labeled." html: https://lendriskanalytics.com/insights/modified-and-not-paying.html --- # What the tape said: modified, and not paying [← All insights](https://lendriskanalytics.com/articles.html) What the Tape Said · Issue 14 Research brief · 13 min read Credit unions · NCUA 5300 · Loan modifications # Modified, and not paying. NCUA 5300 call reports, March 2023 to March 2026 · Public record through September 15, 2026 · LendRisk Analytics Issue 13 read the extension field on securitized auto loans and found that every lender resets the clock, and that what happens in the six months after is the number that separates them. Credit unions file something stricter. One line on the 5300 carries every loan a credit union has modified for a borrower in financial difficulty, held on the line for twelve months. Since 2024 a second line carries the part of that book that is already late again under its new terms. So the filing answers the question the securitization tapes made me work for. The modified book has more than doubled since the definition changed in 2023. One dollar in four on it is not paying. And the line tells you something specific, narrower than I expected, about which delinquent books are going to lose money next year. $10.2B Loans modified for borrowers in difficulty, March 2026, up from $4.4B when the definition changed 25% Of that balance already 30+ days late or in nonaccrual under the modified terms 3.2× Next-year charge-offs, top quartile of modifiers against credit unions reporting none 9 of 9 Annual cohorts since the definition changed with the same ordering at the top ## Bottom line Here is the exercise. Take every federally insured credit union with at least $10 million of loans in both March 2025 and March 2026. That is 2,942 of them. Rank the ones reporting a nonzero modified balance by that balance as a share of total loans, cut them into quartiles, and put the 962 reporting nothing in their own group. Then wait a year and measure net charge-offs on the whole book. | Group, March 2025 | n | Median modified / loans | Delinquency, March 2025 | Charge-offs, year to March 2025 | Charge-offs, year to March 2026 | |---|---|---|---|---|---| | Reports none | 962 | 0.000% | 0.548% | 0.421% | 0.416% | | Quartile 1 | 495 | 0.026% | 0.566% | 0.559% | 0.615% | | Quartile 2 | 495 | 0.098% | 0.674% | 0.661% | 0.604% | | Quartile 3 | 495 | 0.230% | 0.765% | 0.685% | 0.692% | | Quartile 4 | 495 | 0.799% | 1.168% | 1.375% | 1.325% | Charge-offs and delinquency are dollar-weighted within group: total net charge-offs over total loans, annualized. Quartiles are set among reporters; the top quartile begins at about 0.44% of loans. Every figure in this brief was rebuilt from the raw NCUA archives for publication. Read the two right-hand columns together. The top quartile charged off 1.325% over the following year, three times the non-reporters. It had also charged off 1.375% in the year before it was ranked. So this is not the finding from Issue 12, where a full lot predicted losses a board could not yet see. The modified line does not run ahead of the losses. It sits on top of them. The question is what it adds once you already know that, and the answer is in Section III. The read The 5300 modification line is a workout book, filed quarterly, with its own redefault rate attached. Twelve months of modifications, and the share already late again. **Three things are true about it at once.** It has more than doubled in three years. It is concentrated in a few hundred institutions. And in a book whose delinquency is already above the median, the size of it tells you which credit unions lose twice as much as their peers. ## I · What the line is, and when it changed Schedule A, Section 2, item 26 of the [Form 5300](https://ncua.gov/files/publications/regulations/call-report-form-march-2025.pdf): Account 1001F, total outstanding troubled debt restructured loans or modifications to borrowers experiencing financial difficulty. The [instructions](https://ncua.gov/files/publications/regulations/call-report-instructions-september-2025.pdf) say what goes there for a credit union that has adopted CECL: "the number and amortized cost of loan modifications resulting in a new loan or a continuation of the current loan. Report modifications for 12 months from the modification date or until the loan is paid off, charged-off, sold, or otherwise settled. Any subsequent modification resets the timeline." Modifications count if they take the form of "principal forgiveness, an interest rate reduction, a significant payment delay, or a term extension (or a combination thereof)." Item 27, Accounts DL0148 and DL0149, added in March 2024: "the number and dollar amount of modified loans to borrowers experiencing financial difficulty and, under their modified repayment terms, are past due 30 days or more or are in nonaccrual status." That is a redefault line. The securitization tapes in Issue 13 made me follow each extended loan forward six months to get it. Here the credit union files it. Before CECL the same account held troubled debt restructurings under the old accounting definition. [ASU 2022-02](https://dart.deloitte.com/USDART/home/publications/deloitte/heads-up/2022/fasb-issues-asc-326-update) removed the TDR concept for CECL adopters, most credit unions adopted at the start of 2023, and the system total dropped from $7.3 billion in December 2022 to $4.4 billion in March 2023 as the definition changed underneath it. Nothing in this brief compares a number after that line to a number before it. Chart 1 · The modified book since the definition changed Account 1001F summed across all federally insured credit unions, March 2023 to March 2026. From March 2024 the dark portion is Account DL0149, the modified balance already 30+ days late or in nonaccrual. $4.41B at March 2023, $10.95B at December 2025, $10.15B at March 2026. As a share of total loans: 0.288% to 0.587%. System delinquency over the same span: 0.525% to 0.845%. Source files: FS220H and FS220P in each quarterly archive at [ncua.gov](https://ncua.gov/analysis/credit-union-corporate-call-report-data/quarterly-data). Two and a half times in three years, against delinquency up 1.6 times. The December 2025 reading is the highest for this account since the series began in 2016, and the only readings above it were TDR balances under the old, broader definition. Whether that is more members in trouble or more credit unions working with them is not something the line can settle. What it can settle is what came next for the loans on it. ## II · One in four is late again Of the $10.15 billion modified at March 2026, $2.56 billion was already 30 or more days past due or in nonaccrual under the modified terms. That is 25.2%. It was 22.9% in the first quarter the line existed and 27.4% at the December 2025 peak. It runs with size: 8.6% of the modified balance at credit unions under $50 million, 21.3% at $1 to $10 billion, 28.5% above $10 billion. | Asset band, March 2026 | Reporting a modified balance | Modified | Not in compliance | Share | |---|---|---|---|---| | Under $50M | 497 | $48M | $4M | 8.6% | | $50 to 100M | 327 | $75M | $10M | 14.0% | | $100 to 200M | 369 | $111M | $14M | 12.9% | | $200 to 500M | 380 | $280M | $42M | 14.9% | | $500M to 1B | 240 | $360M | $70M | 19.4% | | $1 to 10B | 409 | $3,219M | $686M | 21.3% | | Over $10B | 24 | $6,062M | $1,729M | 28.5% | Accounts 1001F and DL0149, credit unions reporting a nonzero modified balance, March 2026 archive. Twenty-four credit unions above $10 billion hold 60% of the modified dollars and 68% of the non-compliant ones. Put that next to Issue 13. On the securitized subprime shelves, 29% of loans extended while under 60 days late were back at 60+ within six months. Here, at any given quarter, 25% of the modified book is 30+ or nonaccrual. Different clocks, different thresholds, and the credit union line only counts borrowers already in difficulty, so the two are not one number. But they land in the same place. A quarter of the loans given a second set of terms are not keeping them. **Inference** The non-compliance share rising with size is probably two things. Big credit unions carry more indirect auto and unsecured paper, which is where modifications concentrate. And small credit unions know the member, so the modification they grant is more often the one that works. The filing cannot separate those. The number a small credit union should take from the table is that the system's 25% is not its number; its band runs at 9 to 15%. ## III · Where the line says something, and where it does not Split the March 2025 cohort at the median of reported delinquency, 0.569% of loans. Inside each half, rank the reporters by modified share into quartiles again. Then look at next year's charge-offs. Chart 2 · Modified share only matters in the half that is already delinquent March 2025 cohort, n=2,942, split at the median delinquency ratio. Bars are net charge-offs in the year to March 2026, dollar-weighted within group. Low-delinquency half, non-reporters through quartile 4: 0.323 / 0.354 / 0.460 / 0.416 / 0.445%. High-delinquency half: 0.601 / 0.866 / 0.750 / 0.894 / 1.508%. The same split on all nine annual pairs gives quartile 4 in the low half between 0.39% and 0.45% every time, and quartile 4 in the high half between 1.44% and 1.71%. In the clean half, nothing. A credit union with below-median delinquency and a heavy modified line charges off about the same as one with a light line, in every one of the nine cohorts. The field does not find hidden losses in a clean book. That is the opposite of what the repossessed inventory line did in Issue 12, and it is worth saying plainly rather than burying, because the masking story from the securitization tapes does not carry over to this filing. In the delinquent half, a lot. Quartile 4 charges off 1.51% against 0.75 to 0.89% for the other three quartiles and 0.60% for non-reporters. Twice the losses of a peer with the same delinquency profile and a lighter workout book. That holds in all nine pairs. **Inference** The reason the two lines behave differently is what they count. Repossessed inventory is a physical fact that arrives before the charge-off. The modified line is a decision the credit union already made about a borrower it already knows is in trouble. So in a clean book a small modified line is just good servicing. In a delinquent book a large one is the size of the problem the credit union has already admitted to, and it scales the loss. The line is a marker, not a lead. ## IV · Nine cohorts, one ordering at the top | Annual pair | n | Reports none | Q1 | Q2 | Q3 | Q4 | Q4 / none | |---|---|---|---|---|---|---|---| | Mar 2023 to Mar 2024 | 3,090 | 0.375% | 0.689% | 0.603% | 0.665% | 1.274% | 3.4x | | Jun 2023 to Jun 2024 | 3,087 | 0.408% | 0.535% | 0.732% | 0.692% | 1.231% | 3.0x | | Sep 2023 to Sep 2024 | 3,078 | 0.421% | 0.537% | 0.714% | 0.689% | 1.260% | 3.0x | | Dec 2023 to Dec 2024 | 3,060 | 0.433% | 0.547% | 0.658% | 0.661% | 1.342% | 3.1x | | Mar 2024 to Mar 2025 | 3,033 | 0.453% | 0.525% | 0.748% | 0.709% | 1.417% | 3.1x | | Jun 2024 to Jun 2025 | 3,011 | 0.410% | 0.499% | 0.650% | 0.709% | 1.331% | 3.2x | | Sep 2024 to Sep 2025 | 2,998 | 0.393% | 0.515% | 0.574% | 0.715% | 1.277% | 3.2x | | Dec 2024 to Dec 2025 | 2,972 | 0.408% | 0.505% | 0.591% | 0.700% | 1.258% | 3.1x | | Mar 2025 to Mar 2026 | 2,942 | 0.416% | 0.615% | 0.604% | 0.692% | 1.325% | 3.2x | Net charge-offs over the following year, dollar-weighted within group, every annual pair available since the definition change. Quartile 4 beats the non-reporters in all nine, by 3.0 to 3.4 times. Quartiles 1 through 3 sit between 0.50% and 0.75% with no consistent order among them. The ordering is only clear at the top. Below the top quartile the line does not sort anything. A modified balance under about 0.4% of loans is noise, the same lesson as the small repo number in Issue 12: a screen built on this field needs a floor, not a flag on any nonzero value. Head to head, it is the weakest of the three standard measures. Spearman correlation with next-year charge-offs, across the nine pairs: prior charge-offs +0.60 to +0.79, reported delinquency +0.38 to +0.55, modified share +0.19 to +0.23. Holding delinquency fixed, modified share keeps +0.13 to +0.17. Holding delinquency and prior charge-offs fixed, +0.05 to +0.12. It carries something of its own. Not much. It is not a size effect. Apply the top-quartile threshold inside each asset band separately and the credit unions above it charge off more than the rest of their band in all 54 band-pair cells. The narrowest is the $200 to 500 million band in the March 2024 pair, 0.500% against 0.497%, which is a tie in practice. ## V · The redefault line, credit union by credit union Account DL0149 lets you ask the Issue 13 question of each institution. Among reporters with the line filled in, 1,197 of 2,016 in the March 2026 cohort report zero non-compliance. Among the rest, the median non-compliant share is 22%. Split the reporters into those reporting zero, and the nonzero group into halves, and look at next year's charge-offs, for the five pairs the line has existed. | Annual pair | Reports zero non-compliance | Nonzero, lower half | Nonzero, upper half | Median share, upper half | |---|---|---|---|---| | Mar 2024 to Mar 2025 | 0.657% | 0.563% | 0.971% | 21.9% | | Jun 2024 to Jun 2025 | 0.602% | 0.550% | 0.913% | 20.2% | | Sep 2024 to Sep 2025 | 0.540% | 0.573% | 0.879% | 22.5% | | Dec 2024 to Dec 2025 | 0.531% | 0.567% | 0.912% | 26.6% | | Mar 2025 to Mar 2026 | 0.620% | 0.544% | 0.965% | 22.6% | Net charge-offs over the following year, dollar-weighted, credit unions with $10M+ loans reporting a nonzero modified balance and a value in DL0149. The nonzero group is split at its median non-compliant share. A credit union whose modified loans are failing at above the median rate charges off 0.88 to 0.97% the following year. One reporting zero non-compliance charges off 0.53 to 0.66%. That is the per-institution version of the Exeter-against-Honda spread in Issue 13, smaller, because the credit union line only ever counts borrowers who were already in difficulty. ## VI · Half the lines are blank, and the dollars are in a few hundred hands Of 4,250 federally insured credit unions filing March 2026, 2,246 report a nonzero modified balance. Among those with $5 million or more of loans, reporting runs from 37% under $50 million in assets to 100% above $10 billion. The small band is the least likely to fill it in and carries higher delinquency than any band below $10 billion, 0.98% against 0.71 to 0.83% in the middle bands. Same shape as the repo line. Whether a blank means no modifications or an unanswered question, the filing does not say, and every non-reporter group above is a blend of both. The dollars sit in very few places. Of the $10.15 billion, $7.18 billion is at 202 credit unions whose modified line is 1% of loans or more. $5.35 billion is at 34 credit unions at 3% or more. The top quartile in the headline cohort, 495 institutions, holds $8.37 billion, 82% of the system total, on $451 billion of loans. **Inference** For an examiner or an acquirer this is a short list. Two hundred institutions hold seven in ten modified dollars, and for each of them the filing already carries the share that is late again. For a credit union CFO it is a comparison: the band's non-compliance rate in Section II is the number to put beside your own, and the 0.4% of loans threshold in Section IV is roughly where the line starts to mean something. ## VII · What this does not show Five things, so nobody has to find them for me. 1 **The line marks losses. It does not forecast them.** The top quartile was already charging off 1.375% the year before it was ranked. Prior charge-offs outrank it on every test. Its use is inside the delinquent half of the population, and as a per-institution redefault read. 2 **A blank is read as zero.** About a third of credit unions with $10 million or more of loans report nothing. Some have nothing to report. Some did not answer. The non-reporter group is a mix and the filing cannot separate it. 3 **The series breaks at March 2023.** Before CECL the account held TDRs under a different rule. The 2016 to 2022 readings are a different measurement and are not compared to anything after the change. 4 **The nine cohorts overlap.** The same institutions appear in most pairs. Nine repetitions show the result is stable through time, not that it has been confirmed nine independent times. 5 **Whole-book charge-offs, not vehicle.** Modifications are filed for the whole book, so the outcome is the whole book's net charge-offs. Issue 12 tested a vehicle field against vehicle charge-offs. The two briefs are not comparable line for line. One outside reference point, for context rather than support. The [Philadelphia Fed](https://www.philadelphiafed.org/-/media/FRBP/Assets/Consumer-Finance/Reports/cfi-report-april-2026-do-recent-auto-loan-delinquency-rates-overstate-borrower-distress.pdf) found in April that the stock of severe auto delinquency is rising while the inflow of newly delinquent borrowers is stable, and pointed at loss-mitigation practice as the likely reason. The bureau data it used cannot see a modification. This filing can, and the picture it gives is a workout book growing faster than delinquency, with a quarter of it already failing its new terms. Falsifiable This fails if, on the March 2026 to March 2027 pair, the top quartile of modifiers charges off less than twice the non-reporters, or if quartile 4 inside the high-delinquency half stops exceeding the other three quartiles. Both tests are mechanical, the accounts are named above, and the archives are public. **Anyone with the files can run it.** ## Proof: every figure, traced All figures were rebuilt from the raw NCUA call report archives for this publication. Recomputed means the number was rebuilt from the quarterly files and matched. Derived means calculated from recomputed figures, arithmetic shown. Verified means read straight from the cited public document. | Claim as stated | Source | Status | |---|---|---| | Account 1001F definition: modifications for borrowers experiencing financial difficulty, held for 12 months, in the four listed forms; DL0149: modified loans 30+ days past due or in nonaccrual under modified terms; TDR reporting before CECL adoption | NCUA Form 5300 instructions, effective September 30, 2025 , Schedule A Section 2 items 26 and 27; Form 5300, March 2025 | Verified | | TDR concept removed for CECL adopters by ASU 2022-02, effective for fiscal years beginning after December 15, 2022 | Deloitte Heads Up on ASU 2022-02 | Verified | | System modified balance $7.29B at 2022-12, $4.41B at 2023-03, $10.95B at 2025-12, $10.15B at 2026-03; 0.288% of loans at 2023-03, 0.587% at 2026-03; delinquency 0.525% to 0.845% | FS220H.txt field ACCT_1001F summed over CU_TYPE 1 and 2 in each quarterly archive from ncua.gov ; loans and delinquency from FS220A and FS220 totals in the same archives | Recomputed | | 2.5x and 1.6x | 0.587 / 0.288 = 2.04 on the loan-share basis; 10.15 / 4.41 = 2.30 in dollars, 10.95 / 4.41 = 2.48 at the December peak; delinquency 0.845 / 0.525 = 1.61. "Two and a half times" refers to the December 2025 peak | Derived | | Non-compliant $2.56B at 2026-03, 25.2%; 22.9% at 2024-03; 27.4% at 2025-12; by band 8.6 / 14.0 / 12.9 / 14.9 / 19.4 / 21.3 / 28.5%; 24 credit unions above $10B hold $6.06B modified and $1.73B non-compliant | FS220P.txt field ACCT_DL0149 against ACCT_1001F, same archives | Recomputed | | 4,250 federally insured credit unions at 2026-03, 2,246 reporting a nonzero modified balance; reporting share by band 37% to 100%; small band delinquency 0.98% | FOICU.txt CU_TYPE 1 and 2; matches NCUA's published first-quarter 2026 count | Recomputed | | Headline cohort n=2,942, groups 962 / 495 / 495 / 495 / 495; charge-offs 0.416 / 0.615 / 0.604 / 0.692 / 1.325%; prior-year 0.421 / 0.559 / 0.661 / 0.685 / 1.375%; delinquency 0.548 / 0.566 / 0.674 / 0.765 / 1.168%; median modified share 0.000 / 0.026 / 0.098 / 0.230 / 0.799%; quartile 4 threshold 0.441% | Charters with $10M+ loans at both 2025-03 and 2026-03; net charge-offs from ACCT_550 less ACCT_551, annualized by cycle month, over loans | Recomputed | | 3.2x | 1.325 / 0.416 = 3.19 | Derived | | Nine annual pairs, 2023-03 through 2025-03 starts: quartile 4 above non-reporters 9 of 9, ratio 3.0x to 3.4x; quartiles 1 to 3 between 0.50% and 0.75%; Spearman modified +0.19 to +0.23, delinquency +0.38 to +0.55, prior charge-offs +0.60 to +0.79; partials +0.13 to +0.17 and +0.05 to +0.12 | Same construction on every pair; replication table in the working files (annual_pairs.csv, band_check.csv) | Recomputed | | Delinquency split, March 2025 pair: low half 0.323 / 0.354 / 0.460 / 0.416 / 0.445%, high half 0.601 / 0.866 / 0.750 / 0.894 / 1.508%, median delinquency 0.569%; across nine pairs low-half quartile 4 0.39 to 0.45%, high-half quartile 4 1.44 to 1.71% | Cohort split at median delinquency ratio; quartiles re-cut inside each half (dq_split.csv) | Recomputed | | Top group above the rest of its band in 54 of 54 band-pair cells; the narrowest is $200-500M, 2024-03 pair, 0.500% against 0.497% | Cohort-wide quartile 4 threshold applied within each of six asset bands, nine pairs | Recomputed | | Non-compliance terciles: zero group 0.53 to 0.66%, upper half 0.88 to 0.97%, median share in the upper half 20 to 27%; 1,197 of 2,016 reporters at zero | DL0149 / 1001F per charter, five pairs from 2024-03 (noncompliance_pairs.csv) | Recomputed | | Concentration: 202 credit unions at 1%+ hold $7.18B; 34 at 3%+ hold $5.35B; top quartile 495 institutions hold $8.37B, 82%, on $451B of loans | 2026-03 archive, charters with $10M+ loans | Recomputed | | Philadelphia Fed: stock of severe delinquency rising while inflow is stable; loss-mitigation practice as a candidate explanation; bureau tradeline data | Cheney, Hunt, Lambie-Hanson, Santucci and Zhou, April 2026 | Verified | | Issue 13 figure: 29% of subprime extensions granted under 60 DPD were 60+ again within six months | Issue 13 , Section IV | Verified | **Sources & notes** **Data.** NCUA 5300 call report quarterly archives, cycles 2016-03 through 2026-03, downloaded from ncua.gov. Files used: FOICU.txt for charter type (federally insured only, CU_TYPE 1 and 2), FS220H.txt for Accounts 1000F and 1001F, FS220P.txt for Accounts DL0148 and DL0149, FS220 and FS220A for loans and delinquency, FS220I for charge-offs and recoveries, AcctDesc.txt for field definitions. The 2016 to 2022 readings are TDR balances under the pre-CECL definition and appear only in the sentence that says so. **Method.** Annual pairs join each start quarter to the same quarter one year later on charter number. Cohort floor $10 million of loans at both ends. A blank modified balance is read as zero. Charge-offs are net of recoveries and annualized by cycle month. Group rates are dollar-weighted. Quartiles are set among reporters only; non-reporters are their own group. Correlations are Spearman; partials by rank regression residuals. The delinquency split uses the cohort median of total delinquent loans over total loans. Every figure was rebuilt from the raw archives immediately before publication, and the notebook that does it is kept with the working files. **Companion issues.** [Issue 13](https://lendriskanalytics.com/insights/current-on-tape.html) reads the extension field on securitized auto loans and follows extended loans forward six months. [Issue 12](https://lendriskanalytics.com/insights/the-lot-behind-the-branch.html) reads the repossessed inventory line on the same filing as this one. Where this brief goes beyond what the filings state, it is labeled as an inference. Point-in-time reading of public filings through September 15, 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. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html) LR LendRisk Analytics Independent market research Related [Issue 13 Current, on tape.](https://lendriskanalytics.com/insights/current-on-tape.html) [Issue 12 The lot behind the branch.](https://lendriskanalytics.com/insights/the-lot-behind-the-branch.html) --- title: "What the tape said: current, on tape" url: https://lendriskanalytics.com/insights/current-on-tape.html publisher: LendRisk Analytics series: What the Tape Said issue: 13 published: 2026-09-13 kind: Research brief description: "Fourteen auto lenders, forty-four securitizations, 1.15 million loans on their latest monthly tapes. Every one of them grants payment extensions, and every extension turns a past-due account current. Reported 60+ delinquency against the same figure with recently extended loans added back, lender by lender, with what happened to those loans six months later. Every figure recomputed from the filed loan-level records; inferences labeled." html: https://lendriskanalytics.com/insights/current-on-tape.html --- # What the tape said: current, on tape [← All insights](https://lendriskanalytics.com/articles.html) What the Tape Said · Issue 13 Research brief · 15 min read Auto ABS · Regulation AB II loan tapes · Payment extensions # Current, on tape. Form ABS-EE loan-level filings, January 2021 to July 2026 · Public record through September 13, 2026 · LendRisk Analytics Issue 8 walked through how an extension works. The servicer moves a missed payment to the end of the loan, nothing is due next month, and the account reports as current. It also said nobody adds up the field that records this. So I did. Forty-four registered auto securitizations, fourteen lenders, one record per loan per month, 1.15 million loans on the latest tapes. Every one of those lenders grants extensions. Prime captives included. What separates them is how much of the book has been through one, and what the loan does in the six months after. 1,845 bps What extensions took out of the subprime 60+ number, July 2026 318 bps Same thing on the prime shelves, on top of a reported 0.73% 42% Of Exeter loans extended while under 60 days late are 60+ again within six months 6.1× AmeriCredit's gap against GM Financial's own prime shelf, same servicer, same month ## Bottom line Here is the whole exercise. Take the July 2026 tape for every registered auto deal I hold, twenty-nine of them. For each loan, read two fields: the days past due the servicer reports, and the number of months the loan was extended that period. Work out 60+ delinquency the way every index does, weighted by balance. Then work it out again with one change. Any loan that is under 60 days late today but got an extension in the last six months gets counted as delinquent. The second number is not a prediction. It is the most the extension mechanism could have taken out of the first. | Shelf tier, July 2026 tapes | Deals | Loans | Reported 60+ | Adjusted 60+ | Gap | Extended in last 6 months | Extended ever | |---|---|---|---|---|---|---|---| | Prime | 19 | 629,959 | 0.73% | 3.91% | 318 bps | 3.3% | 8.5% | | Subprime | 10 | 253,145 | 8.10% | 26.55% | 1,845 bps | 21.0% | 45.2% | Balance-weighted across the deals in each tier. Prime: Ally, CarMax, Carvana, Ford, GM Financial, Honda, Hyundai, Nissan, Toyota, Volkswagen, World Omni. Subprime: AmeriCredit, Bridgecrest, Exeter, Santander Drive. "Extended ever" is the share of today's balance sitting in loans that have at least one extension anywhere on the tape. Every number here was rebuilt from the filed EX-102 records before publishing. On the subprime shelves, one dollar in five of what is outstanding today was extended in the last six months. Nearly half has been extended at some point. The reported delinquency number sits on top of that. The prime shelves run the same mechanism at about a sixth of the scale. The read The 60+ number in a servicer report, a rating agency index, or a headline is the count after the servicer has already decided which accounts to reset. On every shelf here, that decision mostly happens before the account ever shows as late. More than nine in ten extensions go to loans fewer than 30 days past due. **This is a piece about follow-through, not concealment.** The field that records the reset is public. The six months after it tell you whether the reset held. ## I · What the field says, and how two ways of filing it nearly hid half the market [Item 1125 of Regulation AB](https://www.law.cornell.edu/cfr/text/17/229.1125) lists what a registered auto deal has to file about every loan, every month. Item 3(f)(25) is days past due. Item 3(f)(6) is the payment due next period. Item 3(j) only applies "if the loan has been modified from its original terms," and inside it, Item 3(j)(2) asks for "the number of months the loan was extended during the reporting period." Read that phrase twice. It is a per-month entry, not a running total, and it only has to be there when something was modified. Santander says the reset out loud in its own filing. The [EX-103 narrative](https://www.sec.gov/Archives/edgar/data/1383094/000119312526352812/sdart241ex103.xml) attached to Santander Drive 2024-1's July 2026 tape defines the next payment due as the amount needed "for the receivable to be considered current," and then: "If 'next reporting period payment amount due' is reported as 0.00, no interest or principal is due in the next reporting period for the receivable to be considered current because the obligor either made a payment in advance or was granted a payment extension." [Exeter](https://www.sec.gov/Archives/edgar/data/2005087/000092963826003306/eart2024-1_exhibit103.xml) and [AmeriCredit](https://www.sec.gov/Archives/edgar/data/2020251/000202025126000030/exh103loanv2.xml) file nearly the same sentence. Issuers fill in Item 3(j)(2) two different ways, and that difference is why this brief looks the way it does. Exeter, AmeriCredit, Santander, GM Financial and Hyundai write the field on every loan every month. Zero when nothing happened, a 1 or a 2 in the month an extension was granted, zero again after. Ally, CarMax, Carvana, Ford, Honda, Nissan, Toyota, Volkswagen, World Omni and Bridgecrest leave the tag out entirely unless something was modified, and fill it in only in that month. Both are fine under the rule. They are not the same to a parser. The desk I built in August looked for an increase over the prior month's value. For the second group there is no prior value, so the increase came out as nothing, and those shelves read as extension-free. The first draft of this piece was going to say prime lenders do not extend. That was wrong and the mistake was mine. The rule that holds up is the plain one: a positive value in the month is an extension in the month. **Inference** I am putting the correction in the body rather than a footnote because it is the same lesson as the article. A field can be public, complete, and filed correctly and still read as empty if you assume one filing convention. Anyone building an extension series across issuers has to check how each one writes the field before comparing two of them. One narrative does not match its tape. Carvana's [EX-103 for CRVNA 2024-P2](https://www.sec.gov/Archives/edgar/data/1999671/000199967126000032/crvna24p2ex103.xml) lists Item 3(j)(2) as "No response," yet the filed EX-102 carries values of one to three months on a few hundred loans each period, next to modification type codes. I used the values. Honda's [narrative](https://www.sec.gov/Archives/edgar/data/890975/000119312526361400/harot251ex103_0821-1606.xml) says modification type "4" is an extension and that Item 3(j)(2) can go negative when a prior extension is reversed. AmeriCredit says the same. Negative values were treated as no extension. ## II · Same servicer, two shelves GM Financial services a prime shelf and the AmeriCredit subprime shelf. It files both with explicit zeros, and it filed July 2026 tapes for both on the same schedule. Whatever is different between the two, it is not the servicer, the filing style, or the month. Chart 1 · One servicer, two shelves, July 2026 [GM Financial Consumer Automobile Receivables Trust 2024-1](https://www.sec.gov/Archives/edgar/data/2003007/000200300726000027/0002003007-26-000027-index.htm) and [AmeriCredit Automobile Receivables Trust 2024-1](https://www.sec.gov/Archives/edgar/data/2020251/000202025126000030/0002020251-26-000030-index.htm), July 2026 EX-102 tapes. Balance-weighted 60+ delinquency, as reported and with loans extended in the last six months added back. GMCAR 2024-1: 23,608 loans outstanding, 4.3% of balance extended in the last six months, 10.9% ever. AMCAR 2024-1: 34,667 loans, 26.5% in the last six months, 56.2% ever. Weighted average origination scores from each deal's first tape: 775 and 587. Six-month redefault after an extension, all live deals on each shelf: 11.0% prime, 18.5% AmeriCredit. The prime shelf reports 0.55% and reads 4.64% adjusted. The subprime shelf reports 3.86% and reads 28.69%. Reported, the two are seven times apart. Adjusted, six times. So the extension mechanism is not what makes subprime look worse than prime. It is what makes subprime look better than it is. On that shelf, for every dollar reported 60+ there are more than six dollars that were extended in the last six months and are not. ## III · Fifteen lenders, one ledger Chart 2 · Reported against extension-adjusted 60+ delinquency, by lender Latest July 2026 tape of every live deal, balance-weighted within lender. Dark bar is reported. The full bar adds back loans extended in the last six months. Subprime shelves in red. Twenty-nine deals, 883,104 loans, $15.6 billion outstanding. Six-month lookback throughout. Section V shows three and twelve. | Lender, July 2026 | Deals | Reported 60+ | Adjusted 60+ | Gap, bps | Extended, 6 mo | Extended, ever | 60+ within 6 mo of extension | |---|---|---|---|---|---|---|---| | Exeter | 4 | 9.27% | 34.90% | 2,562 | 30.2% | 53.7% | 41.7% | | AmeriCredit (GM Financial) | 2 | 3.92% | 26.80% | 2,288 | 24.4% | 58.2% | 18.5% | | Bridgecrest (DriveTime) | 1 | 11.23% | 29.15% | 1,791 | 22.9% | 41.2% | 42.4% | | Santander Drive | 3 | 8.75% | 19.91% | 1,117 | 12.3% | 33.3% | 24.0% | | CarMax | 3 | 1.82% | 6.82% | 500 | 5.3% | 16.0% | 19.6% | | Ford Credit | 1 | 0.28% | 5.18% | 490 | 5.0% | 9.2% | 5.0% | | Ally | 3 | 0.97% | 4.81% | 384 | 4.0% | 10.3% | 8.4% | | GM Financial prime | 2 | 0.46% | 4.05% | 360 | 3.8% | 8.9% | 11.0% | | Hyundai | 1 | 0.43% | 3.94% | 351 | 3.6% | 8.2% | 6.9% | | World Omni | 2 | 0.76% | 4.15% | 339 | 3.5% | 8.9% | 7.5% | | Carvana | 3 | 1.88% | 5.23% | 336 | 3.7% | 11.2% | 23.4% | | Toyota | 1 | 0.41% | 2.29% | 188 | 1.9% | 4.8% | 10.6% | | Volkswagen | 1 | 0.32% | 2.10% | 179 | 1.8% | 4.8% | 8.9% | | Nissan | 1 | 0.09% | 1.42% | 133 | 1.3% | 3.0% | 4.5% | | Honda | 1 | 0.33% | 1.45% | 112 | 1.2% | 3.7% | 4.1% | Last column: of loans that were under 60 days late when extended, the share that hit 60+ in the following six months, counted across every live deal on the shelf rather than July's alone. Filing indexes for the July 2026 tapes: [Exeter 2024-1](https://www.sec.gov/Archives/edgar/data/2005087/000092963826003306/0000929638-26-003306-index.htm), [2025-1](https://www.sec.gov/Archives/edgar/data/2049379/000092963826003311/0000929638-26-003311-index.htm); [AmeriCredit 2023-1](https://www.sec.gov/Archives/edgar/data/1963240/000196324026000029/0001963240-26-000029-index.htm), [2024-1](https://www.sec.gov/Archives/edgar/data/2020251/000202025126000030/0002020251-26-000030-index.htm); [Bridgecrest 2025-1](https://www.sec.gov/Archives/edgar/data/2050168/000110465926097653/0001104659-26-097653-index.htm); [Santander 2023-6](https://www.sec.gov/Archives/edgar/data/1999133/000119312526352810/0001193125-26-352810-index.htm), [2024-1](https://www.sec.gov/Archives/edgar/data/2004812/000119312526352812/0001193125-26-352812-index.htm), [2025-1](https://www.sec.gov/Archives/edgar/data/2049903/000119312526352849/0001193125-26-352849-index.htm); [CarMax 2024-1](https://www.sec.gov/Archives/edgar/data/2003263/000200326326000040/0002003263-26-000040-index.htm); [Ford 2025-A](https://www.sec.gov/Archives/edgar/data/2057342/000205734226000033/0002057342-26-000033-index.htm); [Ally 2024-1](https://www.sec.gov/Archives/edgar/data/2010413/000201041326000041/0002010413-26-000041-index.htm); [GM Financial 2024-1](https://www.sec.gov/Archives/edgar/data/2003007/000200300726000027/0002003007-26-000027-index.htm), [2025-1](https://www.sec.gov/Archives/edgar/data/2047316/000204731626000028/0002047316-26-000028-index.htm); [Hyundai 2025-A](https://www.sec.gov/Archives/edgar/data/2056104/000110465926099085/0001104659-26-099085-index.htm); [World Omni 2024-A](https://www.sec.gov/Archives/edgar/data/2007987/000110465926102655/0001104659-26-102655-index.htm); [Carvana 2024-P2](https://www.sec.gov/Archives/edgar/data/1999671/000199967126000032/0001999671-26-000032-index.htm), [2025-P1](https://www.sec.gov/Archives/edgar/data/2037956/000203795626000032/0002037956-26-000032-index.htm); [Toyota 2025-A](https://www.sec.gov/Archives/edgar/data/2047571/000119312526370139/0001193125-26-370139-index.htm); [Volkswagen 2025-1](https://www.sec.gov/Archives/edgar/data/2054483/000110465926098967/0001104659-26-098967-index.htm); [Nissan 2025-A](https://www.sec.gov/Archives/edgar/data/2063629/000119312526356412/0001193125-26-356412-index.htm); [Honda 2025-1](https://www.sec.gov/Archives/edgar/data/2052479/000119312526361400/0001193125-26-361400-index.htm). Each deal's full filing history is one click further into EDGAR. Three things in that table that the tier averages hide. The captives are not at zero. Honda and Nissan are the floor, with roughly 1.1 to 1.3% of balance extended in six months, and even there the adjusted number is four to fifteen times the reported one, because the reported one is so small. Ford reports 0.28% and reads 5.18%. Nobody on this list has an extension policy of none. CarMax and Carvana sit between the tiers on every column. CarMax reports 1.82% and reads 6.82%. Carvana reports 1.88% and reads 5.23%. Their six-month redefault rates, 19.6% and 23.4%, are closer to Santander's 24.0% than to any captive's. An average score in the low 700s got these pools a prime label. It did not get them a prime cure rate. The subprime shelves do not agree with each other. Santander's gap is 1,117 basis points on 12.3% of balance extended in six months. Exeter's is 2,562 on 30.2%. Bridgecrest, in the registered market for the first time with its [2025-1 trust](https://www.sec.gov/Archives/edgar/data/2050168/000110465926097653/0001104659-26-097653-index.htm), reports the highest raw delinquency of the fifteen at 11.23% and still carries a 1,791 point gap on top of it. Exeter is where the extension is closest to the norm. On its [2022-1 trust](https://www.sec.gov/Archives/edgar/data/1907570/000092963826003295/0000929638-26-003295-index.htm), 75% of the balance still outstanding has been extended at least once. **Inference** Rating agencies build tier indices from these same filings. The Fitch subprime 60+ index for the first quarter of 2026, covered in [an earlier note](https://lendriskanalytics.com/insights/q1-2026-dpd.html), was 6.90%. The reported figures here run higher because these ten shelves lean deep. The adjusted figures say something else: the index and the reported number share a denominator that has already been cleaned. Neither is wrong. Both are after the reset. ## IV · What happens in the six months after The adjusted number is a ceiling, and it needs to be said that way every time. Not every extended loan would have gone 60+. The measurement that turns the ceiling into an estimate is the one the [Philadelphia Fed](https://www.philadelphiafed.org/-/media/FRBP/Assets/Consumer-Finance/Reports/cfi-report-april-2026-do-recent-auto-loan-delinquency-rates-overstate-borrower-distress.pdf) said in April was the open question. Its report found that "while the stock of severe auto delinquencies is rising, the flow of new delinquencies into this stage is fairly stable," that the share of subprime loans getting an extension "reached approximately 3.5 percent last year," and offered this: "If extensions temporarily return borrowers to current status, but many borrowers subsequently fall behind again, the result is a cycle that keeps loans in the delinquent population longer." Its data are bureau tradelines, which cannot see the extension. The tapes can. So take every extension granted on the forty-four live deals to a loan that was under 60 days late at the time, with at least six more months of tape to watch. That is 513,907 of them. Ask two things of each. Did the loan hit 60+ in the next six months. Was it charged off. | Extensions granted while under 60 DPD | Events | 60+ within 6 months | Charged off within 6 months | Granted while already 30+ | |---|---|---|---|---| | Prime shelves | 156,405 | 13.4% | 3.4% | 0.8% | | Subprime shelves | 357,502 | 29.2% | 8.5% | 6.1% | Event-weighted across the 31 prime and 13 subprime live deals. Charge-off is zero balance code 4 inside the window. The last column is the share of all extensions that went to a loan already 30 or more days late when it was extended. On the prime shelves, seven in eight extended loans do not reach 60+ inside six months. That is what a hardship tool doing its job looks like. On the subprime shelves it is seven in ten. On Exeter and Bridgecrest it is under six in ten: 41.7% and 42.4% of extended loans are 60+ again within six months, and 12% to 15% are charged off in the same window. On those shelves an extension is, more often than anywhere else in the table, a loss that has been pushed down the road. The last column changes how you should read the reported number. On every shelf, extensions overwhelmingly go to loans that have not yet shown as 30 days late. 99.2% of prime extensions, 93.9% of subprime ones. Exeter is the outlier at 10.1% granted to loans already 30+. So the reset does not mostly pull accounts out of the delinquency buckets. It mostly keeps them from ever going in. That is why the flow of new delinquencies can look flat while the adjusted stock climbs, which is exactly the shape the Philadelphia Fed found in the bureau data and could not explain from it. **Inference** Multiply the ceiling by the redefault rate and the subprime gap of 1,845 points comes out to roughly 540 points of deterioration that was real and deferred, on top of the 8.10% reported. Same arithmetic on prime is about 40 points on 0.73%. Treat those as order-of-magnitude figures, not a second index. They assume the six-month redefault rate is the right conversion, and Section V is where that assumption gets tested. ## V · Why six months, and what three or twelve would say The lookback is a choice. A loan extended two years ago and current today is not being masked. A loan extended last month is. I picked six months before running the numbers. Here is what the other two obvious choices give you, same tapes, same month. | Gap, reported to adjusted 60+, July 2026 | 3-month lookback | 6-month | 12-month | |---|---|---|---| | Prime shelves | 179 bps | 318 bps | 591 bps | | Subprime shelves | 1,057 bps | 1,845 bps | 3,105 bps | | Exeter 2024-1 | 1,646 bps | 2,728 bps | 4,085 bps | | AmeriCredit 2024-1 | 1,390 bps | 2,483 bps | 4,021 bps | | Santander 2024-1 | 616 bps | 1,105 bps | 2,137 bps | | GM Financial 2024-1 | 211 bps | 409 bps | 715 bps | | CarMax 2024-1 | 395 bps | 725 bps | 1,369 bps | The order of the lenders is the same under all three. The ratio between tiers is 5.9x at three months, 5.8x at six, 5.3x at twelve. The choice moves the level. It does not move the ranking. That is the property you want from a convention. The six-month redefault window in Section IV is matched to the six-month lookback on purpose. A loan the extension held for six months is a loan the six-month adjustment should stop counting. ## VI · Where the record goes dark Everything above comes from registered deals, because only registered deals file Form ABS-EE. Private 144A placements do not. Search EDGAR's full-text index for ABS-EE filings by shelf and the [Tricolor](https://www.sec.gov/edgar/search/#/q=%22Tricolor%22&forms=ABS-EE) trusts return nothing. Neither do CPS, Westlake, GLS, Flagship or First Investors. Tricolor failed in September 2025 and its loan tapes were never public. The lenders most likely to run a heavy extension book are, by how the market is built, the ones this method cannot see. Bridgecrest is the exception that makes the point. DriveTime's DT Auto trusts were 144A. Its 2025 Bridgecrest trust is the first tape from that book anyone outside the deal has been able to read. **Inference** The registered subprime market here is four lenders. If you want the extension rate for subprime auto as an asset class, you do not have it, and without a change in what 144A issuers disclose you will not get it. What the four do show is enough to say the practice is universal inside the tier, and that what happens after it is the number that tells the lenders apart. ## VII · What this does not show Five things, so nobody has to find them for me. 1 **The adjusted number is an upper bound, and it is presented as one.** It counts every recently extended loan as if it would have been delinquent. Section IV gives the rate at which that held over six months. Anyone quoting 26.55% as a delinquency rate has misread this. 2 **Securitized pools are a slice of each lender's book, not the book.** Loans that charged off or paid before the cutoff never show up, and a deal ends at its clean-up call. The comparisons are like-for-like across pools. None of them is a statement about the lender's whole portfolio. 3 **Pool age differs.** A 2025 deal has had fewer months for extensions to pile up than a 2022 deal. That is why "extended ever" rises with age, and why the lender table weights across each shelf's live deals rather than picking one. The six-month column is the age-neutral one. 4 **An extension is not misconduct.** Every issuer discloses that the field resets the amount due, several describe their limits, and Issue 8 covered the disclosure failure that does get punished. This piece measures a mechanism the rules allow. What it argues is that the number after the mechanism should not be read as the number before it. 5 **Four registered subprime lenders are not the subprime market.** Section VI is the reason. Stretching the tier figures to cover the 144A issuers would be a guess, and I am not making it. Falsifiable This fails if the same construction run on the July 2027 tapes shows the subprime shelves' six-month redefault share at or below the prime shelves', or the subprime gap between reported and adjusted 60+ under 1,000 basis points at a six-month lookback. Both tests are mechanical, the fields are named above, and every filing is linked. **Anyone with the tapes can run it.** ## Proof: every figure, traced Every figure was rebuilt from the filed EX-102 asset data files for this publication, after the correction in Section I. Recomputed means the number was rebuilt from the loan-level records and matched. Derived means calculated from recomputed figures, arithmetic shown. Verified means read straight from the cited public document. | Claim as stated | Source | Status | |---|---|---| | Item 3(j) applies "if the loan has been modified"; Item 3(j)(2) is "the number of months the loan was extended during the reporting period"; Item 3(f)(6) is the next period payment due; Item 3(f)(25) is days past due | 17 CFR 229.1125, Item 3 ; element name paymentExtendedNumber per the SEC ABS XML technical specification | Verified | | Santander: a next payment due of 0.00 means the receivable is considered current because the obligor "either made a payment in advance or was granted a payment extension" | SDART 2024-1 EX-103 , July 2026 filing, Item 3(f)(6); same language in Exeter and AmeriCredit | Verified | | Carvana narrative lists Item 3(j)(2) as "No response"; Honda and AmeriCredit describe negative values as reversals; Honda uses modification type 4 for extensions | CRVNA 2024-P2 EX-103 , HAROT 2025-1 EX-103 , AMCAR 2024-1 EX-103 | Verified | | Two filing conventions: explicit zeros every month (Exeter, AmeriCredit, Santander, GM Financial, Hyundai) versus tag omitted unless modified (the other ten); values are 1 to 2 in the grant month and 0 or absent after | Population share of Item 3(j)(2) across every row of every panel: 100% for the first group, 0.1% to 2.9% for the second; loan-level sequences inspected on EART 2024-1, AMCAR 2024-1, SDART 2024-1, GMCAR 2024-1, HART 2025-A | Recomputed | | Universe: 44 live deals from 14 lenders, 1,148,856 loans on their latest tapes; 29 deals with a July 2026 tape, 883,104 loans, $15.56B outstanding; four deals excluded for mid-life filing gaps (AMCAR 2022-1, EART 2021-1, GMCAR 2022-1, WOSAT 2023-A) | EX-102 files fetched from each trust's EDGAR filing list, e.g. Exeter 2024-1 ; de-duplicated on asset number and period end because servicers re-file amended tapes | Recomputed | | Tier figures, July 2026: prime 19 deals, 629,959 loans, 0.732% reported, 3.914% adjusted, 318 bps, 3.32% extended in six months, 8.5% ever; subprime 10 deals, 253,145 loans, 8.101%, 26.546%, 1,845 bps, 21.00%, 45.2% | Balance-weighted on reportingPeriodActualEndBalanceAmount; 60+ from currentDelinquencyStatus; extension events from paymentExtendedNumber > 0 | Recomputed | | Lender table, all fifteen rows | Same construction, balance-weighted across each lender's July 2026 deals; deal-level values in the working file | Recomputed | | GMCAR 2024-1: 0.55% reported, 4.64% adjusted, 409 bps, 23,608 loans, 4.3% / 10.9% extended, WA score 775. AMCAR 2024-1: 3.86%, 28.69%, 2,483 bps, 34,667 loans, 26.5% / 56.2%, WA score 587 | GMCAR 2024-1 July 2026 ABS-EE ; AMCAR 2024-1 July 2026 ABS-EE ; scores from each deal's first tape, balance-weighted | Recomputed | | 6.1x: AmeriCredit's gap against the prime shelf | 2,483 / 409 = 6.07 | Derived | | Redefault: 156,405 prime events, 13.4% 60+ and 3.4% charged off within six months, 0.8% granted at 30+; 357,502 subprime events, 29.2%, 8.5%, 6.1%; Exeter 41.7%, Bridgecrest 42.4%, Carvana 23.4%, CarMax 19.6%, Santander 24.0%, AmeriCredit 18.5%, GM Financial prime 11.0% | Events are periods with paymentExtendedNumber > 0 on a loan under 60 DPD with six further tapes available; outcome is any subsequent currentDelinquencyStatus ≥ 60, or zeroBalanceCode 4, within six periods | Recomputed | | Roughly 540 bps and 40 bps of deferred deterioration | 1,845 × 0.292 = 539; 318 × 0.134 = 43 | Derived | | Lookback table: prime 179 / 318 / 591; subprime 1,057 / 1,845 / 3,105; the five named deals as printed; tier ratios 5.9x / 5.8x / 5.3x | Same July tapes, lookback windows of 3, 6 and 12 periods | Recomputed | | Exeter 2022-1: 75% of outstanding balance extended at least once | EART 2022-1 July 2026 ABS-EE ; 75.3% of end balance in loans with any positive paymentExtendedNumber on the tape | Recomputed | | Philadelphia Fed: stock rising while flow stable; approximately 3.5% of subprime loans extended last year; the redefault hypothesis; bureau tradeline data | Cheney, Hunt, Lambie-Hanson, Santucci and Zhou, April 2026 , Federal Reserve Bank of Philadelphia Consumer Finance Institute; data are the New York Fed Consumer Credit Panel / Equifax auto tradelines | Verified | | Fitch subprime 60+ index 6.90%, Q1 2026 | LendRisk note on the Q1 2026 print | Verified | | No ABS-EE filings for Tricolor, CPS, Westlake, GLS, Flagship or First Investors; Bridgecrest 2025-1 is registered | EDGAR full-text search, form ABS-EE , run per shelf name; BLAST 2025-1 July 2026 ABS-EE | Verified | **Sources & notes** **Data.** Form ABS-EE, exhibit EX-102 asset data files, for forty-four registered auto loan trusts, monthly from each deal's first tape through the July 2026 period end, fetched from EDGAR by trust CIK. Fields used: assetNumber, reportingPeriodActualEndBalanceAmount, currentDelinquencyStatus, paymentExtendedNumber, zeroBalanceCode, obligorCreditScore. Amended tapes de-duplicated on asset number and period end. Deals with a gap in the monthly sequence excluded. **Method.** Delinquency is balance-weighted throughout, matching the rating agency indices. An extension event is a positive paymentExtendedNumber in a period. Adjusted delinquency adds to the 60+ balance every loan under 60 days past due that had an event in the trailing lookback window, six months unless stated. Redefault follows each event on a loan under 60 days past due for six further periods. Tier membership follows each shelf's prospectus positioning; CarMax, Carvana and World Omni are counted prime. **Correction.** The desk built in August detected extensions as an increase over the prior month's value, which found nothing for issuers who leave the tag out unless a modification occurred. The rule was replaced before any figure in this brief was computed. The earlier read was never published. **Companion issues.** [Issue 8](https://lendriskanalytics.com/insights/current-on-paper.html) covers the extension mechanism, the Car-Mart non-reliance finding and the COVID extension surge. [Issue 12](https://lendriskanalytics.com/insights/the-lot-behind-the-branch.html) covers the credit union filing line that reads the loss after the reset. Where this brief goes beyond what the filings state, it is labeled as an inference. Point-in-time reading of public filings through September 13, 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. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html) LR LendRisk Analytics Independent market research Related [Issue 8 Current, on paper.](https://lendriskanalytics.com/insights/current-on-paper.html) [Issue 12 The lot behind the branch.](https://lendriskanalytics.com/insights/the-lot-behind-the-branch.html) --- title: "What the tape said: the lot behind the branch" url: https://lendriskanalytics.com/insights/the-lot-behind-the-branch.html publisher: LendRisk Analytics series: What the Tape Said issue: 12 published: 2026-08-31 kind: Research brief description: "One line on the NCUA call report counts cars already repossessed and not yet sold. Roughly seven in ten credit unions leave it blank. Across eleven consecutive annual cohorts, the institutions filling it in charged off more the following year, every time. Every figure recomputed from the raw filings; inferences labeled." html: https://lendriskanalytics.com/insights/the-lot-behind-the-branch.html --- # What the tape said: the lot behind the branch [← All insights](https://lendriskanalytics.com/articles.html) What the Tape Said · Issue 12 Research brief · 14 min read Credit unions · NCUA 5300 · Repossessed inventory # The lot behind the branch. NCUA 5300 call reports, 2022–2026 · Public record through August 31, 2026 · LendRisk Analytics Delinquency reports a promise already broken. There is a different line on the same quarterly filing, one field on schedule FS220P, that counts something physical: consumer vehicles the credit union has already repossessed and not yet sold. Cars sitting on a lot, waiting to become a charge-off. Roughly seven in ten federally insured credit unions leave it blank. Among the ones that fill it in, the size of that number, relative to the vehicle book, sorts next year's losses. It has done so in eleven consecutive annual cohorts without an exception at the top of the scale. $585M Repossessed consumer vehicles on credit union books, March 2026 69% Of 4,250 federally insured credit unions report zero or nothing in the field 2.0× Next-year charge-off rate, top repo quartile against non-reporters 11 of 11 Annual cohorts in which the top quartile charged off more than every group below it ## Bottom line Take every federally insured credit union with at least $10 million of vehicle loans in both March 2025 and March 2026. That is 2,284 institutions. Rank the ones reporting repossessed vehicle inventory by that inventory as a share of their vehicle book, cut them into quartiles, and put the non-reporters in their own group. Then wait a year and measure vehicle net charge-offs. | Group, March 2025 | n | Median repo / vehicle book | Vehicle NCO, year to March 2026 | |---|---|---|---| | Reports none | 1,159 | 0.000% | 0.877% | | Quartile 1 | 282 | 0.027% | 0.959% | | Quartile 2 | 281 | 0.086% | 1.349% | | Quartile 3 | 281 | 0.173% | 1.394% | | Quartile 4 | 281 | 0.438% | 1.753% | Charge-off rates are dollar-weighted within group: total net vehicle charge-offs over total vehicle loans, annualized. Every figure in this brief was recomputed from the raw NCUA files for publication, not carried over from working notes. Twice the loss rate, top of the scale against the bottom, from one field most institutions do not fill in. The same construction run on every available annual pair back to September 2022 produces the same ordering all eleven times. The read Delinquency is a payment that stopped. Repossessed inventory is a loss the institution has already committed to and not yet booked. The gap between those two is lead time, and one line on the 5300 measures it. **The honest version of this finding is narrower than it sounds:** repo inventory is a weaker standalone predictor than the credit measures everyone already watches. Its value is what it says about the institutions those measures call clean. ## I · What the field measures, and what it does not Account AS0024 on schedule FS220P: consumer vehicle foreclosed and repossessed assets. When a credit union takes a car back, the loan comes off the books and the car goes on, held at its estimated value until sold. The difference between loan balance and sale proceeds becomes the charge-off. So the field is a photograph of losses in transit: committed, sized approximately, not yet recognized. The obvious objection: if the car is already on the lot, the loss is already made, so predicting next year's charge-offs with it sounds like predicting rain with a wet umbrella. The answer is that the objection concedes the point. A measure that mechanically precedes recognition is exactly what lead time is. Delinquency has the same relationship to charge-offs and nobody calls it a tautology; it just sits earlier in the same sequence. Payment stops, car comes back, car sells, loss books. The field reads the third position in that sequence, the last one before the number everyone waits for. What the field does not do is beat the standard measures head to head, and this brief will not pretend otherwise. In the March 2025 cohort, ranked by Spearman correlation with next-year vehicle charge-offs: prior charge-offs +0.691, prior vehicle delinquency +0.480, repo inventory +0.387, indirect-lending concentration +0.372. Repo inventory is the weakest of the four as a standalone signal. Controlling for prior delinquency it retains +0.337; controlling for delinquency, prior charge-offs and concentration together it retains +0.156. It carries some information of its own. Not a lot. **Inference** A signal that ranks third of four on its own has one legitimate use: the cases where the signals above it say nothing. That is where this one earns its place, and the next section is the evidence. ## II · The quadrant that matters Split the cohort twice. Once at the median of prior vehicle delinquency, 0.576%, the number a board sees. Once at the median repo ratio among reporters, 0.122%, the number most boards do not. Four cells: Chart 1 · Next-year charge-offs by prior delinquency and prior repo inventory March 2025 cohort, n=2,284, vehicle NCO dollar-weighted within cell, year to March 2026. Non-reporters counted as low repo; the reporters-only version is given below the chart. Splits at the cohort medians: prior vehicle delinquency 0.576%, repo ratio among reporters 0.122%. Reporters only, same construction: low-DQ low-repo 0.752% against low-DQ high-repo 1.200%, a gap of +0.45pp, so the result is not an artifact of bucketing non-reporters as low. Across all eleven annual cohorts the low-DQ gap is positive eleven times, median +0.37pp, median ratio 1.67x. The top-right cell is the finding. One hundred ninety-six institutions whose delinquency sits below the cohort median, the profile a board reads as healthy, and whose lots are full. They charged off 1.11% over the following year against 0.67% for their clean-lot peers. Two thirds more loss, from a group the standard measure cannot separate. That gap is positive in all eleven cohorts. **Inference** Delinquency can be quiet while the lot fills, because extensions and modifications reset the past-due clock without touching the car. Issue 8 documented that mechanic in the securitized market. Here it is in a depository filing: the payment record says current, the yard says otherwise. ## III · Eleven cohorts, one ordering Chart 2 · Top-quartile and non-reporter charge-offs, every available annual pair Same construction repeated on each annual pair from September 2022–23 through March 2025–26. Top line is quartile 4, bottom line is non-reporters. Q4 exceeds non-reporters in 11 of 11 pairs; the spread runs 0.71 to 1.15 percentage points, median 1.04. Quartile 2 beats non-reporters, quartile 3 beats quartile 2, and quartile 4 beats quartile 3 in all eleven. Spearman correlation sits between +0.358 and +0.393 in every pair. The eleven cohorts share most of their member institutions, so this is stability of one population through time, not eleven independent experiments. One honest wrinkle, stated rather than smoothed: the bottom quartile of reporters is indistinguishable from the non-reporters. Q1 lands below the non-reporter group in seven of the eleven pairs, median difference minus 0.03 points. A little repo inventory means nothing. The gradient begins at quartile 2, which in the current cohort means a repo ratio around 0.09% of the vehicle book and up. The ordering also survives the obvious confound. Big institutions report more and charge off more, so the gradient could have been a size effect wearing a costume. Within each of four asset bands separately, under $200 million, $200 to $500 million, $500 million to $1 billion, and over $1 billion, the high-repo half of reporters exceeds the non-reporters by 0.5 to 0.7 points. The spread lives inside every size class. ## IV · A correction, and what concentration turned out to do An earlier version of this analysis tested indirect-lending concentration against contemporaneous delinquency, found almost nothing, and said so. That test was wrong, and the error was mine: it compared the two measures on different clocks and different outcomes. On identical terms, concentration measured in March 2025 against vehicle charge-offs in the year to March 2026, concentration is monotonic in all eleven cohorts, with a median spread of 0.45 points from the lowest band to the highest and a standalone correlation of +0.372 against repo inventory's +0.387. Essentially tied. Under full controls, concentration retains slightly more independent signal, +0.182 against +0.156. **Inference** The lesson generalizes past this dataset: two measures can only be compared on the same clock against the same outcome. The corrected result does not weaken the repo finding, but it does mean the field is one of two useful early lines on the filing rather than the only one, and an examiner who wants a two-variable screen should take both. ## V · Seven in ten lots are dark Now the part that turns a statistical note into a supervisory question. Among credit unions with at least $5 million of vehicle loans in March 2026: Chart 3 · Who fills the field in Share of credit unions with $5M+ vehicle loans reporting any repossessed vehicle inventory, March 2026, by asset band. The delinquency of each band is printed beside its bar. Reporting counts: 443 of 1,664 under $200M; 255 of 499 at $200–500M; 185 of 282 at $500M–$1B; 367 of 466 above $1B. Delinquency is dollar-weighted within band. The institutions least likely to fill the field in are the small ones, and the small band carries the highest vehicle delinquency of the four. Whatever sits on those lots is invisible to anyone reading the filings, including, at one remove, the examiners who set the exam calendar from them. A zero in this field means either an empty lot or an unanswered question, and the filing does not say which. **Inference** The non-reporter group in every table above is a blend of clean books and books that do not answer. That blend charged off 0.877% in the current cohort, higher than the bottom two reporter quartiles in several pairs, which is what a mixture of clean and unmeasured would look like. The one-line version for a supervisor: the field is cheap, it is already on the form, and the pattern above is an argument for making it non-optional. ## VI · What this does not show Five things, so nobody has to find them for me. 1 **Repo inventory is the weaker standalone measure.** Prior charge-offs and prior delinquency both outrank it. Anyone running a single-variable screen should not pick this variable. Its use is conditional: the low-delinquency, high-repo cell, and the reporting gap. 2 **The eleven cohorts overlap.** The same institutions appear in most pairs, so eleven repetitions demonstrate persistence, not independence. There is no way around this with public data; the honest claim is stability, not eleven separate confirmations. 3 **Survivorship is present and disclosed.** Of 2,409 eligible institutions in the headline pair, 56 dropped out before the outcome year, 2.3%. The dropped group had weaker capital and earnings but lower vehicle delinquency than the survivors, so the direction of the bias is not obvious, and at 2.3% its size is small. 4 **The gradient starts at quartile 2.** A small positive repo number is noise. Any screen built on this field needs a floor, not a flag on any nonzero value. 5 **Nothing here identifies cause.** A full lot can mean aggressive repossession policy, slow disposal, a weak local auction market, or genuinely worse paper. The filing cannot distinguish these. What it can say is that whichever mix is present, the following year's charge-offs are higher, and that is the property a monitoring signal needs. One outside data point, for corroboration rather than support: Depository360, which runs charter-departure work on the same filings, found that delinquency separates ceased from continuing credit unions in all thirteen of its cohorts, but by margins around 0.12 points. Our own March cohorts reproduce that scale: +0.12, +0.09, +0.09. The headline credit measure separates reliably and thinly. That is the gap this field, and concentration beside it, exist to fill. Falsifiable The claim fails if the quartile-4 group stops exceeding the non-reporter group in the next two annual cohorts, March 2027 and March 2028, or if the low-delinquency high-repo cell stops carrying a positive gap. Both tests are mechanical, the construction is fully specified above, and the data is public. **Anyone with the raw files can run it.** ## Proof: every figure, traced All figures were recomputed from the raw NCUA call report archives for this publication. Recomputed means the number was independently rebuilt from the raw quarterly zip files and matched. Derived means calculated from recomputed figures, arithmetic shown. | Claim as stated | Source | Status | |---|---|---| | 4,250 federally insured credit unions filing March 2026; $585.0M repossessed consumer vehicle assets; 1,304 reporting a nonzero value (31%, so 69% report zero or nothing); $479.6B total vehicle loans | NCUA 5300 call report, cycle 2026-03: FOICU.txt (CU_TYPE 1 and 2), FS220P.txt field ACCT_AS0024, FS220A.txt fields ACCT_370 + ACCT_385. Matches NCUA's published Q1 2026 institution count exactly | Recomputed | | Headline cohort: 2,284 institutions with $10M+ vehicle loans in both 2025-03 and 2026-03, complete cases; group sizes 1,159 / 282 / 281 / 281 / 281 | Same files, both cycles, joined on CU_NUMBER | Recomputed | | Group charge-offs 0.877 / 0.959 / 1.349 / 1.394 / 1.753%, dollar-weighted; median repo ratios 0.000 / 0.027 / 0.086 / 0.173 / 0.438% | Net charge-offs from FS220I (ACCT_550C1 + 550C2 less 551C1 + 551C2), annualized, over vehicle loans | Recomputed | | 2.0x top quartile versus non-reporters | 1.753 / 0.877 = 2.00 | Derived | | Correlations with next-year vehicle NCO: prior NCO +0.691, prior DQ +0.480, repo +0.387, concentration +0.372; repo partials +0.337 and +0.156; concentration partial +0.182 | Spearman on the headline cohort; partials by rank regression residuals. Full-controls values | Recomputed | | 2x2 cells: 0.668 / 1.111 / 1.432 / 1.834%, n = 946 / 196 / 776 / 366; splits at DQ 0.576% and repo 0.122% | Headline cohort, splits at cohort medians | Recomputed | | Reporters-only check: 0.752% versus 1.200%, gap +0.45pp | Same construction restricted to the 1,125 reporters | Recomputed | | Eleven annual pairs, 2022-09 through 2025-03 starts: Q4 above non-reporters 11/11, Q2 above non-reporters 11/11, Q3 above Q2 11/11, Q4 above Q3 11/11; spread 0.71–1.15pp, median 1.04; rho +0.358 to +0.393; low-DQ gap positive 11/11, median +0.37pp | All fifteen cached quarterly archives, 2022-09 through 2026-03; replication table in the working files (repo_replication_FICU.csv) | Recomputed | | Q1 below non-reporters in 7 of 11 pairs, median −0.03pp | Same replication table | Recomputed | | Within asset bands, high-repo half exceeds non-reporters by 0.5–0.7pp in all four bands | Headline cohort split at $200M / $500M / $1B; recomputed spreads +0.69 / +0.60 / +0.67 / +0.53 | Recomputed | | Concentration on identical terms: monotonic 11/11, median spread +0.45pp, rho +0.372 standalone in the headline pair, +0.246 median across pairs | 2025-03 indirect share of vehicle book against 2026-03 vehicle NCO; concentration_fair_test.csv in the working files | Recomputed | | Reporting by band: 443/1,664 (27%), 255/499 (51%), 185/282 (66%), 367/466 (79%); band delinquency 0.86 / 0.77 / 0.76 / 0.79% | Cycle 2026-03, institutions with $5M+ vehicle loans | Recomputed | | Survivorship: 56 of 2,409 dropped (2.3%); dropped group weaker capital and earnings, lower vehicle DQ | Eligible set at 2025-03 against filers at 2026-03 | Recomputed | | Ceased-versus-continuing delinquency gaps +0.12 / +0.09 / +0.09pp in our March 2023–2025 cohorts | Computed in-house from the same archives. Depository360's independent charter-departure work (thirteen cohorts, ~0.12pp) cited as corroboration only, not as a source for any figure here | Recomputed | | Field definition: ACCT_AS0024, "Consumer Vehicle Foreclosed and Repossessed Assets," schedule FS220P | AcctDesc.txt inside each NCUA quarterly archive | Recomputed | **Sources & notes** **Data.** NCUA 5300 call report quarterly archives, cycles 2022-09 through 2026-03, fifteen quarters, downloaded from ncua.gov. Files used: FOICU.txt for charter type (federally insured only, CU_TYPE 1 and 2; 86 privately insured charters excluded), FS220A.txt for vehicle loan balances, FS220P.txt for repossessed vehicle assets, FS220I.txt for vehicle charge-offs and recoveries, FS220C.txt for indirect balances, AcctDesc.txt for field definitions. **Method.** Annual pairs join each start quarter to the same quarter one year later on charter number. Cohort floor $10 million of vehicle loans at both ends. Charge-offs are net of recoveries and annualized by cycle month. Group rates are dollar-weighted. Quartiles are set among reporters only; non-reporters are their own group. Correlations are Spearman. Every figure in the body was recomputed from the raw archives immediately before publication. **Corroboration.** Depository360, "Why Credit Union Charters Disappear," on delinquency separation at charter departure; referenced as independent corroboration of the thinness of the headline measure, with the equivalent figures computed in-house from our own archives. **Companion issues.** Issue 8 covers extension mechanics and the reported-versus-actual gap in securitized subprime. Issue 10 covers the credit union funding turn. Issue 11 covers the past-due arithmetic. Where this brief reasons beyond what the filings state, it is labeled as an inference. The concentration correction in Section IV is stated in the body, not hidden in a footnote. Point-in-time reading of public filings through August 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. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html) LR LendRisk Analytics Independent market research Related [Issue 8 Current, on paper.](https://lendriskanalytics.com/insights/current-on-paper.html) [Issue 10 The one-way door.](https://lendriskanalytics.com/insights/the-one-way-door.html) --- title: "Past due, unchanged" url: https://lendriskanalytics.com/insights/past-due-unchanged.html publisher: LendRisk Analytics series: What the Tape Said issue: 11 kind: Operator study description: "Eleven buy-here-pay-here stores, thirty months of real monthly numbers. The share of customers behind held steady. What each dollar of loss cost in interest moved, bottomed in late 2025, and is climbing back. What these operators did about it." html: https://lendriskanalytics.com/insights/past-due-unchanged.html --- # Past due, unchanged [← All insights](https://lendriskanalytics.com/articles.html) What the Tape Said · Issue 11 Operator study · 9 min read Eleven stores · Thirty months · Real numbers # Past due, unchanged. Monthly operator data, January 2024 through June 2026 · No store identified · LendRisk Analytics Ten issues of this series have been built on things anyone can look up. Court filings, bank disclosures, rating agency numbers. This one is different. It is built on the monthly numbers eleven buy-here-pay-here stores actually turn in, month after month, for two and a half years straight. Same eleven stores at the start and at the finish. What that lets you do, which no public report can, is hold a real set of books still and watch which numbers move and which ones do not. The one almost everybody watches did not move at all. Something else did. 31.7 → 32.4% Customers behind. Basically flat across thirty months $1.14 → $0.96 Interest collected for every dollar lost. Low was $0.83 36.6 mo Average term written, down from 38.9. These stores tightened 11.2% Cash down as a share of price. Same at the end as the start How to read this Everything here comes from eleven stores that reported every month from the first half of 2024 through the first half of 2026. Stores that joined partway through are left out, because otherwise you are measuring a changing roster instead of a changing business. When numbers get combined, **the dollars get added up first and then divided**. Averaging percentages would let a quiet month count the same as a heavy one, and that is how a book ends up looking better on paper than it does in the bank account. **Here is the whole thing in two sentences.** Across thirty months, the share of customers sitting behind at these eleven stores did not really move. Over those same thirty months, what a dollar of loss cost them in interest did move, it got worse through 2025, and it has been coming back since. ## I · The number that held Start with the number on every Monday morning report in this business. Customers behind, as a share of active accounts. These eleven stores came into 2024 at 31.7 percent and finished the first half of 2026 at 32.4 percent. Draw a line through all thirty months and it is flat. Not gently sloping. Flat. That deserves to be said out loud, because holding a past-due number steady for two and a half years is work. Collectors were doing their jobs. Nobody let the book drift. If the only thing you had looked at over that stretch was the past-due column, you would have concluded the business was running exactly the way it ran in 2024. Chart 1 · Two numbers from the same eleven stores Top is the number everybody watches. Bottom is what a dollar of loss actually cost in interest. Same stores, same months. Hover any month for both numbers Same eleven stores, dollars added before dividing. The two panels sit one above the other, not on top of each other, because they are different kinds of number: one is a percentage of accounts, the other is dollars per dollar. Lines in different units should never be read against each other. The thin pale lines are the raw month-by-month readings, which bounce around because a light loss month makes the ratio jump. The heavy lines connect the half-year figures quoted in the text. ## II · What these stores did while it held The past-due number did not stay flat by accident, and the deal terms show it. Average term written came down from 38.9 months to 36.6 months. That is just over two months shorter, and it is the textbook move when customers are getting squeezed. Cash down finished at 11.2 percent of the selling price against 11.2 percent at the start, which means the down payment requirement held through a stretch when plenty of people would have been tempted to let it slide. Units sold were about six percent lower, which is what it looks like when a store turns down deals it does not like. Chart 2 · Shorter contracts, and a down payment that did not slip What these eleven stores were actually writing, half-year by half-year. Hover a point for the exact figure Term is weighted by units sold in each block. Cash down is total down collected divided by total selling price. The zigzag in the right panel is seasonal rather than a change in policy: down payments run higher in first halves and lower in second halves in every year here, which lines up with tax refund timing, so compare first half to first half. Down payment as submitted may include pickup payment arrangements at some stores, so read this as the terms written rather than strictly cash in the drawer at delivery. The two panels use different scales and are not comparable to each other. Shorter contracts at a similar amount financed means a bigger monthly payment, and that shows up too. The average payment written went from about $500 to about $534, up close to seven percent, while the term came down almost six percent. That is the tradeoff these stores made, and on the evidence of the past-due line it worked. ## III · The number that moved Now the other one. For every dollar these stores lost after the auction paid them back, how many dollars of interest did they collect? In the first half of 2024 it was **$1.14**. Interest more than covered the losses. By the second half of 2025 it was **$0.83**. In the first half of 2026 it was back to **$0.96**. That is the bottom panel of Chart 1, and out of everything measured across these thirty months, it is the only number with a real trend behind it. Two things are true about that at once, and both matter. It fell below the line where interest pays for the losses, and it has been climbing back for two straight blocks. The worst of it was late 2025. **What this means** Past due counts customers. This counts money. When the customer count holds and the money gets tighter, the arithmetic behind each account that goes bad has changed. It does not mean anyone got worse at their job, and section II is the evidence they did not. It means the same reading on the past-due column was carrying a different result underneath it. ## IV · Where it came from Two things moved, and neither of them is something you decide at the desk. **Losses got bigger.** For every $100 out on the street, these stores were writing off $18.32 a year after the auction paid them back at the start of the window. That went to $24.54 at the worst of it, and it is $21.03 now. **The auction paid back less.** For every $100 written off, the lot was returning $32.63 at the start. That fell to $27.20 and has recovered to $30.95. That swing alone moves the whole picture, and it is set at the auction lane, not in your office. What did not move much is what these stores were earning. Interest income ran right around 20 percent of money on the street the whole way through, easing only slightly. So the tighter number is not a story about earning less. It is a story about what the earnings were being measured against getting bigger for a while. | Half-year | Customers behind | Lost per $100 out | Auction back per $100 written off | Interest per $1 lost | |---|---|---|---|---| | 2024 · first half | 31.7% | $18.32 | $32.63 | $1.14 | | 2024 · second half | 33.0% | $19.94 | $30.81 | $1.03 | | 2025 · first half | 30.3% | $22.86 | $28.72 | $0.88 | | 2025 · second half | 33.1% | $24.54 | $27.20 | $0.83 | | 2026 · first half | 32.4% | $21.03 | $30.95 | $0.96 | Eleven stores, dollars added before dividing. Loss and interest figures are stated at an annual rate against money on the street at the start of each month. The last row beats the row above it on every single column, which is the most useful thing in this table. ## V · What a customer behind actually cost Here is the same idea in the plainest form available. Take the dollars written off, after the auction, and divide by the number of customers sitting behind. That is roughly what one behind account was costing. Chart 3 · Dollars written off per customer behind Same share of the book behind, different amount of money attached to it. Hover a bar for the detail Net dollars written off in each half-year divided by the count of accounts behind in that half-year, across the same eleven stores. This is a rough comparison of two totals rather than a tracked account-by-account figure, since the panel does not carry the fields to follow an individual account from behind to written off. Treat it as a sense of direction, not a per-account cost you could book. It ran $527 in the first half of 2024, climbed to $672 through 2025, and came back to $549. Roughly the same share of the book was behind at the start and the end. What was attached to those accounts got heavier for a while, and it is getting lighter again. ## VI · Not one or two stores This is not one outlier dragging a group average around. Taking each store on its own and comparing the first half of 2024 with the first half of 2026, eight of eleven had the same or fewer customers behind, nine of eleven had a tighter interest-to-loss number, and seven of eleven had both at the same time. Chart 4 · Every store, same comparison Hollow dot is the first half of 2024. Filled dot is the first half of 2026. The note on the right is what that store's past-due count did. Hover a row for that store One row per store, lettered and ordered by where they started on past due, which is not a ranking of anything else. Anything past $2.45 is drawn at the edge and marked; one store started well above that on a very small loss base, so its opening figure is unstable and should be read as high rather than as a level. Past-due change is in percentage points and called about the same when the move is under 1.6 points. ## VII · What to do with this on your own book None of this says the past-due report is wrong. It answers the question it was built to answer, which is how many customers are behind this month, and it answers it honestly. It just is not the only vital sign, and across these thirty months it was the one holding still. The number that moved is one you can run yourself in about ten minutes, from figures you already have. Take the interest you collected for the month. Divide it by what you wrote off that month after subtracting whatever the auction gave back. That is dollars of interest per dollar of loss. Above $1.00 the interest paid for the damage. Below it, something else did. Run it for twelve months and add the dollars up before you divide, because one quiet month will make a single month look spectacular and tell you nothing. That is the whole method. ## Where this comes up short **Eleven stores is eleven stores.** Enough to say what happened to these eleven. Not enough to speak for the industry, and it is not trying to. **Only one of these numbers is solid enough to call a trend.** The interest-per-dollar-lost move holds up when you test it properly. The loss rate and the auction recovery move in the direction described but do not clear the same bar on a month-by-month basis, which is why the half-year table is the honest way to show them. **Past due is counted differently at different stores.** Some systems start counting at one day, some at thirty. That is why it is only ever used here to compare a store to itself over time, and never one store against another. **Auction figures are as reported.** Whether a store books what the car actually brought or a value assigned when it came back is not something this data can tell. **A growing book looks better on past due.** New accounts have not had time to fall behind, so a store adding accounts quickly will show a lower past-due share for the same underwriting. Account counts moved around quite a bit across these eleven, and this data cannot separate that out. It is the biggest open question here and worth saying plainly. **Thirty months is not a cycle.** It covers one stretch of pressure and the start of a recovery, and that is all. ## Three things worth checking ### 1 Do you know what a dollar of loss costs you in interest right now, and what it cost two years ago? If the answer moved and nothing on your monthly sheet moved with it, the sheet is not wrong, it is just quiet. ### 2 When you look at your past-due column, do you know whether the accounts behind it are carrying more money than they used to? Same percentage, heavier accounts, is a real thing and it happened here. ### 3 When your numbers improve, do you know which half improved? These stores got better in 2026 mostly because the auction started paying again, not because they were earning more. Both are good news. They are not the same news, and only one of them is yours to keep. The past-due report is the oldest number in this business and there is nothing wrong with it. Across thirty months at eleven stores it gave the same answer every time while the money behind that answer moved, went the wrong way for about a year and a half, and started coming back. Worth having a second number next to it. **Sources & notes** Every figure comes from monthly operating numbers submitted by eleven independently owned buy-here-pay-here stores taking part in a performance benchmarking group, covering January 2024 through June 2026, used with their permission and reported only as a group. No store, owner, location or business name appears anywhere here, and no number is traceable to any one of them. Stores are lettered in Chart 4 in the order they started on past due, which is not a ranking of anything else. How the numbers were built. Everything is calculated from the raw figures each store submits rather than from any summary already sitting in the group's own reports, so a number here may differ from one circulated inside the group. Where stores are combined, dollars are summed first and then divided. Loss and interest rates are stated at an annual rate against money on the street at the start of each month. Interest per dollar lost is interest collected divided by what was written off after subtracting auction proceeds. To be included, a store had to report in both the first half of 2024 and the first half of 2026, which takes the group from fifteen stores to eleven. Related reading. Issue 9 covers why cash leaves a buy-here-pay-here book on day one and comes back slowly. Issue 8 covers what a current flag does and does not tell you when a payment gets moved. If you run this on your own book and get a different answer, that is the useful outcome and worth a conversation. Same if you think the growing-book problem above sinks the whole thing. [Send it through](https://lendriskanalytics.com/contact.html). LR LendRisk Analytics Independent market research Continue reading [Issue · 09 · Deep study The *cash problem.*](https://lendriskanalytics.com/insights/the-cash-problem.html) [Issue · 08 · Method study Current, *on paper.*](https://lendriskanalytics.com/insights/current-on-paper.html) --- title: "What the tape said: the one-way door" url: https://lendriskanalytics.com/insights/the-one-way-door.html publisher: LendRisk Analytics series: What the Tape Said issue: 10 published: 2026-08-23 kind: Research brief description: "Credit unions began securitizing their own loans in November 2019. Twenty-five deals and roughly $8.4 billion later, the dollars are still small. The structure is not. Why a funding tool adopted under liquidity stress has not been retired now that the stress has eased. Every figure sourced; inferences labeled." html: https://lendriskanalytics.com/insights/the-one-way-door.html --- # What the tape said: the one-way door [← All insights](https://lendriskanalytics.com/articles.html) What the Tape Said · Issue 10 Research brief · 18 min read Credit unions · Originate-and-hold · The distribution turn # The one-way door. Public record through August 23, 2026 · LendRisk Analytics Credit unions have sold conforming mortgages into the secondary market for decades. Consumer paper was the part that stayed home: member deposits fund member auto loans, and the loans sit on the balance sheet until they pay off. In November 2019 a $2.2 billion credit union in Tampa broke the loop, selling bonds backed by its own auto loans to Wall Street investors. Twenty-five or so deals and roughly $8.4 billion later, the dollars are still small enough to dismiss. The structure is not. Every door this shift has opened, from the first deal to the first multi-seller trust to the first sub-$1 billion participant to the first non-auto collateral, opened recently, and none has closed. For four years issuance tracked the liquidity cycle almost exactly. Then it stopped tracking it. ~$8.4B Cumulative credit union ABS across at least 25 deals since November 2019 (Dechert count) <0.5% Share of the $1.73 trillion credit union loan book that cumulative issuance represents 68.8% → 81%+ The loan-to-share swing, early 2021 to late 2022, that pushed issuers through the door $1.4B First half of 2026, against $2.1 billion for all of 2025 and a $2.6 billion record year ## Bottom line The claim is one sentence. Credit unions are early in a shift from holding what they originate to distributing it, the shift is small in dollars and cyclical in its trigger, and it is one-directional in its structure, because every capability it has built outlives the conditions that built it. Three things sit on the record. First, the deals: from GTE Financial's November 2019 debut through PenFed's fourth transaction in June 2026, the July 2025 Alloya multi-issuer trust, Corporate One's January 2026 multi-seller deal, and Lafayette Federal's February 2026 home-improvement transaction, the first credit union ABS on collateral other than autos. Second, the trigger: a liquidity swing that took the system loan-to-share ratio from a record-low 68.8% in early 2021 to above 81% by late 2022, on loan growth near 19%. Third, the fact that carries the forward argument: the trigger has eased and the behavior has not. By the first quarter of 2026, system net income was up 30.5% year over year and loan-to-share had fallen to 81.5%. Over the same stretch, the first half of 2026 produced roughly $1.4 billion, against $2.1 billion for all of 2025. The read A funding tool adopted under stress does not get retired when the stress eases. It gets institutionalized. That is what the extension data showed in servicing ([Issue 8](https://lendriskanalytics.com/insights/current-on-paper.html)), and it is what the issuance data is starting to show in funding. The dollars say niche. The structure says ratchet: the legal architecture exists, the playbooks are written, the multi-seller rails are built, and the traditional alternative, selling participations to smaller credit unions, is losing its buyer base to four decades of consolidation that is not reversing either. **Inference, labeled as one:** the shift is premature to declare and pointless to bet against, because every mechanism that would undo it is itself in decline. Chart 1 · Credit union auto ABS activity by year Deals or issuers pricing each year, with published dollar totals annotated beneath. The unit each source reports is marked, because they are not the same unit. Dollars are shown as a secondary row because the sources disagree on them by scope; the deal count they do not dispute. Deal counts: 2019 through 2022 sum to four, matching Dechert's count of four credit union securitizations over that span exactly (GTE 2019, Unify 2021, PenFed and Oregon Community 2022). There was no credit union securitization in 2020. 2023 count and $2.0B per KBRA and Dechert, which agree. 2024 per KBRA, "Record-High 2024 Issuance," six credit unions and $2.4B; S&P counts five issuers and $2.07B in its narrower rated-auto universe. 2025 per S&P, six issuers and $2.1B. **Units differ by year and the difference is not cosmetic:** 2019 through 2023 are transaction counts, while the 2024 and 2025 figures are issuer counts, which coincide with transaction counts only in years when no credit union priced twice. No deal or issuer count is published for the first half of 2026, so that bar is drawn open; four transactions are individually confirmable (Corporate One in January, Lafayette Federal in February, Oregon Community's 2026-1, and PenFed in June) and the $1.4B is the published aggregate. Read the shape rather than any single bar. One deal, then a year with none at all, then a slow build to a step change in 2023 that has held for three years and is already matched in the first half of 2026. The dollar row underneath is deliberately secondary. Sources disagree on annual dollars by as much as 25%, because they are counting different universes: KBRA and Dechert tally the market and land within a rounding error of each other, while S&P counts only what it rates and reports a smaller number for the same year. The deal count is where they converge, so the deal count carries the chart. ## I · The first deal, and the roster since GTE Financial is the right place to start, because GTE's own history contains the whole thesis in miniature. Before it ever issued a bond, GTE distributed the old way. Its CFO, Brad Baker, has said the credit union sold roughly $202 million of auto loans through participations between 2014 and 2018, in individual pieces ranging from $1 million to $23 million. Participations were the ceiling of what a credit union could do with its own paper: private, bilateral sales to other depositories, funded by those depositories' deposits. Then in November 2019, GTE became the first credit union in the country to sponsor a securitization, pledging roughly 9,000 new- and used-car loans with a balance of $185.27 million as collateral for GTE Auto Receivables Trust 2019-1, a Rule 144A transaction through Stifel Nicolaus. Same institution, same collateral, a different instrument, and a different buyer base entirely. Two different sizes circulate for that deal, and the capital structure settles it. The A classes were a $36.3 million money-market tranche, $58.19 million of A-2 and $52.94 million of A-3, which sum to $147.4 million. Add the $22.3 million single-A and $5.3 million BBB subordinates and total notes come to $175.0 million, against the $185.27 million pool. So $147.4 million is the senior classes, $175 million is total notes, and $185.27 million is collateral. Both circulating figures were right about different things. This brief uses $175 million, total notes. Then nothing. GTE returned to market on April 24, 2023, with a $202.8 million deal, three and a half years after the first one, through the Stifel and BofA Securities program that ran four of that year's credit union transactions. The gap is the part worth sitting with. The pioneer did not become a habitual issuer on the strength of the experiment working. It waited out 2020, 2021 and 2022, and came back in the middle of the liquidity squeeze, in the same year six other credit unions priced deals. The tool had been legally available since the 2017 safe harbor, six years by then, and stayed largely unused until conditions made it necessary. Three and a half years is GTE's own gap between deals, not the industry's wait. Baker's own account of what the first deal did is worth quoting, because it names three distinct motives rather than one. The securitization, he said, "increased our liquidity, leveled out our concentration of auto loans and provided transactional income, as well as providing a higher rate of servicing income throughout the life of the security." Liquidity is the cyclical reason. Concentration management and fee income are structural ones, and they do not go away when deposits return. Nothing about that first deal was practically possible before 2017. The NCUA's securitization safe harbor, published in the Federal Register on June 30, 2017 and effective July 31, 2017, is what made it work: a rule at 12 CFR Part 709 defining how the NCUA Board, acting as liquidating agent or conservator, treats financial assets a credit union transfers in a securitization or participation. Without that clarity, capital markets buyers would not touch credit union paper, because they could not be certain they owned what they paid for. A separate 2017 NCUA legal opinion confirmed federal credit unions' authority to issue the securities at all, under Section 107(17) of the Federal Credit Union Act. The rule then sat unused for two and a half years. GTE used it first. Chart 2 · Seven doors, none of them closed Each marker is a first, plotted on a linear month scale from November 2019 through August 2026. Every date is from a named primary or trade source; see the verification table. Linear scale, 7.28 pixels per month, November 2019 at x=70 and August 2026 at x=660. There is no marker for 2020 because no credit union securitized that year. The hollow marker at March 2021 denotes a deal whose size was never published. The shaded column spans the liquidity squeeze described in Section III. The January and February 2026 markers sit seven pixels apart and are labeled at different heights for that reason, not because they differ in kind. | Date | Deal | The first it represents | |---|---|---| | Nov 2019 | GTE Financial auto ABS (Stifel) | First credit union securitization | | 2020 | No transactions | The one year with nothing; the tool sat idle again | | Mar 2021 | UNIFY Auto Receivables Trust 2021-1, ~$300M | A second issuer; the tool is not proprietary to GTE | | 2022 | PenFed and Oregon Community | Two issuers in one year for the first time | | Apr 2023 | GTE returns, $202.8M | The pioneer comes back after three and a half years, inside a seven-deal year | | 2023 | Seven deals, ~$2.0B, +167% y/y | The inflection year: GTE and Oregon Community repeat, Veridian and GE Credit Union debut | | Jul 2025 | Alloya Auto Receivables Trust 2025-1, $150M | First multi-issuer deal: Blaze, Consumers and Interra contribute $50M each; seven note tranches | | Jan 2026 | Corporate One multi-seller, $335.1M pool | Wright-Patt, Everwise and Day Air; Day Air is the first credit union under $1 billion in assets inside an ABS | | Feb 2026 | Lafayette Federal MRQI 2026-HI1, $459M | First credit union ABS on collateral other than auto loans | | Jun 2026 | PenFed Auto Receivables Owner Trust 2026-A, $354M | Fourth deal from one issuer, with the programmatic intent stated on the record | Cumulatively, at least 25 offerings and approximately $8.4 billion, per Dechert, which advised on 22 of the 25 it counts. Dechert profits from the trend it tallies, which is worth holding in mind against its arithmetic. The shape of that timeline is itself an argument. One deal in November 2019, one more in March 2021, and then almost nothing until 2023, when seven priced at once and the pioneer finally came back. The rule landed in 2017 and sat unused for two and a half years. GTE proved it worked in 2019, and for three more years almost nobody followed. What changed was not the law or the technology. It was the liquidity, and the fit is close enough to test directly. Loan-to-share stood at 84.1% in the fourth quarter of 2019, near a pre-pandemic high, when GTE went to market fully loaned up. Pandemic deposits then flooded in and the ratio collapsed to 75.6% by the third quarter of 2020 and to a record-low 68.8% by early 2021. Credit unions did not need funding, and 2020 produced no credit union securitization at all. As deposits drained back out and the ratio climbed above 81% by late 2022, seven deals priced in 2023 and the pioneer returned. Through 2023 the instrument behaved exactly as a liquidity valve should. Section IV is about what happened after that. Read the table by its right-hand column and the pattern is hard to miss. First deal, a second issuer, the inflection year that also produced the first repeat issuances, the first cooperative structure, the first institution under $1 billion, the first collateral that was not a car loan, and the first stated commitment to keep going. Seven doors in under seven years, and the only year that produced nothing at all was 2020. ## II · Why this instrument is different in kind The trade press files securitization under loan sales, continued, as though it were a faster participation. The differences are the whole argument, and there are four of them. Chart 3 · Where the money comes from Schematic. A participation moves paper inside the credit union deposit base. A securitization imports funding from outside it. Schematic, drawn to show the funding path rather than any deal's economics. The trust structure and its statutory protection are described in the text. ### The buyer is a different economy A participation buyer is another credit union, funded by its own members' deposits. So the participation market's capacity is capped by the deposit health of the credit union industry itself. When deposits tighten system-wide, sellers multiply and buyers disappear in the same quarter, which is what happened in 2022 and 2023. A securitization buyer is an insurance company, an asset manager, a bond fund. Selling a participation redistributes paper within the industry's funding base. Issuing a bond brings new water into the pool. ### The legal architecture is stronger, and the weak version is in court right now A participation is a contract, and its central promise, that the buyer truly owns its share of the loans, has a known failure mode: recharacterization. If a court decides the sale was in substance a disguised loan, the buyer holds nothing but a claim against a bankrupt seller. That is a live dispute, not a law-school hypothetical. It sits at the center of the PrimaLend Chapter 11 covered in [Issue 9](https://lendriskanalytics.com/insights/the-cash-problem.html), where participations sold to an affiliated entity are contested as true sales versus disguised financings. The credit union industry has its own scar tissue on the same point. Taxi medallion participations spread the losses of a few New York originators across credit unions nationwide during the 2010s. Melrose Credit Union was placed in conservatorship on February 10, 2017 and liquidated on August 31, 2018, and the NCUA's Share Insurance Fund absorbed more than $750 million from medallion-related failures, a cost borne by credit unions that never originated or bought a medallion loan. The NCUA's own participation guidance is unusually blunt about the mechanism, warning that a seller "can create systemic underwriting and loan servicing risk to a large group of credit unions," and prescribing, in its words, caveat emptor and diversification. A securitization runs instead through a true sale into a bankruptcy-remote trust that the 2017 safe harbor addresses directly, with rated tranches, credit enhancement and a trustee. The structure is engineered against precisely the failure the participation market is litigating. ### The disclosure regime is heavier, though still lighter than the public market's Credit union ABS deals have come to market as Rule 144A private placements, so the loan-level Schedule AL disclosure that registered auto ABS files on Form ABS-EE does not apply. What does apply is rating-agency presale scrutiny and ongoing surveillance, investor reporting, and Rule 15Ga-2 findings filings on EDGAR. The Corporate One deal drew ratings from both S&P and KBRA. PenFed files ABS-15G through PenFed Auto Receivables Funding, LLC. That is far more third-party examination than any participation receives, where the buyer's entire visibility is the tape the seller hands over. Baker also put a number on the thing that makes participations expensive in a way the dollar figures never show. Participations, he said, "take about as much time for each transaction regardless of size," because "it takes time to search for and connect with each participating financial institution." That is the operative constraint. A $1 million participation and a $23 million one cost roughly the same in effort, and GTE did both, so the average deal carried a fixed overhead the seller could not amortize away. A shelf inverts that: the setup cost is heavy once and near zero thereafter. ### The machine runs again A shelf, once built, is repeatable. PenFed has now used its four times and said on the record that it intends to keep using it. Corporate One closed its first multi-seller deal in January 2026 and has stated an ambition of two to four transactions a year. Nobody builds a machine to use it once. **Inference** The instrument-level differences all point the same way. Securitization gives an issuer an exit from the participation market's structural constraints: its deposit-capped buyer base, its recharacterization risk, its bilateral grind. Institutions that have made that exit have little reason to walk back through the door, and so far none of them has. ## III · What pushed them through the door The trigger was cyclical and the record on it is clean. Between early 2021 and late 2022, the system loan-to-share ratio moved from a record-low 68.8% to above 81%, as loan growth ran near 19% in 2022 while pandemic-era deposits drained away. Members moved cash to Treasuries and higher-yielding accounts, credit unions repriced certificates upward to slow the outflow, and cost of funds jumped. Underneath the funding squeeze sat the asset side of the trap: balance sheets full of 2021-vintage auto and mortgage paper written at the low rates of that year, performing perfectly and completely immobile. The loans were fine. They were also fuel that could not be burned twice. This brief does not attach a specific term structure to that vintage, because it has not sourced one. The participation data from the peak shows both the surge and the strain. Credit unions sold $22.7 billion of non-real-estate loans in 2021, up 69% from $13.5 billion in 2020. But the top ten sellers accounted for $13.8 billion of that, and PenFed alone sold $6.4 billion, up nearly fourfold from $1.6 billion the year before. Oregon Community Credit Union, a $2.8 billion institution, sold $1.6 billion in 2021 against $143.3 million in 2020, an eleven-fold jump. Distribution was already concentrating in the hands of a small number of institutions, and several of them, PenFed and Oregon Community among them, went on to become securitizers. The capability was being built inside the old instrument before it moved to the new one. Then there is the slower force underneath the cycle, and it is the one that does not reverse. The count of federally insured credit unions fell from 4,411 to 4,250 in the year to the first quarter of 2026, roughly 160 institutions in twelve months, with essentially nothing chartered to replace them. New charter formation collapsed to near zero across the 2000 to 2020 period while closures continued at roughly 100 to 200 a year, and the net count has been negative since the mid-1980s. The institutions disappearing are the small ones. NCUA treats a credit union under $100 million in assets as small, and 56% of 2026 mergers to date involved institutions under $50 million. That is the profile of the traditional participation buyer: more deposits than local loan demand, and no capacity to originate at scale. Two things cut against the tidy version of this and both belong on the record. First, merger approvals have slowed rather than accelerated, from a peak of 263 in 2014 to 157 in 2025, with 27 in the first quarter of 2026 against 35 in Q1 2025 and 26 in Q1 2024. Second, and more importantly, these are mostly not failures. In the first quarter of 2026 the NCUA approved 22 mergers for expanded services, three for inability to find officials and two for poor financial condition, and the expanded-services share has risen from 69% in 2025 to 82%. Healthy institutions are choosing to combine. **Inference** The reason does not change the arithmetic. Whether a small credit union merges from weakness or from strategy, it stops being available as a participation counterparty either way, and no new charter arrives to take its place. That is why this is a structural claim rather than a distress claim, and it holds even in the healthy scenario the merger data actually describes. **Inference** Even a full liquidity normalization does not restore the old model, because the old model's counterparties are fewer every quarter. The cyclical trigger opened the door. The demographic trend removes the road back. ## IV · The divergence that carries the argument Here is the frame this brief exists to put on the record, and it is labeled as interpretation throughout. If issuance were purely a liquidity valve, it should be closing. The pressure that opened it has eased. Loan-to-share fell to 81.5% in the first quarter of 2026 from 81.8% a year earlier. System net income ran 30.5% above the prior year at $20.4 billion annualized. Total assets reached $2.48 trillion. By any ordinary reading, a well-capitalized, profitable, re-liquefying system does not need to distribute. Instead the valve widened. The first half of 2026 produced roughly $1.4 billion, against $2.1 billion for the whole of 2025 and a $2.6 billion record for 2024. Two thirds of a full year's volume in half a year, and it came with two categories that had never existed: a multi-seller deal carrying an institution under $1 billion in assets, and a securitization backed by something other than car loans. Continue that pace and 2026 clears the record. This brief does not annualize it, because the second half has not happened. Chart 4 · Four years of tracking, then a break System loan-to-share above, credit union auto ABS activity below, on a shared timeline from the first deal. Through 2023 the two move together. After 2023 they separate, and that separation is the argument. Loan-to-share, all NCUA quarterly system data: Q4 2019 at 84.1%, Q3 2020 at 75.6%, Q1 2021 at 68.8% (record low), above 81% by late 2022, Q1 2025 at 81.8%, Q1 2026 at 81.5%. Points are published readings; the segments between them are connections, not data, and the ratio is seasonal, so first-quarter readings compare like with like. Lower panel as in Chart 1, with the same unit caveat: 2019 through 2023 are transactions, 2024 and 2025 are issuers, and no count is published for the first half of 2026. The 2026 bar is half a year and is drawn as such. What it establishes is that the valve stayed open, and widened, after the pressure dropped. That is what a ratchet looks like in its early years. ### The pattern this cycle keeps producing Subprime auto has been generating series that rise under stress and settle above their old floor rather than round-tripping. Extension usage is the cleanest example, and [Issue 8](https://lendriskanalytics.com/insights/current-on-paper.html) documented it: the tool spread under COVID, servicers built the capability, and usage never returned to its pre-pandemic level even after the emergency passed. America's Car-Mart ran a modification program across 28.9% of its receivables before disclosing it. In each case a behavior adopted under pressure became the baseline once the pressure passed, because the institutions had built the capability, normalized it internally, and had no incentive to unbuild it. **Inference** Credit union funding is joining that list, and the divergence above is the strongest single piece of evidence for it. The cyclical reading has to explain why the valve stayed open in 2025 and widened in 2026 while every pressure gauge improved. It has not yet. ### The scale objection answered itself The multi-seller structure deserves its own paragraph, because it dissolves the objection that used to be the whole case against this thesis. In June 2014, then-NCUA Chairman Debbie Matz put the constraint plainly: "Most credit unions do not yet originate enough loans to sponsor securitizations, but for those that do, it is prudent to propose specific safety and soundness provisions." True then, and still true now, and now beside the point. The Alloya trust pooled $50 million each from three credit unions, none of which could have issued alone at scale. Six months later, Corporate One's $335 million transaction carried Day Air, a credit union under $1 billion in assets, into the ABS market alongside Wright-Patt and Everwise, with Stifel and Bank of America as joint leads and ratings from both S&P and KBRA. What was a structural barrier in 2014 became a pooling problem in 2025, and the corporate credit unions solved it the way aggregators always do. **Inference** The multi-seller rail is the mechanism by which a mega-institution capability propagates down-market. Its existence is the difference between "a dozen big credit unions found a niche" and a template the other four thousand can reach. The supervisor has noticed. The NCUA told attendees at its Capital Markets Symposium that securitization "is essential for larger credit unions to manage their balance sheets and tap into a broader investor base." An infrastructure layer has formed around the flow: corporate credit unions running the trusts, underwriters from Stifel to J.P. Morgan to Bank of America, and two law firms with assembly-line deal practices. Infrastructure is the most durable form of institutionalization there is. It does not dismantle itself when a ratio improves. ## V · The counter-case, taken seriously Four objections deserve a straight answer, because the honest version of this thesis is narrower than the trade press version. 1 **The dollars are tiny.** Roughly $8.4 billion cumulative against a $1.73 trillion loan book is under half of one percent. Correct, and this brief leads with it. The claim is about direction and structure, not stock. 2 **The trigger was cyclical, so the behavior may be too.** This is the strongest objection, and the Q1 2026 data genuinely supports it. A profitable, re-liquefying system does not need to distribute. The answer is the divergence in Chart 4 plus the demographic point: the participation alternative's buyer base shrinks regardless of the cycle. If issuance falls back as loan-to-share does, the cyclical reading wins. 3 **Real distribution is old news.** Also correct. Credit unions have sold conforming mortgages into the secondary market through the GSE channel for decades, and that flow dwarfs the ABS numbers. What is new is distribution of consumer collateral, autos above all, under the credit union's own name, into capital markets, on repeatable rails. That did not exist in any form before November 2019. 4 **Fewer than a dozen institutions have done it, and the loudest voices are paid to cheer.** Both true. Standalone issuance remains a large-credit-union activity. And the sources describing a boom, the law firms, the platforms and the corporate credit unions, all earn fees from more of it. Their deal facts are reliable and checkable; their forecasts are advocacy, and this brief weights them accordingly. The academic counterweight exists and it cuts cautionary. A 2026 study in *Risks*, using US credit union call report data from 1994 to 2024, finds that credit unions with heavier non-core lending exposure grow faster in loans and membership but carry weaker financial buffers, including lower net worth ratios, alongside higher delinquency. The sharpest finding in it is directly on point for this brief: of the non-core categories the authors test, **loans held for sale** show the strongest association with adverse buffer and asset-quality patterns, more so than purchased loans or lease receivables. Distribution is a capability, not a virtue. An originate-to-distribute credit union is a different risk animal from an originate-and-hold one, and the call report data already says so. Falsifiable, both directions This reading is wrong, and the cyclical reading right, if annual credit union auto issuance falls below roughly $1.5 billion for two consecutive years while loan-to-share sits below 80%, meaning the tool gets shelved exactly when it is least needed. It becomes hard to argue with if sustained annual issuance clears $4 to $5 billion with fifteen or more distinct issuers, or if multi-seller deals become quarterly and routinely carry institutions under $1 billion in assets. **The next two years of data land between those markers, and this brief will be revisited against them.** ## Proof: every figure in this brief, traced Each numeric or quoted claim in the body appears below with the source it came from and its verification status as of August 23, 2026. Verified means the figure was located at the named source and matches. Derived means it was calculated here from a verified figure, with the arithmetic shown. Scope-split means sources report different numbers because they are counting different universes, and the brief says which. Corrected marks a figure an earlier draft got wrong and this one fixes on the record. Source-named marks a quotation attributed to a specific named article that could not be retrieved automatically to re-check the wording. | Claim as stated | Source | Status | |---|---|---| | 25 credit union securitizations, ~$8.4 billion cumulative; Dechert advised on 22 of them | Dechert OnPoint, February 2026 | Verified | | 2023: seven transactions, $2.03 billion, +167% year over year | Dechert OnPoint, February 2026 | Verified | | Counts by year: 1 / 0 / 1 / 2 / 7 transactions (2019–2023), then 6 and 6 issuers (2024, 2025); no count published for H1 2026 | 2019–2022 sum of four matches Dechert's stated count exactly; 2023 per KBRA and Dechert; 2024 per KBRA (six credit unions priced, $2.4B, titled "Record-High 2024 Issuance"); 2025 per S&P (six issuers). Units are not interchangeable and are marked as such in the charts | Verified | | H1 2026: four individually confirmable transactions (Corporate One January, Lafayette Federal February, Oregon Community 2026-1, PenFed June); $1.4 billion aggregate | Deal sources as listed elsewhere in this table; Asset Securitization Report on OCCU 2026-1 ($307.33M, seven note classes, KBRA-rated, OCCU's fourth deal); aggregate per Credit Union Times, July 6, 2026. An earlier draft claimed six H1 2026 transactions including First Community and Space Coast; Space Coast's $700 million deal was August 2025, not 2026, and the First Community deal could not be dated to 2026. Both were removed | Corrected | | 2024: $2.4 billion across six credit unions (KBRA, "Record-High"); Dechert independently ~$2.4 billion; S&P $2.07 billion across five in its rated-auto universe, +20%, Suncoast and PenFed $1.12 billion combined | KBRA ; Dechert OnPoint; S&P Global Ratings | Scope-split | | 2025: $2.1 billion across six issuers, +1.6% on 2024 | S&P Global Ratings , via trade coverage | Verified | | First half of 2026: ~$1.4 billion pace | Credit Union Times, July 6, 2026 | Verified | | GTE Financial, November 2019, first credit union securitization; ~9,000 loans, $185.27M collateral; Stifel | Asset Securitization Report ; American Banker | Verified | | GTE 2019-1: deal named GTE Auto Receivables Trust 2019-1, structured as a Rule 144A transaction via Stifel Nicolaus; money market $36.3M, A-2 $58.19M, A-3 $52.94M (A classes $147.4M); subordinates $22.3M and $5.3M; total notes $175.0M; pool $185.27M; ~9,000 loans; 5.54% overcollateralization; weighted-average non-zero FICO 727; ~98% Tampa-area concentration; sponsor a $2.2 billion institution | Asset Securitization Report ; American Banker . Trust name, 144A structure and Stifel role all stated in ASR; $2.2 billion asset size stated in the same coverage | Verified | | GTE repeat issuance was April 2023 at $202.8 million, inside the 2023 count, not a 2020 deal | Auto Finance News, April 2023 ; Dechert, October 2023 on GTE and OCCU as second-time issuers | Corrected | | GTE $2.2 billion in assets at the time of the first deal | American Banker | Verified | | GTE sold ~$202 million of auto loans via participations 2014–2018, in pieces of $1M–$23M; CFO Brad Baker | CU Management, July 2020 | Verified | | Baker: the securitization "increased our liquidity, leveled out our concentration of auto loans and provided transactional income, as well as providing a higher rate of servicing income throughout the life of the security" | CU Management, July 2020 . Recovered verbatim through search indexing of that article; the page itself blocks automated retrieval | Source-named | | Unify Financial, March 2021: $300 million offering of prime auto loans, second credit union ever to sponsor an auto ABS, issued through UNIFY Auto Receivables Trust 2021-1; ~85% of the pool with final payments 75 to 84 months from origination | Credit Union Times, March 26, 2021 ; Asset Securitization Report ; CU Today ; Cadwalader | Verified | | No credit union securitization priced in 2020 | Dechert counts four deals across 2019–2022, which the named roster (GTE 2019, Unify 2021, PenFed and OCCU 2022) accounts for exactly | Derived | | GTE returned April 24, 2023 with a $202.8M deal, three and a half years after the first; Stifel and BofA Securities ran four 2023 credit union deals (GTE and OCCU repeating, Veridian and GE Credit Union debuting) | Auto Finance News, April 2023 ; Dechert, October 2023 | Verified | | Baker on participations: they "take about as much time for each transaction regardless of size" because "it takes time to search for and connect with each participating financial institution" | CU Management, July 2020 . The publisher blocks automated retrieval; the wording is as indexed and as supplied, not re-verified at the page | Source-named | | NCUA safe harbor published June 30, 2017, effective July 31, 2017, 12 CFR Part 709 | Federal Register, 82 FR, June 30, 2017 | Verified | | 2017 NCUA opinion on incidental authority under FCUA §107(17) | Dechert OnPoint, February 2026 | Verified | | Alloya Auto Receivables Trust 2025-1: $150M, Blaze, Consumers and Interra at $50M each, seven tranches of class A, B, C and D notes, first multi-issuer credit union auto ABS; subordination of 13.65% / 9.10% / 3.60% / 0% on classes A through D | Asset Securitization Report , citing S&P Global Ratings; S&P presale . The seven-tranche figure is the note structure and is unrelated to any oversubscription claim, which this brief does not carry | Verified | | Corporate One: closed January 21, 2026; 16,607 loans, $335.1M pool; Wright-Patt, Everwise, Day Air; S&P and KBRA; Stifel and Bank of America joint leads; two-to-four deals a year stated | Auto Finance News ; Corporate One release ; CU Times | Verified | | Day Air is the first credit union under $1 billion in assets inside an ABS | Corporate One / Cadwalader coverage | Verified | | Lafayette Federal MRQI 2026-HI1, $459M home improvement, closed February 3, 2026, first non-auto collateral | Dechert OnPoint, February 2026 | Verified | | PenFed 2026-A: $354M, closed June 22, 2026, fourth deal; J.P. Morgan structuring lead, Goldman joint lead, CIBC co-manager | PenFed release via PR Newswire | Verified | | Heintzman: "continue establishing PenFed as a programmatic issuer and leveraging securitization as a tool to help us serve our members by diversifying liquidity and funding options" | PenFed release, June 2026 | Verified | | PenFed files Form ABS-15G through PenFed Auto Receivables Funding, LLC | SEC EDGAR | Verified | | Q1 2026: 4,250 federally insured credit unions, down from 4,411; loans $1.73T; assets $2.48T; net income $20.4B annualized, +30.5%; loan-to-share 81.5% from 81.8% | NCUA, June 2026 | Verified | | Loan-to-share 68.8% record low in early 2021, above 81% by late 2022 on ~19% loan growth | NCUA data via TruStage Credit Union Trends Report | Verified | | Participations: $22.7B of non-real-estate loans sold in 2021, up 69% from $13.5B in 2020; top ten sellers $13.8B against $4.8B the prior year; PenFed $6.4B from $1.6B; Oregon Community ($2.8B in assets at the time) $1.6B from $143.3M | Credit Union Times, Jim DuPlessis, March 25, 2022 . All figures including both base years are in that article; the $2.8 billion is OCCU's asset size as of that reporting, and it has grown since | Verified | | NCUA participation guidance: a seller "can create systemic underwriting and loan servicing risk to a large group of credit unions"; caveat emptor and diversify | NCUA, Evaluating Loan Participation Programs | Verified | | Melrose Credit Union conservatorship February 10, 2017; liquidated August 31, 2018 | NCUA; American Banker ; CU Times | Verified | | Share Insurance Fund absorbed more than $750 million from medallion-related failures | NCUA medallion FAQ | Verified | | Matz, June 2014: "Most credit unions do not yet originate enough loans to sponsor securitizations…" | NCUA, June 2014 | Verified | | NCUA at its Capital Markets Symposium: securitization "is essential for larger credit unions to manage their balance sheets and tap into a broader investor base" | ORSNN, republishing Asset-Backed Alert coverage | Verified | | Loan-to-share 84.1% in Q4 2019, 75.6% in Q3 2020, 68.8% record low in Q1 2021 | NCUA quarterly credit union data summaries for the respective quarters; NCUA Quarterly U.S. Map Review | Verified | | NCUA treats institutions under $100 million as small; 56% of 2026 mergers to date involved credit unions under $50 million; Q1 2026 approvals split 22 for expanded services, three for inability to find officials, two for poor financial condition; expanded-services share rose from 69% in 2025 to 82% | CUCollaborate on NCUA Q1 2026 Merger Activity and Insurance Report ; Credit Union Times | Verified | | Merger approvals: 263 in 2014, 157 in 2025, 27 in Q1 2026 against 35 in Q1 2025; new charter formation near zero since 2000 | CUCollaborate ; CU Times ; NCUA chartering and merger data | Verified | | Risks 2026: higher non-core exposure associated with faster loan and membership growth but weaker financial buffers, "including lower net worth ratios and weaker economic solvency, alongside higher delinquency"; loans held for sale show the strongest adverse buffer and asset-quality association, ahead of purchased loans and lease receivables | Risks 14(2):32, MDPI , US call report data 1994–2024 | Verified | | Car-Mart modification program across 28.9% of gross receivables | Car-Mart FY2025 Form 10-K, as documented in Issue 8 | Verified | | PrimaLend participation recharacterization dispute | Bankr. N.D. Tex., Case 25-90013, as documented in Issue 9 | Verified | Claims that could not be placed in this table were removed from the brief rather than softened. Figures cut for lack of verification include a 2022 Federal Home Loan Bank borrowing count, a vendor survey of executives "actively securitizing", a national total for taxi medallion participations, loss performance on a specific PenFed transaction, oversubscription multiples on the Alloya deal, lifetime volumes for corporate credit union participation platforms, a derived 2022 dollar figure, and a phantom GTE transaction dated April 2020 that was a misdating of its April 2023 return. **Sources & notes** **System data.** NCUA quarterly system performance release, Q1 2026 (June 2026). NCUA Quarterly Credit Union Data Summary 2026 Q1. TruStage Credit Union Trends Report for the loan-to-share low and 2022 loan growth. NCUA chartering and merger activity data, with CUCollaborate and Credit Union Times analyses of quarterly merger approvals. **Deals.** Dechert OnPoint, "Credit Union-Sponsored Securitizations: Market Developments and Legal Considerations" (February 2026), for the cumulative count, the 2023 inflection and the Lafayette Federal transaction; Dechert advised 22 of the 25 deals it counts, noted throughout as an interest. S&P Global Ratings for 2024 and 2025 credit union auto ABS issuance in its rated universe, and for the Alloya presale. KBRA, "Credit Union Auto ABS: Record-High 2024 Issuance," and "Credit Union Auto ABS: Continuing Momentum" (March 2025), for the market-wide 2023 and 2024 counts and totals. Asset Securitization Report and American Banker on GTE. ORSNN, republishing Asset-Backed Alert, on GTE's two deals and the NCUA Capital Markets Symposium statement. Auto Finance News, CUInsight, Cadwalader and Credit Union Times on Corporate One. PenFed press releases and SEC EDGAR filings on PenFed 2026-A. CU Management (July 2020) for Brad Baker's participation history and quotation. Credit Union Times (July 6, 2026) for the first-half 2026 pace and the 2024 record figure. **Regulatory.** NCUA securitization safe harbor, Federal Register, June 30, 2017, effective July 31, 2017, 12 CFR Part 709. NCUA Office of General Counsel opinion on incidental authority under FCUA §107(17). NCUA Supervisory Letter and guidance, "Evaluating Loan Participation Programs." NCUA Board materials, June 2014, for the Matz statement. NCUA taxi medallion FAQ and related statements for the Share Insurance Fund loss and the Melrose and LOMTO actions. **Participations.** Credit Union Times, Jim DuPlessis (March 25, 2022), for 2020 and 2021 sale volumes and the seller concentration figures. **Counter-case.** *Risks* 14(2):32 (MDPI, 2026), "Mission Drift or Strategic Expansion? Non-Core Lending, Risk, and Capital in US Credit Unions," on 1994–2024 call report data. **Companion issues.** Issue 8 covers extensions and the reported-versus-actual gap. Issue 9 covers the 2023–2026 failure roster as cash events, including PrimaLend. Where this brief reasons beyond what a document states, it is labeled as an inference. Figures that could not be tied to a named source are excluded rather than softened, and the verification table below marks the status of every one that remains. Point-in-time reading of the public record through August 23, 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. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html) LR LendRisk Analytics Independent market research Related [Issue 8 Current, on paper.](https://lendriskanalytics.com/insights/current-on-paper.html) [Issue 9 The cash problem.](https://lendriskanalytics.com/insights/the-cash-problem.html) --- title: "The cash problem" url: https://lendriskanalytics.com/insights/the-cash-problem.html publisher: LendRisk Analytics series: What the Tape Said issue: 9 published: 2026-08-17 kind: Research brief description: "In buy-here-pay-here, cash leaves the day you sell the car and comes back over four to five years. Seven failures from 2023 to 2026, re-read as cash events: American Car Center, U.S. Auto Sales, Tricolor, PrimaLend, Automotive Credit, FinBe, and America's Car-Mart. Every figure sourced; inferences labeled." html: https://lendriskanalytics.com/insights/the-cash-problem.html --- # The cash problem [← All insights](https://lendriskanalytics.com/articles.html) What the Tape Said · Issue 9 Research brief · 14 min read The 2023-2026 failure roster, re-read as cash events # The cash problem. Public record through August 17, 2026 · LendRisk Analytics In buy-here-pay-here, a company can be profitable on paper and dead at the bank, and that is how the model works rather than a paradox in it. Cash leaves the day you sell the car. Cash comes back over the next four to five years. Grow, and the gap widens. Seven operators failed or nearly failed between 2023 and 2026, and every one of them ran out of cash or ran out of a lender willing to renew. 28% Bad debt as a share of sales in 2024, up from 21% in 2022 (SGC benchmarks) $616K New receivable a 40-car-a-month store commits every month, at the Fed's $15,402 average 7 Failures or near-failures, 2023-2026, all triggered by funding, not demand +150% Rise in banks' reported default probability on BHPH borrowers in one quarter ## Bottom line Across the roster, American Car Center, U.S. Auto Sales, Tricolor, PrimaLend, Automotive Credit Corp, FinBe, and America's Car-Mart, the proximate cause of death was cash exhaustion or a funding facility that would not renew. Customer demand held up in every case. Car-Mart's own CEO called fiscal 2026 "a liquidity and capital-structure story, not a credit-quality one," and the company's collections rose 2.2% in the same year its going-concern warning printed. Credit stress supplied the shock. Subprime 60-plus delinquency hit a record 6.90% in January 2026, then normalized to 6.11% at the March tax-refund trough and 5.67% by June, down 64 basis points from a year earlier. Bad debt at the benchmark operators jumped from 21% of sales to 28% in two years and stayed there. That shock reached every operator in the sector. The ones that died were the ones whose funding could not absorb it and whose reporting gave them no warning it was coming. The read Every BHPH sale consumes cash today against collections spread over roughly 55 months. That makes growth itself a drain: the model burns cash fastest exactly when the sales numbers look best. **Credit stress decides how hard the wave hits. Funding architecture and cash visibility decide who drowns.** And almost no small or mid-size operator runs the liquidity monitoring that would flag the problem while it is still fixable. ## The arithmetic of one deal The operator buys the car at auction, pays for reconditioning, sales tax, and commissions, and puts it on the lot. All of that is cash out on day one. Then the operator finances the sale in-house and collects it back one payment at a time. Per the Federal Reserve's May 2026 FEDS Note, the average BHPH subprime origination carries a principal balance of $15,402 on a 55-month term at a 25.39% weighted-average rate: a longer, larger, higher-rate contract than traditional subprime auto paper. Across the twelve portfolios in the Cash Clock section below, the cash buried in a typical deal runs about $8,300 on average, with per-book averages from roughly $1,700 to $13,400. One deal's cash position, month by month Illustrative deal: about $6,100 of cash out at sale, recovered through payments over a 55-month contract at a realistic pay pace. The deal spends its first year and a half underwater. Illustrative, not a real account: about $6,100 cash in deal, inside the roughly $1,700 to $13,400 per-deal range observed across the twelve portfolios in the Cash Clock section below, on the Fed note's 55-month average term at a realistic pay pace. Every new sale opens another one of these curves. Now multiply. A store selling 40 cars a month opens 40 of those curves every month, each one underwater for its first year and a half. That is why the capital required to run an identical store has climbed so steeply since 2019. The unit economics did it: the Fed puts the average BHPH origination at $15,402 over a 55-month term, and a store writing 40 of those a month is committing roughly $616,000 of new receivable every month before a dollar of it comes back. Nothing about the customer changed. The cash the same storefront has to carry did. Then the credit cycle arrived on top of it. SGC CPAs, the Houston firm that has tracked BHPH benchmarks since 1999, reports bad debt climbing from 21% of vehicle sales in 2022 to 24% in 2023 and 28% in 2024, the worst reading in its published table. SGC footnotes its composite as drawn from "the best performing operators in the industry," so read every figure in it as the best-case path; the unmonitored average operator sits somewhere below it, on a series nobody publishes. Sales stayed strong. What broke was bad debt, which jumped from 21% of sales in 2022 to 28% in 2024. And SGC partner Steven Carstens named a danger that is funding rather than credit: "the greatest threat facing the industry right now is the number of lenders leaving the space." The same report tells dealers to focus on "reducing debt levels and running lean." Its prescription was pure liquidity management: reduce debt and run lean, "even if that means contracting in size." Bad debt as a share of vehicle sales, benchmark operators SGC's composite of some of the industry's best performers. The average operator's numbers are worse than these. Source: SGC CPAs (Shilson, Goldberg, Cheung & Associates), 2024 Buy Here-Pay Here Financial Benchmarks. **Inference**Put the three facts together: a structurally cash-hungry model, a rising capital requirement per unit of capacity as origination balances grew, and the banks that fund the sector walking out. The marginal operator in 2024-2026 faced a liquidity squeeze independent of its own credit performance. An operator who held charge-offs flat still needed materially more cash to run the same book, and found it harder to borrow. ## The seven failures Line the roster up by date and read each one for what pulled the trigger. FEB 2023 **American Car Center closes in a day.** The chain, counted at roughly 40 to 50 dealerships across about ten states in contemporaneous coverage, pulled a subprime bond sale reported at $222 million from the market, then told employees the next day it was shutting its doors. Chapter 7 followed on March 14. The loan book did not vanish overnight. The funding did. APR-AUG 2023 **U.S. Auto Sales follows the same script.** The 39-dealership chain closed its stores in April 2023; Westlake took over servicing of the roughly $740 million portfolio within weeks, and the company was subsequently liquidated, with contemporaneous reporting warning its securitizations risked the first principal losses in auto ABS in decades. AUG 2025 **Automotive Credit Corp pauses originations "indefinitely."** The 33-year-old Michigan lender cited "internal and external financial conditions" while insisting "this is not a closure of business." Halting new loans is the classic liquidity defense: stop the cash going out and hope collections carry the book. SEP 2025 **Tricolor files Chapter 7.** Prosecutors allege ~$2.2 billion pledged against ~$1.4 billion of real collateral. Double-pledging is what a borrower does after running out of unpledged collateral to borrow against. It is the fraudulent version of running out of borrowing base. OCT 2025 **PrimaLend files Chapter 11** with a credit facility maturing two days after the petition, after over-advances on its CIBC facility in August 2024 and again in January 2025, a ~$34 million participation sale to cure the first one, and a skipped interest payment on its $75 million notes. DEC 2025 **FinBe USA stops originating** after its late-December sale by Bepensa Capital. KBRA raised projected losses on its 2025 securitization to 23.5% from 21.24%, and Westlake was added as subservicer in August 2026. FY2026 **America's Car-Mart posts a $139 million loss with going-concern doubt**, closes 60 of 154 dealerships, and operates under a covenant waiver running to September 7, 2026. Its collections rose. Its funding structure failed. ### Tricolor ran out of collateral it had not already pledged The fraud allegations describe a company borrowing against collateral it no longer had free. The banking damage was real and specific: JPMorgan booked a $170 million charge-off, with CEO Jamie Dimon telling the bank's October 2025 earnings call, "when you see one cockroach, there's probably more." Fifth Third disclosed a $178 million non-cash impairment on its Tricolor exposure. Barclays took an impairment of £110 million, about $148 million, and Origin Bancorp reported $29.5 million of net charge-offs tied to the fraud in its FY2025 10-K. Two former executives pleaded guilty in December 2025; the former COO pleaded guilty in June 2026. ### PrimaLend shows how the squeeze reaches a single-lot dealer PrimaLend's First Day Declaration is the clearest public description of how the squeeze travels down to the single-lot dealer: as consumers default, the dealer loses collections, the delinquent contracts become ineligible collateral on the dealer's borrowing-base line, the line shrinks, and the dealer cannot buy inventory. Each step tightens the next. PrimaLend held roughly $280 million of dealer loans, most of them revolving lines secured by the dealers' own paper, against funded debt reported at roughly $286 million. ### Car-Mart swapped a revolver for a term loan and lost its flexibility The pivotal transaction was October 30, 2025. Car-Mart raised a $300 million five-year term loan from Silver Point at SOFR plus 7.50% with warrants, used part of it to repay and retire its asset-based revolver, which carried roughly $163 million outstanding per contemporaneous reporting. A revolver flexes with cash needs. A term loan is a fixed slug of debt. Car-Mart traded the first for the second and left itself with no revolving liquidity at all, a structure its own FY2026 10-K later flagged as the core vulnerability. By June 2026 the company was cycling through weekly forbearance extensions with covenant floors of $5-7 million of minimum liquidity and a 1.20-1.25x collateral coverage ratio, plus up to $18 million in waiver fees. Those are exactly the liquidity metrics this brief argues operators should watch. Car-Mart got them imposed by lenders, after the fact, at a price. ## The monitoring gap The discipline for watching this risk already exists. Restructuring practice runs on the 13-week cash flow forecast: weekly cash in minus weekly cash out, projected forward, with the lowest point flagged and a variance record kept. Courts and creditors demand it in Chapter 11, which is precisely why the operators who build their first one in bankruptcy have already lost. Rating agencies assess finance companies on funding diversity, unencumbered assets, and coverage. Banks build their own protections into BHPH facilities: per the Fed note, 81% of loans to BHPH dealers are guarantor-backed, 65% are asset-based against over-collateralized receivables, and advances to Tricolor ran at only 60-80% of collateral value. Every one of those tools protects the lender. None of them is something the dealer runs on itself. A DMS produces sales reports, collections reports, and delinquency agings. It does not produce a runway number, a collections-to-obligations coverage ratio, or the distance to the over-advance line, the single number PrimaLend's dealer-borrowers were not watching. The gap between what the discipline requires and what the tooling delivers is the finding. | Operator | Scale | Funding structure at failure | Trigger | |---|---|---|---| | America's Car-Mart | ~$1.28B revenue | Term loan + securitizations, no revolver | Liquidity and collateral-coverage covenants | | Tricolor | ~$1.4B real collateral | Multiple warehouse lines + ABS | Ran out of unpledged collateral | | PrimaLend | $286M funded debt | Two bank facilities + unsecured notes | Over-advances; maturity one day post-filing | | American Car Center | ~40 dealerships | Subprime ABS | $222M bond deal pulled; window shut | | Single-lot dealer | One rooftop | One borrowing-base line | Delinquent paper turns ineligible; line shrinks | The mechanism is identical across four orders of magnitude. What changes with scale is optionality: Car-Mart had months of warning, an investment bank, and asset sales. The single-lot dealer has none of those, and no early-warning system either. ## The counter-case, taken seriously Three objections deserve a straight answer. First, volume really is down: Car-Mart's units fell 14.3% in FY2026, and Cox Automotive projects 15.8 million new-vehicle sales this year, a decline it deepened in June to 2.9% from January's 2.4% while holding the volume forecast. Second, liquidity sits downstream of credit: collections fell because borrowers were stressed, and the record 6.90% subprime delinquency print is real. Third, survivorship: plenty of operators run the identical originate-and-hold model and are still funding. All three are true. None overturns the finding. Car-Mart's volume decline was, on its own numbers, largely self-inflicted: it cut inventory purchases in half to conserve cash while collections rose. Credit is the shock, but the shock was sector-wide while the deaths were concentrated, and mortality tracked funding structure rather than credit performance. And the survivors are disproportionately the operators with diversified funding, revolving liquidity, and scale, which is the thesis confirmed rather than contradicted. **Inference**The working causal model has two factors. Credit stress is the shock, and it is mostly outside an operator's short-run control. Liquidity structure decides survival, and it is something the operator chooses and can watch. That is why the cash framing is more useful to an operator than the credit framing: it points at the levers the operator holds. ## What the clock says on real books Everything above says cash timing decides who survives. This is what cash timing looks like inside real books: twelve BHPH dealer portfolios across eight states, taken from monthly dealer benchmark reporting for June 2026. Cash Clock · twelve dealer portfolios, eight states, June 2026 · identities withheld 3.3 mo Fastest book gets its cash back 19.4 mo Slowest book, same month 5.8× Gap between them $13,423 Cash buried per unit at the slowest, against $8,328 average Same industry, same month, same kind of customer. The fastest operator has its money back in 3.3 months. The slowest waits 19.4. That is almost six times the difference in how long cash sits out the door. **The gap is set when the car is bought, not when it is collected.** The slowest book has $13,423 of its own cash in every unit against a group average of $8,328. That is roughly five thousand dollars a car, on every car, and no amount of collections pressure changes it. It was decided at the auction, in the down payment, and in the term. That is the number the seven failures were losing on, and none of them published it. It takes no new system. It is arithmetic on data every operator already keeps. [Run the clock on your own book →](https://getbookiq.com/cash-clock.html) ## What a right-sized liquidity stack looks like None of this requires a treasury department. It is six numbers, kept weekly, before a lender asks you for them. Each one answers a question you would otherwise be guessing at, and each one has an action attached to it. | The number | How to work it out | What to do when it trips | |---|---|---| | Weekly cash forecast How much cash will I have in each of the next thirteen weeks? | List the cash coming in and the cash going out for each of the next thirteen weeks. Keep last week's version so you can see where your estimate was wrong. | When the lowest week falls under four weeks of operating burn, slow inventory buying immediately. Inventory is the largest outflow you can cut on short notice. | | Runway How many months can I operate if nothing changes? | Take your unrestricted cash and divide it by average monthly net burn, meaning the cash you consume in a month after collections come in. | Under six months, start the funding conversation that week. A new facility takes roughly that long to close, so below six months you are negotiating from weakness. | | Collections coverage Do my collections cover what I owe? | Divide the collections you expect over the next ninety days by the fixed obligations due in the same window: debt service, rent, payroll, and minimum inventory spend. | Under 1.1x, cut discretionary inventory purchases. Under 1.0x you are paying operating costs out of the balance sheet, and a balance sheet has a floor. | | Facility headroom How close am I to over-advancing? | Compare your drawn balance against your borrowing base every week. The gap between them is your headroom. | Under 10%, raise down payments and pull ineligible paper out of the base yourself. Doing it before the lender's audit does it is the difference between a conversation and a demand letter. | | Covenant distance How much room is left before I breach? | For each covenant in your agreement, work out how far today's actual number sits from the trigger, as a percentage of the trigger. | Under a 15% cushion, call the lender before the breach rather than after. A waiver arranged early costs fees. One arranged afterward costs control. | | Funding concentration Can a single lender end me? | Take the share of your total funding that sits in the largest facility, then write down the maturity date of every facility you have. | Above 60% in one facility, or with two maturities landing in the same quarter, open a second lender relationship while you still look healthy enough to be worth one. | One number tripping tells you which lever to pull. Two tripping in the same week is a different situation, and the point to bring in an advisor while you still have choices rather than after a lender picks one for you. ## Limits **The operators most at risk are the least visible.** No public dataset covers single-lot BHPH cash positions, runway, or facility headroom, and no reliable closure series exists for 2023-2026. The best-documented liquidity deaths are a public company and a Chapter 11 debtor. That absence is consistent with the finding but cannot be independently quantified. **Benchmark data skews strong:** SGC's composite explicitly covers "some of the best performing operators," so the average operator's cash economics are worse than every figure above. **The Cash Clock figures are a single-month cross-section:** twelve portfolios in June 2026, reported as observed rather than as an estimate of the wider sector. Months to breakout is a reported field in that data, defined there as average cash in the deal divided by average monthly payment, so it describes deal structure rather than collections performance. Records showing 0% or 100% delinquency, and three series repeating one value for months, were excluded from any testing. **Tricolor figures are allegations** from an indictment and civil suits; two former executives pleaded guilty in December 2025 and the former COO in June 2026, while charges against the founder remain unproven in court as of this writing. **Car-Mart's distress language describes risk, not an outcome:** as of August 15, 2026 it had not filed for bankruptcy and was operating under waivers through September 7, 2026. **Industry counts** (roughly 30,000 licensed dealers, roughly $20 billion in annual originations) are industry estimates cited in the PrimaLend First Day Declaration, not Federal Reserve figures. **Sources & notes** Benchmarks: [SGC CPAs, 2024 Buy Here-Pay Here Financial Benchmarks](https://www.sgcaccounting.com/Resources/BHPHBenchmarks2024.pdf) (Bad Debts/Vehicle Sales 21% in 2022, 24% in 2023, 28% in 2024 per its ratio tables; partner Steven Carstens: "the greatest threat facing the industry right now is the number of lenders leaving the space" and "reducing debt levels and running lean"; composite footnoted as drawn from "the best performing operators in the industry"; figures verified against the published PDF). Sector structure: [Federal Reserve FEDS Note, May 8, 2026](https://www.federalreserve.gov/econres/notes/feds-notes/subprime-auto-lending-trends-in-buy-here-pay-here-auto-lending-20260508.html) (BHPH subprime cells of Table 1: $15,402 average origination balance, 55-month average term, 25.39% weighted-average rate; roughly $32 billion outstanding in 2025, about 2% of the $1.6 trillion auto market; 81% guarantor-backed, 65% asset-based, ~13% SPEs, 60-80% advance rates on Tricolor; bank risk metrics "have all worsened"; default probability up nearly 150%, 2025:Q2 to Q3). Failures: Bloomberg on American Car Center's pulled $222M bond sale and next-day closure (Feb 2023; Chapter 7 March 14, 2023); Bloomberg and Auto Finance News on U.S. Auto Sales (dealerships closed April 2023; Chapter 7 August 2023); SDNY indictment and DOJ releases on Tricolor (~$2.2B pledged vs ~$1.4B real; guilty pleas Dec 16, 2025 and June 24, 2026); PrimaLend First Day Declaration, Bankr. N.D. Tex., Oct 22, 2025 (over-advances, ~$34M participation sale, missed July 15, 2025 interest payment, facility maturing Oct 23, 2025, ~$233.8M dealer loans, ~$286.1M funded debt); Auto Remarketing and trade press on Automotive Credit Corp (Aug 7, 2025) and FinBe USA; America's Car-Mart FY2026 Form 10-K (filed July 14, 2026; $722.4M total debt, comprising $458.7M non-recourse securitization notes plus the Silver Point term loan at its $263.7M net carrying value; the $300M term-loan figure at SOFR+7.50% is face value, which is why the components do not sum to the total; no revolver, going-concern language, 60 of 154 dealerships closed, net loss $139.1M) and the June 19, 2026 First Amendment and Limited Waiver 8-K (weekly minimum liquidity of $7M, $5M at other times; Collateral Coverage Ratio of 1.25:1.00 stepping to 1.20:1.00; fees up to $18M; scheduled termination September 7, 2026, extendable to November 6). Bank losses on Tricolor: JPMorgan $170M charge-off and the Dimon "cockroach" remark (Oct 14, 2025 call); Fifth Third $178M non-cash impairment (its disclosed figure; the Fed note characterizes major exposures as around $200M each); Barclays £110M/~$148M impairment; Origin Bancorp $29.5M net charge-offs per its FY2025 10-K. Zions Bancorporation's contemporaneous $50M charge-off related to the separate Cantor Group matter and is deliberately excluded. Market: [Cox Automotive 2026 outlook, Jan 6, 2026](https://www.coxautoinc.com/insights/cox-automotive-2026-outlook/) (15.8M units, down 2.4% in the January outlook, revised to down 2.9% in June with volume held; "Our 2026 forecast reflects a slowing market, but still a good one"); Fitch Ratings subprime 60+ delinquency: record 6.90% in January 2026, 6.11% at the March tax-refund trough, 5.67% in June, down 64 basis points year over year (via Auto Remarketing, May 21, 2026, and Wolf Street, August 13, 2026; trade sources differ on the January 2025 baseline, 6.45% per Fitch coverage against 6.56% per Wolf Street). The Cash Clock section is this publication's own computation on monthly dealer benchmark reporting, a twelve-portfolio, eight-state cross-section for June 2026. Months to breakout is a reported field in that data, defined there as average cash in the deal divided by average monthly payment. Names, firms and any identifying detail are withheld, and figures are reported at the portfolio level only. All forward-looking and interpretive statements are labeled as inference. LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned. Not investment, legal, or accounting advice. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html) LR LendRisk Analytics Independent market research Continue reading [Issue · 05 · Comparative postmortem Three failures, *one blind spot.*](https://lendriskanalytics.com/insights/three-failures-one-blind-spot.html) [Issue · 06 · Cycle study Three stress cycles, *one missing layer.*](https://lendriskanalytics.com/insights/three-cycles-one-missing-layer.html) --- title: "What the tape said: current, on paper" url: https://lendriskanalytics.com/insights/current-on-paper.html publisher: LendRisk Analytics series: What the Tape Said issue: 8 published: 2026-08-09 kind: Method brief description: "An account gets an extension. The past-due clock resets. The tape shows current. Nothing about the borrower has changed. Where that gap lives in the SEC record, what the Philadelphia Fed found, and the disclosure failure that cost America's Car-Mart a non-reliance finding. Every figure sourced; inferences labeled." html: https://lendriskanalytics.com/insights/current-on-paper.html --- # What the tape said: current, on paper [← All insights](https://lendriskanalytics.com/articles.html) What the Tape Said · Issue 8 Method brief · 16 min read Extensions · Deferrals · Reported vs actual # Current, on paper. Public record through August 9, 2026 · LendRisk Analytics An account gets an extension. The past-due clock resets. The tape shows current. Nothing about the borrower has changed. In April 2026 the Federal Reserve Bank of Philadelphia put a number on what that does to the headline: record auto delinquency is being driven by delinquent loans failing to resolve rather than by new borrowers falling behind, and the headline rate "taken at face value, likely overstates the degree to which auto borrowers' financial health is currently deteriorating." This brief traces where that gap lives in the SEC record, and where the record goes dark. 28.9% Of Car-Mart's gross receivables under a modification program it failed to disclose ~3.5% Share of subprime loans receiving an extension last year, per the Philadelphia Fed 23.44% COVID peak subprime extension share, May 2020, dollar basis (S&P) Zero Regulators aggregating the Schedule AL extension fields into a public series ## Bottom line Extensions are a legitimate servicing tool with a measurement side effect. When a servicer grants one, the amount due next period goes to zero, and a loan that owes nothing is reported current. The borrower's capacity has not changed. The tape has. Three things are now established. The Philadelphia Fed has decomposed the delinquency stock and found the inflow stable while the stock climbs. The fields that would let anyone measure extension usage exist in every registered auto ABS filing and no regulator aggregates them. And America's Car-Mart demonstrated what happens when the disclosure obligation gets missed: a non-reliance finding across two fiscal years, a material weakness, a Nasdaq notice, and securities litigation, over a footnote covering 28.9% of its receivables. What follows separates what the filings actually disclose from what the market infers, and flags one widely circulated category of evidence that does not survive checking. The read The suppression here is mechanical rather than deceptive. An issuer following the rules exactly still produces a delinquency number that understates portfolio condition, because the definition of current is "nothing is due," and an extension makes nothing due. **Santander states the mechanism in its own filing: if the next payment due is reported as 0.00, no payment is required "for the receivable to be considered current" because the obligor "was granted a payment extension."** Nobody has to lie for the tape to mislead. ## I · What the filings actually disclose Form ABS-EE carries the loan-level record for registered auto ABS, adopted under Regulation AB II and effective November 23, 2016. It carries two exhibits: EX-102, the asset data file containing the Schedule AL payload, and EX-103, the asset-related document narrative. The field specifications sit in 17 CFR 229.1125, which requires asset-level disclosure each distribution period covering origination characteristics, scheduled and actual payments, delinquency, modifications, prepayments, charge-offs and performance. The modification and extension data points travel in the XML payload under element names such as modificationTypeCode and paymentExtendedNumber . Those names come from the EDGAR schema and third-party documentation of the ABS-EE dataset rather than from the text of the regulation, which describes the required data points in prose. Schedule AL also carries a general modification indicator recording whether an asset was modified during the reporting period. **Inference**Comparability across issuers is not guaranteed. Issuers may omit fields they consider inapplicable and may populate modification codes on their own conventions, so a cross-issuer extension rate assembled from raw filings carries an unquantified consistency risk. Anyone building this series should validate field population issuer by issuer before comparing shelves. ### The mechanism, in an issuer's own words Santander's EX-103 narrative for Drive Auto Receivables Trust 2024-1 spells it out. Item 3(c)(5), loan maturity date, is "the current final maturity date of the receivable after giving effect to payment and promotional extensions and due date changes." Item 3(f)(6), next reporting period payment amount due, states that where the figure is reported as 0.00, "no interest or principal is due in the next reporting period for the receivable to be considered current because the obligor either made a payment in advance or was granted a payment extension." How an extension resets the clock The borrower's position is unchanged on both sides of the line. The reported status is not. Schematic, not measured data. Mechanism and quoted definitions per the Santander Drive Auto Receivables Trust 2024-1 EX-103 narrative, Items 3(c)(5) and 3(f)(6). Dollar figure illustrative. ### Who discloses their extension policy Consumer Portfolio Services states its limits plainly in its FY2024 10-K: "In certain circumstances we will grant obligors one-month payment extensions to assist them with temporary cash flow problems. In general, an obligor will not be permitted more than two such extensions in any 12-month period and no more than eight over the life of the contract." CPS also notes it counts delinquency "as extended where applicable," which is the reset stated as an accounting convention. **Inference**A disclosed cap of two per year and eight per contract life is the difference between a short-term hardship tool and open-ended ever-greening. The presence or absence of a stated cap is itself a diligence question, and most issuers do not state one. ## II · The suppression effect, measured The Federal Reserve Bank of Philadelphia's April 2026 report, "Do Recent Auto Loan Delinquency Rates Overstate Borrower Distress?" by Cheney, Hunt, Lambie-Hanson, Santucci and Zhou, is the first official decomposition of the question. Its structural finding is that the pool of delinquent loans has three components: borrowers newly delinquent, borrowers delinquent across multiple quarters, and redefaulters who returned to good standing and fell behind again. The stock of severe delinquencies is rising while the flow of new delinquencies stays fairly stable. Loans that once resolved quickly are staying on the books longer, which holds the headline elevated even as the pace of new distress moderates. On extensions the report is direct about prevalence and careful about causation. It states that lenders have expanded their use of loss-mitigation tools such as loan extensions, which let borrowers defer payments to the end of the term, with the share of subprime loans receiving an extension reaching approximately 3.5% last year. On the link to redefaults it says expanded extensions *may* explain the rise in redefaulters, because extensions that temporarily return borrowers to current status are followed by many of those borrowers falling behind again. The conclusion is the sentence quoted at the top of this brief. Stock rising, inflow flat The shape the Philadelphia Fed found underneath the headline. Schematic and qualitative, not measured data. Structure per the Federal Reserve Bank of Philadelphia, "Do Recent Auto Loan Delinquency Rates Overstate Borrower Distress?" April 2026. The report presents expanded extension use as a possible explanation for the redefaulter increase rather than a demonstrated cause. ### What the COVID episode showed, on a consistent basis The pandemic produced the only extension surge large enough to see clearly, and it is worth reporting on one measurement basis at a time. On a dollar basis for public subprime shelves, S&P recorded extension status peaking at 23.44% in May 2020 and falling roughly 30% to 16.29% in June. At month-end June 2020 the individual shelves ran: Santander's DRIVE at 22.58%, SDART at 18.38%, World Omni Select at 8.49% and AmeriCredit at 6.69%. S&P also published a count-basis series for the four public subprime platforms filing Reg AB II loan-level data, which ran 6.82% in March 2020, 15.75% in April and 8.9% in May. Those two series measure different things and should never be spliced into one line. **Inference**The COVID cohort largely cured, and that is the strongest argument extensions work. It is also the weakest available analogue for today, because the cure was powered by stimulus payments and enhanced unemployment insurance. An extension is a bridge, and a bridge needs something on the far side. The COVID spike, and where the rate sits now Subprime extension share on a dollar basis, public shelves, against the pre-pandemic floor. 2019, 2020 and 2024 figures per S&P Global Ratings via Asset Securitization Report, dollar basis for public subprime shelves. The 2025 figure is the Philadelphia Fed's approximately 3.5% share of subprime loans receiving an extension, sourced to Intex Solutions, and is an annual flow rather than a point-in-time share; it is placed here for scale and is not strictly comparable to the S&P bars. ## III · The disclosure failure ASU 2022-02 eliminated troubled-debt-restructuring accounting for entities that had adopted CECL and replaced it with disclosure requirements for loan modifications to borrowers experiencing financial difficulty. Reportable types are principal forgiveness, interest rate reduction, significant payment delay, or term extension, and the requirements sit in ASC 310-10-50-42 through 50-44. America's Car-Mart failed that standard, and the sequence is worth reading as a timeline rather than a headline. | Date | Event | |---|---| | Jul 15, 2025 | Form 12b-25 filed. The FY2025 10-K will be late because management "identified the need to enhance disclosures related to loan modifications for borrowers experiencing financial difficulty." | | Jul 30, 2025 | 8-K discloses that the FY2024 10-K and the FY2024 and FY2025 10-Qs "should no longer be relied upon," for omitting disclosures required by ASC 310-10-50-42 through 50-44. | | Aug 1, 2025 | Nasdaq notice of non-compliance with Listing Rule 5250(c)(1). | | Aug 8, 2025 | Comprehensive FY2025 10-K filed, disclosing a systematic modification program covering $436.1 million, or 28.9%, of gross finance receivables as of April 30, 2025. | The omission carried no change to the balance sheet, income statement or cash flow statement as previously reported. It carried a material weakness in internal control, attributed to an incorrect assessment during initial adoption of ASU 2022-02, ineffective disclosure controls, and turnover in technical accounting resources. Securities litigation followed, with a proposed class period running from July 2023 to September 4, 2025. What matters for the rest of the market is narrower than the litigation: a company running term extensions across more than a quarter of its book did not consider that a reportable modification program, and its auditors and controls did not catch it for two fiscal years. **Inference**Read against what came next, the footnote was a leading indicator. By the FY2026 10-K filed July 14, 2026, Car-Mart reported a $139.1 million net loss, going-concern language, no revolver or warehouse facility, and a footprint cut from 154 dealerships to 94. The [Issue 6](https://lendriskanalytics.com/insights/three-cycles-one-missing-layer.html) and [Issue 7](https://lendriskanalytics.com/insights/the-future-of-subprime.html) accounts pick up that thread. A modification program covering 28.9% of receivables was information about portfolio condition, and it was disclosed a year late. The footnote, and what followed America's Car-Mart from disclosure failure to going concern. Dates per America's Car-Mart SEC filings: Form 12b-25 (July 15, 2025), 8-K (July 30, 2025), Nasdaq notice (August 1, 2025), FY2025 Form 10-K (August 8, 2025), term loan 8-K (October 30, 2025), covenant amendment 8-K (June 22, 2026), FY2026 Form 10-K (July 14, 2026). Spacing is schematic. ### What the CFPB has and has not said The CFPB's auto-finance Supervisory Highlights, in the Fall 2024 special edition and Spring 2022, flagged extension-adjacent conduct: servicers wrongfully repossessing vehicles after borrowers had obtained extensions, deferments or modifications, and deferral notices carrying imprecise conditional statements that misled consumers about final payment amounts. Those are servicing-conduct findings. No regulator has published a finding on extension measurement or on the effect of extensions on reported delinquency. ## IV · The dealer visibility gap Credit Acceptance is the sharpest test case, because dealer economics depend on ongoing pool performance. Under the Portfolio Program, collections on a dealer's pool run a waterfall: first to collection costs, second to the servicing fee, third to reduce the advance balance, and fourth to the dealer as holdback. If collections do not repay the advance balance and other amounts due, the dealer receives no holdback. An extension slows principal paydown and delays the point at which cumulative collections clear the advance balance, which delays or reduces holdback automatically through the waterfall. Nothing obliges the lender to itemize extensions to the dealer, and CACC's Purchase Program Agreement affirmatively disclaims dealer notice rights, stating the dealer "is not entitled to receive any statutory notices concerning Credit Acceptance's collection of a Contract." Across the broader indirect market, dealer portals handle origination, funding and payoffs. Extension and deferral notices go to the consumer. Recourse and repurchase in the mainstream indirect model trigger on origination warranty breaches rather than on servicing events, and the CFPB's 2019 examination procedures confirm servicing transfers wholly to the assignee. **Inference, and an absence-of-evidence finding**An undocumented internal servicing screen inside some lender's dealer portal cannot be ruled out, because those specifications are not public. What can be said is that no public contract, filing or portal documentation reviewed here gives a dealer account-level sight of extension activity. The structural result is that a dealer carrying recourse or holdback exposure has no visibility into a servicing practice that is quietly changing the timing of that pool's realized losses. Who can see an extension The account-level view stops at the servicer. Schematic and qualitative. Synthesized from the Credit Acceptance FY2024 10-K waterfall description and Purchase Program Agreement, CFPB Auto Finance Examination Procedures (2019), and public dealer-portal documentation. Visibility varies by contract; this is an absence-of-evidence finding rather than a measured dataset. ## V · The honest counter-case Here is the strongest good-faith argument that extensions are legitimate loss mitigation, stated at full strength. The COVID cohort largely cured. Extension shares fell from a 23.44% May 2020 peak toward normal within months, and the 2020 and 2021 auto ABS vintages went on to deliver strong recoveries and low losses. A one-month extension for a genuine temporary cash-flow gap avoids an unnecessary repossession and preserves value for the borrower and the investor alike. CPS's disclosed design supports the benign reading. Two extensions per twelve months and eight over a contract life, granted one month at a time, is a structure built for short-term hardship. And the Philadelphia Fed does not characterize extensions as abusive. It frames them as an evolution in loss-mitigation practice, and its headline conclusion is that reported delinquency overstates deterioration. Read plainly, that is a point in favor of the borrowers: the population may be in better shape than the headline suggests, with the measurement capturing longer resolution timelines. **Inference**The counter-case is strongest exactly where it is least needed. A temporary macro shock with a government backstop is the ideal case for forbearance, and 2020 had both. Today's stress is structural affordability against record negative equity, with 29.6% of Q2 2026 trade-ins underwater by an average of $6,884. The mortgage-modification literature is the relevant analogue, and its finding is that redefault rates fall as payment reduction rises. An auto extension defers without reducing anything, which places it at the weak end of that spectrum. ## Limits: what the record does not show **A widely circulated category of loss evidence does not survive checking.** Third-party analytics comparing loss severity on extended against never-extended charge-offs circulates in this market and is the sort of finding this brief would most want to cite. The two studies most often quoted carry publication dates of August 26 and September 2, 2026, both of which fall after this brief's cutoff of August 9, 2026, on a vendor site whose research list also shows entries dated August 12 and August 19, 2026. Figures whose own publication dates have not yet occurred are not used here. Anyone citing them should confirm the dating directly with the vendor before relying on them. **Extension-rate measures are not interchangeable, and mixing them produces nonsense.** A point-in-time share of loans currently in extension, an annual share of loans that received an extension, a dollar-weighted share and a count-weighted share are four different numbers. The Philadelphia Fed's approximately 3.5% is an annual flow. S&P's shelf figures are dollar-weighted point-in-time. Vendor trackers publishing sub-1% active-extension rates are point-in-time count shares. A shelf compared against the wrong benchmark will look four times better or worse than it is. **No standardized cross-issuer extension rate exists from any official source.** The Schedule AL data points exist in every registered auto ABS filing and no regulator aggregates them into a public index. Rating agencies compute them and publish behind paywalls; third-party parsers compute them with undisclosed methodologies. **Field-level consistency was not verified.** Whether every named shelf populates the modification and extension elements identically is unconfirmed, and the XML element names cited here come from the EDGAR schema and dataset documentation rather than the text of 17 CFR 229.1125. **The Philadelphia Fed report's causal claim is hedged and is reported as hedged.** It states extension expansion and the approximately 3.5% figure directly. It offers expanded extension use as a possible explanation for the redefaulter increase rather than a demonstrated cause. This brief does not upgrade that. **Rating-agency figures reach this brief through trade press.** S&P and Fitch extension and delinquency series are cited via Asset Securitization Report and Auto Remarketing where the underlying reports are paywalled. **The Car-Mart securities docket is unconfirmed.** The law-firm investigations and the proposed July 2023 to September 4, 2025 class period are on the record. The consolidated case name, court and lead-plaintiff outcome are not. **No auto-specific peer-reviewed redefault study was located** matching the rigor of the mortgage-modification literature. The mortgage findings are used as an analogue and labeled as one. ## What to do with this 1 **Build the extension-adjusted delinquency series, and fix the denominator first.** Pull the modification and extension elements from EX-102 for the shelves you hold and compute a measure that treats extended-to-current accounts as still past due. Before comparing anything, decide whether you are measuring a point-in-time share or an annual flow, and dollar-weighted or count-weighted, then benchmark only against figures on the same basis. The Philadelphia Fed validated the underlying thesis; the edge is doing this per trust, monthly, on a consistent basis. 2 **The divergence matters more than the level.** An issuer whose extension rate is climbing while its reported 30+ and 60+ delinquency improves is the pattern worth flagging. The level alone tells you little, because a disciplined program with disclosed caps and an undisciplined one can print the same number in a given month. 3 **If you carry recourse or holdback exposure, price the blindness.** No lender documents account-level extension reporting to dealers. Negotiate an extension-activity feed, or at minimum a monthly aggregate extension rate on your own pool, at agreement renewal. Where that fails, model holdback recovery with an explicit drag for extension timing, because the waterfall will apply that drag whether or not you modeled it. 4 **Treat a second ASU 2022-02 non-reliance finding as the sector signal.** Car-Mart is one company's controls failure until it happens twice. A second issuer restating or amending for omitted modification disclosures, or SEC comment letters to other integrated lenders on ASC 310-10-50 completeness, would convert this from an idiosyncratic story into a disclosure-standard shift worth repositioning around. **Sources & notes** **Filings and field structure.** 17 CFR 229.1125, Item 1125, Schedule AL asset-level information, via eCFR, which requires per-period asset-level disclosure covering modifications among other data points. Form ABS-EE adopted under Regulation AB II, effective November 23, 2016. XML element names for modification and extension data points per the EDGAR schema and third-party ABS-EE dataset documentation rather than the regulation text. Santander Drive Auto Receivables Trust 2024-1 EX-103 narrative, SEC accession 0000950131-24-000934, for the Item 3(c)(5) and Item 3(f)(6) definitions quoted. Consumer Portfolio Services FY2024 Form 10-K, SEC accession 0001683168-25-001548, for the extension caps and the "as extended where applicable" delinquency convention. **The suppression finding.** Federal Reserve Bank of Philadelphia, "Do Recent Auto Loan Delinquency Rates Overstate Borrower Distress?" April 2026, by Julia Cheney, Bob Hunt, Lauren Lambie-Hanson, Larry Santucci and Justin Zhou, for the stock-versus-flow decomposition, the three-component delinquent pool, the approximately 3.5% subprime extension share sourced to Intex Solutions, the hedged link between expanded extensions and redefaulters, and the conclusion that the headline rate "taken at face value, likely overstates the degree to which auto borrowers' financial health is currently deteriorating." The American Financial Services Association published an industry response, "Stop Misreading Auto Data," on April 16, 2026. **Extension rates.** S&P Global Ratings via Asset Securitization Report for the dollar-basis public subprime shelf series: 2.06% in September 2019, 23.44% in May 2020, 16.29% in June 2020, and 3.19% in September 2024 against 2.48% in September 2023. Month-end June 2020 shelf detail: DRIVE 22.58%, SDART 18.38%, World Omni Select 8.49%, AmeriCredit 6.69%. S&P's separate count-basis series for the four public subprime Reg AB II platforms ran 6.82% in March 2020, 15.75% in April and 8.9% in May, and is reported separately because it is not comparable to the dollar-basis series. Fitch subprime 60+ index at a record 6.90% for the January 2026 reading; trade sources differ on the year-earlier baseline, with Wolf Street reporting 6.56% and Auto Remarketing 6.45%, and this brief uses 6.56% consistent with the rest of the series. **Accounting and the Car-Mart sequence.** ASU 2022-02, Financial Instruments, Credit Losses (Topic 326), and ASC 310-10-50-42 through 50-44. America's Car-Mart Form 12b-25 (July 15, 2025), Form 8-K and press release (July 30, 2025), Nasdaq non-compliance notice under Listing Rule 5250(c)(1) (August 1, 2025), and FY2025 Form 10-K, SEC accession 0001628280-25-039026 (August 8, 2025), for the $436.1 million and 28.9% figures, the no-impact statement on the primary financial statements, and the material weakness and its three attributed causes. FY2026 Form 10-K (July 14, 2026) for the subsequent going-concern position. Securities investigation notices from Levi & Korsinsky, Glancy Prongay, Howard G. Smith, Hagens Berman and the Law Offices of Frank R. Cruz for the proposed class period. **Servicing conduct and dealer visibility.** CFPB Supervisory Highlights, Fall 2024 auto special edition and Spring 2022, for wrongful repossession after approved extensions and for imprecise deferral notices. Credit Acceptance FY2024 Form 10-K for the Portfolio Program collection waterfall and dealer holdback, and the Purchase Program Agreement exhibit, SEC accession 0000950137-07-006346, for the disclaimer of dealer notice rights. CFPB Auto Finance Examination Procedures (2019) on transfer of servicing to the assignee. Federal Reserve Bank of Minneapolis, "Indirect Lending" (2014), and Boardman Clark LLP indirect-financing commentary on recourse triggers. **Counter-case and context.** Congressional Research Service R46356 on forbearance as a loss-mitigation tool for temporary hardship. Federal Reserve Bank of Philadelphia Working Paper 18-02 on mortgage-modification redefault and the relationship between payment reduction and redefault, used as an analogue and labeled as one. Edmunds Q2 2026 negative-equity data, released July 16, 2026: 29.6% of trade-ins underwater against 26.6% a year earlier, average $6,884, and a $944 average monthly payment for buyers rolling negative equity against a $777 industry average. **Companion issues.** [Issue 4](https://lendriskanalytics.com/insights/cacc-stress-signals.html) is the Credit Acceptance stress-signal brief. [Issue 5](https://lendriskanalytics.com/insights/three-failures-one-blind-spot.html) is the comparative postmortem on Tricolor, PrimaLend and America's Car-Mart. [Issue 6](https://lendriskanalytics.com/insights/three-cycles-one-missing-layer.html) is the three-cycle study. [Issue 7](https://lendriskanalytics.com/insights/the-future-of-subprime.html) covers the cross-facility measurement gap this brief's dealer-visibility section extends. Where this brief reasons beyond what a document literally states, it is labeled as an inference. All five charts are schematic or qualitative and are captioned as such. Figures that could not be tied to a named source, or whose publication dates fall after this brief's cutoff, are excluded and flagged in Limits rather than softened. Point-in-time reading of the public record through August 9, 2026. LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned, and this is not investment, legal, or accounting advice. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). LR LendRisk Analytics Independent market research Continue reading [Issue · 07 · Deep study The future of *subprime.*](https://lendriskanalytics.com/insights/the-future-of-subprime.html) [Issue · 06 · Deep study Three stress cycles, *one missing layer.*](https://lendriskanalytics.com/insights/three-cycles-one-missing-layer.html) --- title: "What the tape said: the future of subprime" url: https://lendriskanalytics.com/insights/the-future-of-subprime.html publisher: LendRisk Analytics series: What the Tape Said issue: 7 published: 2026-08-08 kind: Deep study description: "A forward read on subprime auto. Severity has moved into origination and become forecastable, the industry benchmark is dissolving under composition drift, and the verification layer will be built by a rating agency, a consortium or a vendor. What each outcome costs the operators being measured. Every figure sourced; inferences labeled." html: https://lendriskanalytics.com/insights/the-future-of-subprime.html --- # What the tape said: the future of subprime [← All insights](https://lendriskanalytics.com/articles.html) What the Tape Said · Issue 7 Deep study · 20 min read A forward read · Four calls, one falsifiable # The future of subprime. Public record through August 8, 2026 · Sector data confirmable through March 2026 · LendRisk Analytics Six issues of reading the tape after the fact have left me with a view I want on the record before the data catches up to it. Subprime auto is about to start pricing operators on whether they can prove what sits in their own book, and four things get us there. Severity has moved somewhere it can be forecast. The industry benchmark is dissolving under its own composition. The verification layer nobody built in 1998 is finally going to get built by somebody, on terms that will matter a great deal to whoever is being measured. And the next failure lands in the same layer as the last three. What follows is the evidence for each, the parts I cannot confirm, and one prediction specific enough that you could catch me being wrong about it. 37.74% Full-year 2025 subprime recovery average, lowest in S&P data back to 2007 6.18% Full-year 2025 subprime 60+ DQ vs 5.78% in 2024 (S&P). Annual, no seasonal distortion $2.2B vs $1.4B Tricolor collateral pledged against collateral that existed, per the SDNY indictment 1998 Year the trade body said it was building standardized static-pool reporting **Data currency**Fitch's monthly subprime index is confirmable at 6.90% (January 2026, a record), 6.80% (February), and 6.11% (March, the tax-refund seasonal trough, with annualized net losses of 8.80% and recoveries of 37.48%). April through July 2026 monthly prints are not confirmable from a named source and are treated as unobserved. S&P's full-year 2025 figures are confirmed and carry the argument wherever a current-condition claim is made. The July 2026 employment report and Credit Acceptance's Q2, both released in the first week of August, are included. America's Car-Mart is confirmed through its June 19, 2026 covenant amendment and its FY2026 10-K filed July 14, 2026; any resolution after that date is unobserved. ## Bottom line The claim is one sentence. Operators who can prove what sits in their own book are going to price differently from operators who can only assert it, and that gap is about to become explicit rather than implied. Four things get us there. Each is a call, and each is labeled as one. **Severity has moved into origination, which makes it forecastable for the first time.** Recoveries hit a 2007 low in a year when wholesale prices rose. That combination means loss severity is now set by amount financed, term and rolled negative equity rather than by the auction lane, and every one of those is knowable on the day the loan is written. Anyone still reading Manheim as the severity signal is watching a gauge that has been disconnected from the engine. **The industry benchmark is dissolving.** Deep subprime is now roughly a third of S&P's subprime composite, and Fitch attributes part of its own record to composition effects. An index whose definition drifts cannot serve as a benchmark, so proving outperformance is about to require like-for-like static pools that most operators do not produce. **The verification layer will be built, and the only open question is who owns it.** The Structured Finance Association has a task force, vendors are selling into the gap, and rating-agency criteria have not moved yet. Those three routes produce three different market structures and three different answers to who pays. **The next failure surfaces in the same place as the last three.** Term ABS absorbed the whole 2025 stress without structural failure. The pre-securitization stack did not. That asymmetry has not been fixed, so the next event is a warehouse withdrawal rather than a bond downgrade. The read The industry keeps reading 2025 as a wave of failures and drawing the wrong lesson from it. One fraud that nobody could see, plus two distressed operators the market saw perfectly well and priced in advance. **If detection failed once rather than three times, then better underwriting was never the missing piece. What was missing is the ability to prove, to someone outside your walls, that the collateral you say you hold is real, singular, and paying the way you say it is.** Everything below is what follows from taking that seriously. ## I · What 2025 settled The numbers are on the record and [Issue 6](https://lendriskanalytics.com/insights/three-cycles-one-missing-layer.html) works through them. Full-year 2025 subprime 60-day delinquency at 6.18% against 5.78%, a record. Recoveries at 37.74%, the lowest annual figure in S&P's data back to 2007. Fitch's monthly index at a record 6.90% in January 2026, easing to 6.11% by March on the tax-refund seasonal. Term ABS absorbed all of it: a second consecutive record issuance year at $127 billion, with spreads widening at the BBB and BB subprime end around the Tricolor headlines and senior paper holding throughout. The three failures are covered case by case in [Issue 5](https://lendriskanalytics.com/insights/three-failures-one-blind-spot.html). What matters here is the shape they made together, because the whole forward argument rests on it. | Case | Warning available | Who saw it | Outcome | |---|---|---|---| | PrimaLend | 14 months | CIBC, through borrowing-base over-advance | Chapter 11, liquidated | | Car-Mart | Priced in advance | Silver Point, from audited public filings | Default, unresolved | | Tricolor | Effectively none | One analyst, at a mezzanine lender | Chapter 7, criminal case | Two of the three were seen clearly and acted on. The one that was not involved information no counterparty could independently reach. Detection did not fail three times. It failed once. Time from the first signal a counterparty could act on to the terminal event. Schematic, not measured data. Dates from the DOJ SDNY indictment, PrimaLend's First Day Declaration and case coverage, and America's Car-Mart SEC filings. Warning-date placement reflects the public record; private creditor knowledge predates public disclosure by an unknown interval. ## II · Severity moved into origination, so it can be forecast Here is the finding with the longest forward reach, and it is the one getting the least attention. Recoveries fell to their worst annual level since 2007 in a year when Manheim rose 2.1%. Those two facts together rule out the auction lane as the cause. What is left is the amount financed: larger balances, longer terms, and negative equity rolled forward from the last car onto the next one. Loss severity used to be discovered at remarketing, months after default, and it was genuinely hard to predict because it depended on where used-car prices happened to be. That is no longer where it is being set. **Inference**If severity is determined by loan structure rather than by auction outcomes, then severity became computable at origination. Advance rate against actual collateral value, term, and rolled negative equity are all known on the day the contract is written. An operator can price expected loss-given-default per deal at the point of sale, which was never reliably possible when the answer depended on the auction eighteen months later. The same shift makes the wholesale index a broken proxy: watching Manheim to gauge severity now measures a variable that has been disconnected from the outcome. Recoveries at a 2007 low while wholesale prices hold In 2008 and 2009, recoveries fell because auction prices fell. In 2025 they fell to a comparable level while Manheim rose. Manheim anchors per Cox Automotive; path smoothed between anchors. Recovery figure and the 2007 to 2024 comparison per S&P Global Ratings via Auto Remarketing; the shaded zone represents the licensed claim that 2025 is the series low, not per-year values, which sit in paywalled S&P data. Both series are pool-level composites subject to composition drift. Loss is built at origination now Negative equity carried into the next loan, by quarter. Edmunds negative-equity data, Q2 2025 through Q2 2026. Figures reflect new-vehicle trade-ins; subprime and BHPH used-vehicle terms differ but move with the same affordability pressure. Quarter-to-quarter comparisons are affected by seasonality, which is why the record claim is stated against prior second quarters only. ## III · The benchmark is dissolving The second forward consequence is that the number everyone quotes is losing its ability to mean anything. Fitch attributes part of its own record to a trailing average "driven by collateral deterioration and composition effects as stronger pre-pandemic vintages amortized and were replaced by weaker post-pandemic cohorts," with weakness "most acute in the 2022-2023 vintages." S&P notes deep subprime now makes up roughly a third of its subprime composite. Both indices measure whoever happens to be issuing, in whatever mix they happen to be issuing. **Inference**An index whose composition drifts cannot function as a benchmark for very long. If a lender's book improves while the composite worsens because deep subprime issuers took a larger share of the composite, the lender has no way to demonstrate that from public data, and no counterparty has a way to check it. The consequence lands on funding: as the composite becomes less usable, the burden of proof shifts onto the individual operator, and the only instrument that carries that proof is a like-for-like static pool by vintage. Operators who already produce those will be able to show a lender exactly where they sit against their own history. Operators who cannot will be priced against a composite that no longer describes them. ## IV · Who builds the verification layer The third consequence is the one with the most money attached, and the mechanism is already visible. The Structured Finance Association stood up a Fraud Mitigation Task Force and, with Ernst & Young, published survey findings on March 31, 2026 in which respondents "identified double-pledging of collateral as the most concerning fraud risk," with fictitious loans and originator misrepresentation also cited frequently. The SFA committed to "establish asset class working groups (Auto ABS, Consumer Loan ABS, RMBS and CRE/CMBS) to explore sector-specific nuances, map financial transactions, and develop recommendations for improved collateral tracking throughout the lifecycle." Respondents flagged legacy systems and inconsistent data quality as what slows progress. Meanwhile KBRA, S&P, Moody's and DBRS continue to apply existing auto-ABS criteria and participated in that survey as respondents rather than as authors of new de-duplication requirements. Three routes are open, and they are not equivalent for anyone being measured. | Route | What triggers it | What it costs the operator | |---|---|---| | Rating agency | A criteria change making collateral-uniqueness certification a condition of rating | Fastest and most binding. Compliance becomes the price of term-market access, and the cost falls on issuers immediately. | | Industry consortium | The SFA working groups producing an adopted standard | Slowest and cheapest, and the 1998 precedent argues it may not finish. Voluntary standards bind the disciplined and miss everyone else. | | Commercial vendor | A registry reaching enough coverage that lenders require it in credit agreements | Fast, and it puts a private party between you and your funding, pricing access to proof of your own collateral. | Assessment of routes visible as of August 2026, labeled as inference. No rating agency has announced a criteria change; the SFA working groups are formed but have published no standard; vendor coverage at scale is unverified. **Inference**The route matters more than the timing. A rating-agency criteria change makes verification a gate on funding, which favors operators who can already produce the evidence and prices everyone else out quickly. A vendor-owned registry creates a toll on proving your own book. A consortium standard is the cheapest outcome for operators and the least likely to arrive, on the evidence of the 1998 attempt at the same thing. In every version, the operator who already runs cross-facility reconciliation and static-pool reporting is on the right side of the change, because the layer verifies what that operator is already producing. Where each participant can see, and where nobody can Every row sees its own slice. One column is dark for everyone. Schematic and qualitative. Visibility varies by contract and is illustrative of structure rather than a measured dataset. Synthesized from the Federal Reserve FEDS Note (May 8, 2026), the SDNY indictment, and the SFA and Ernst & Young survey (March 31, 2026). That dark column is where Tricolor lived. Twenty-eight years between the diagnosis and the build Regulation eventually standardized the securitized layer. The layer where the 2025 failures happened stayed empty. Schematic, not measured data. Track A per the ABI Journal (May 1998) and SEC Regulation AB II (2014). Track B is a qualitative representation of the absence of an industry standard. Whether the 1998 NAFA effort was formally adopted is not confirmable from a named source; see Limits. ## V · The exposures that arrive next Three things follow that are not yet priced anywhere. ### The last channel that has not fired The July 2026 employment report, released August 7, showed payrolls falling 23,000 against a consensus looking for a gain of 83,000. Unemployment ticked down to 4.1%, and it did so because people left the labor force: participation fell to 61.4%, its lowest in more than five years, and it has declined 0.7 points since January. May and June were revised down a combined 103,000, bringing the twelve-month average job gain to 34,000. This is a labor market cooling from the bottom up, which is the cohort subprime lends to. **Inference, with a discipline flag**The temptation is to convert that into a delinquency forecast using a single elasticity. Resist it. The available parameters measure different quantities and are unstable across regimes. Fritsch and Prescott estimate a relative sensitivity of default probability, roughly 16% in 2006 against roughly 3% post-2020. The Richmond Fed's figure of roughly 0.54 is percentage points of default rate. Presenting those as a range would be a units error. What can be said directionally is that record delinquency and worst-since-2007 recoveries were both produced with this channel switched off, so the base a labor shock would land on is higher than in any prior cycle. The magnitude is not knowable from these parameters, and any threshold derived from them is uncalibrated. ### Where the next failure surfaces Term ABS absorbed the entire 2025 stress without structural failure while the pre-securitization stack produced all three casualties. Warehouse lines and private credit reprice and withdraw on confidence rather than on cash flow, and they still fund against borrower-reported collateral that no independent party continuously verifies. Nothing in the record since has changed that asymmetry. **Inference, and a falsifiable one**The next subprime auto event should present as a warehouse withdrawal or a facility non-renewal rather than as a bond downgrade, and it should become visible through a borrowing-base or covenant disclosure before it appears in any ABS performance series. If the next failure instead arrives through the term market, this reading is wrong and the structural argument in [Issue 6](https://lendriskanalytics.com/insights/three-cycles-one-missing-layer.html) needs revisiting. ### The measurement gap becomes a legal exposure Demand is moving toward thinner-file borrowers. Fitch noted that transactions with higher exposure to thin-file and undocumented immigrant borrowers experienced greater stress, reflecting income disruption and payment discontinuity following immigration-related borrower departures. Underwriting is moving with it, toward cash-flow and bank-transaction data that suits thin-file borrowers better than a thin-file score does. The affordability pressure driving that shift sits in the chart below, and it lands on one cohort. **Inference**Alternative-data underwriting in a product this close to protected classes carries disparate-impact exposure, and defending against that claim requires exactly the evidence this sector does not standardize: like-for-like cohort performance, documented at origination, reproducible by a third party. An operator who cannot produce a static pool by vintage for a credit committee also cannot produce one for a regulator or a plaintiff. The same missing layer that made Tricolor possible is what leaves the next generation of underwriting undefendable. One product, two economies The same payment curve that is an inconvenience at the top is a default at the bottom. Delinquency per Fitch Ratings via Auto Remarketing and Wolf Street; payments per Edmunds Q2 2026 (released July 16, 2026). Delinquency series are securitized-pool composites; payment figures are national averages across all credit tiers, so the two panels describe the same pressure rather than the same population. ## Limits: what the record does not show **Fitch monthly prints after March 2026.** Confirmed at 6.90% (January), 6.80% (February) and 6.11% (March). April through July are not confirmable from a named source and are treated as unobserved. Do not infer a summer trajectory from the March number, which is a tax-refund trough. Fitch itself expected the March improvement to be short-lived. **The January 2025 baseline is disputed between trade sources.** Wolf Street reports 6.56%, which is arithmetically consistent with the stated 34-basis-point year-over-year move to 6.90%. Auto Remarketing reports 6.45%. This study uses 6.56% and flags the conflict rather than picking silently. **The three build routes are an assessment, not a forecast with probabilities attached.** No rating agency has announced a post-Tricolor criteria change as of this writing. The SFA working groups are formed and have published no standard. Vendor coverage at scale is unverified, which is why no vendor is named in the body. Which route arrives first, or whether any does, is genuinely open. **Severity computable at origination is an inference about capability, not a published method.** The claim rests on the observed divergence between recoveries and wholesale prices. No public study was located that decomposes 2025 subprime severity into loan-structure and collateral-value components, so the size of each contribution is unquantified here. **Credit Acceptance detail not independently confirmed.** The quarterly forecast-change figures, the liquidity position, and the core subprime market-share movement cited in some coverage of the Q2 release could not be tied to a named primary source and are excluded from the body. The reported income, EPS, revenue, dealer count, volume and buyback figures are confirmed. The specific NYAG reserve figures that circulate in coverage, near $82.6 million in contingent losses against a potential $75.5 million payment, could not be confirmed and are excluded; the unresolved New York tail is confirmed through the April 24, 2025 CFPB withdrawal and the pending motion to dismiss. **America's Car-Mart's resolution.** Confirmed through the June 19, 2026 amendment and the July 14, 2026 10-K. Whether the September window produced Chapter 11, rescue capital, or a going-concern resolution is unobserved as of August 8, 2026. The debt components are stated at face value against a carrying total and do not sum. **Tricolor bank losses.** JPMorgan disclosed a charge-off of approximately $170 million and Fifth Third disclosed $178 million, for roughly $348 million combined. The Federal Reserve note characterizes the exposures as around $200 million each. Where those two accounts differ, the disclosed figures are used and the difference is noted here rather than averaged. **Figures carried from trade press rather than primary documents.** PrimaLend's funded-debt breakdown, the Fed note's BHPH delinquency comparison and its guarantee and structure percentages, and Tricolor's borrower-profile percentages are reported in coverage this study could not tie to a primary document. They are excluded from the body or attributed in place, and any that fail primary confirmation in a future pass get removed rather than softened. **The unemployment-to-delinquency elasticity.** Deliberately not quantified. The available parameters measure different quantities and are unstable across regimes. Any single number would be false precision, and any threshold derived from them would be uncalibrated. **The 1998 NAFA effort's fate.** The May 1998 ABI Journal confirms the effort was in progress. No primary source confirms formal adoption. That it was superseded by SEC Regulation AB II in 2014 rather than adopted as a trade standard is an inference, labeled as one. ## Three questions 1 **Can you price expected severity on a deal at the point of sale?** Severity is now set by advance rate against real collateral value, term, and rolled negative equity, all known the day the contract is written. If your severity assumption is still an auction-derived average applied after the fact, you are estimating a number you could be computing. 2 **When the composite stops describing you, what do you show instead?** Deep subprime is a third of S&P's subprime composite and the mix keeps moving. The operator who can hand a lender a like-for-like static pool by vintage gets judged on their own book. Everyone else gets judged on someone else's. 3 **When the verification layer arrives, are you being verified or doing the verifying?** A rating-agency criteria change, a consortium standard, and a vendor registry are three different bills. In all three, the operator already running cross-facility reconciliation and principal-paydown monitoring is describing what they already do. The rest are rebuilding under deadline, while funding is priced against the gap. The 2025 record settled what kind of problem this is. One detection failure, two operators the market read correctly, and a severity channel that moved somewhere nobody is watching. What follows from that is a market where proof of your own book becomes the thing that prices you, and the infrastructure to produce that proof gets built by someone within the next few years. That is the future of subprime. The operators who get there early will do it because they decided to know their own data before anyone made them. **Sources & notes** **Delinquency and recovery.** Fitch Ratings subprime and prime 60+ indices via Auto Remarketing ("Fitch: Stress in subprime surfaces through auto ABS trends," May 21, 2026) and Wolf Street (February 17, 2026): January 2026 record 6.90%, February 6.80%, March 6.11% with annualized net losses of 8.80% and recoveries of 37.48%, trailing-twelve-month average 6.25% from 5.98%, prime 0.4% and unchanged from January 2018 against a 0.9% Great Recession peak, pandemic trough 2.58% in May 2021. The 385-month framing per independent analysis of Fitch data by Bill Ploog and Auto Finance News. S&P Global Ratings full-year 2025 auto ABS review via Auto Remarketing: recoveries 37.74%, "the lowest seen on an annual basis going back to 2007"; 60-day delinquency 6.18% against 5.78%; annualized losses 8.88% against 8.51%; ABS issuance $127 billion in 2025 with $122 billion forecast for 2026. New York Fed Household Debt and Credit, Q1 2026. **Macro and affordability.** BLS Employment Situation for July 2026, released August 7, 2026: payrolls −23,000, unemployment 4.1%, participation 61.4%, May and June revised down a combined 103,000, twelve-month average 34,000. Federal Reserve, fed funds target 3.50% to 3.75%. Cox Automotive Manheim Used Vehicle Value Index: June 2026 at 212.9, up 2.1% year over year and 0.1% month over month, mid-July at 211.5. Kelley Blue Book and Cox Automotive average transaction price report for June 2026, released July 14, 2026: ATP $49,758, average payment $763, average rate 9.58% from 9.53% in May. Edmunds Q2 2026 negative-equity data, released July 16, 2026: 29.6% of trade-ins underwater from 26.6% a year earlier, average $6,884, prior quarter $7,183, Q2 2025 $6,754, negative-equity payment $944 against a $777 industry average. **Failures and survivors.** U.S. DOJ, SDNY, Tricolor indictment unsealed December 17, 2025 and the superseding eight-count indictment; Kollar and Seibold guilty pleas of December 16, 2025 per DOJ releases; Goodgame's June 24, 2026 guilty plea to six counts and his statement to the court per Reuters and Bloomberg; the Rakoff dismissal per Reuters, June 10, 2026. Wilmington Trust's loan-verification-agent role across nine securitizations, its failure to spot repeated VINs, its resignation as trustee, and the investor suit naming Wilmington Trust and backup servicer Vervent, per Bloomberg Law and Auto Finance News. PrimaLend First Day Declaration and case coverage: over-advance beginning August 2024, roughly $34 million of participations sold, CIBC default notices in February 2025, Chapter 11 filed October 22, 2025, plan confirmed February 20, 2026 with credit bids to CIBC and Amarillo National Bank. SEC EDGAR for America's Car-Mart: FY2026 Form 10-K filed July 14, 2026 reporting a $139.1 million net loss, approximately $722.4 million of total indebtedness including a $300.0 million Silver Point term loan and approximately $458.7 million of non-recourse securitization notes, 94 dealerships from 154, and going-concern language; the June 19, 2026 covenant amendment 8-K; the October 30, 2025 term loan 8-K. Credit Acceptance Q2 2026 release and earnings call, August 4, 2026, with the FactSet consensus of $12.20 per MarketScreener. **Structural.** Federal Reserve FEDS Note, Chyruk, Cox, Liu, Wang and Zoulalian, "Subprime Auto Lending: Trends in Buy Here Pay Here Auto Lending," May 8, 2026: roughly $32 billion in BHPH receivables, deep subprime above 50% from approximately 70% in 2018, balance growth of 214% against 34%, approximately 78% subprime origination share against 27%, weighted-average rate 25.39% against 14.60%, 16.63 times the active-repossession likelihood with roughly 5% of balances in active repossession in Q3 2025, more than $2 billion in bank commitments across roughly a dozen lenders and 82 obligors, probability of default up nearly 150% from Q2 to Q3 2025, and Tricolor advances at 60% to 80% of stated value. ABI Journal, "Subprime Auto Finance: The Year of the Bankruptcies," May 1998 (Buenzow, Pate, Sadarangani), for the static-pool prescription and the NAFA reporting-guidelines status. SEC Regulation AB II, 2014. Structured Finance Association and Ernst & Young Fraud Mitigation Survey findings, March 31, 2026. **Regulatory.** CFPB withdrawal from the joint action with the New York Attorney General against Credit Acceptance, April 24, 2025, with the case limited to New York consumers and the motion to dismiss pending. **Companion issues.** [Issue 4](https://lendriskanalytics.com/insights/cacc-stress-signals.html) is the Credit Acceptance stress-signal brief, covering the vintage table, the funding-cost series, and the regulatory tail. [Issue 5](https://lendriskanalytics.com/insights/three-failures-one-blind-spot.html) is the comparative postmortem on Tricolor, PrimaLend and America's Car-Mart, and is the case-level source for the detection argument here. [Issue 6](https://lendriskanalytics.com/insights/three-cycles-one-missing-layer.html) is the three-cycle study covering 1997-98, 2008-09 and 2022-26, and is the source for the warehouse and term-ABS split. Where this study reasons beyond what a document literally states, it is labeled as an inference. Charts 2, 3 and 4 are schematic or qualitative and are captioned as such; Chart 1's shaded zone represents a licensed comparative claim rather than per-year values. Figures that could not be tied to a named source are flagged in Limits rather than softened. Allegations in the Tricolor indictment are unproven as to Daniel Chu, who has pleaded not guilty; nothing here asserts wrongdoing by any lender. Point-in-time reading of the public record through August 8, 2026, with sector delinquency and recovery data confirmable through March 2026. LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned, and this is not investment, legal, or accounting advice. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). LR LendRisk Analytics Independent market research Continue reading [Issue · 06 · Deep study Three stress cycles, *one missing layer.*](https://lendriskanalytics.com/insights/three-cycles-one-missing-layer.html) [Issue · 05 · Comparative postmortem Three failures, *one blind spot.*](https://lendriskanalytics.com/insights/three-failures-one-blind-spot.html) --- title: "What the tape said: three stress cycles, one missing layer" url: https://lendriskanalytics.com/insights/three-cycles-one-missing-layer.html publisher: LendRisk Analytics series: What the Tape Said issue: 6 published: 2026-08-07 kind: Deep study description: "A deep study of the U.S. macroeconomy and subprime auto across 1997-98, 2008-09, and 2022-26. Three macro regimes, the same three causes of death, and the measurement layer the sector was told to build in 1998 and still has not. Every figure sourced; inferences labeled." html: https://lendriskanalytics.com/insights/three-cycles-one-missing-layer.html --- # What the tape said: three stress cycles, one missing layer [← All insights](https://lendriskanalytics.com/articles.html) What the Tape Said · Issue 6 Deep study · 18 min read Three cycles · 1997-98 · 2008-09 · 2022-26 # Three stress cycles, one missing layer. Public record through August 7, 2026 · Sector data confirmable through Q1 2026 · LendRisk Analytics Twelve subprime auto lenders went bankrupt in 1997 while GDP grew almost 4%. In 2008 the economy broke and the senior bonds paid anyway. Across 2025 subprime auto delinquency ran 6.18% against 5.78% the year before, an all-time high, and in January 2026 it touched a 32-year record. Unemployment never left the low fours the whole time. Three completely different economies, and the lenders died the same three ways in all of them. The autopsy that explained why was published in 1998. The fix it prescribed still has not been built. 37.74% Full-year 2025 subprime ABS recovery average, lowest in S&P data back to 2007 6.18% Full-year 2025 subprime DQ vs 5.78% in 2024 (S&P). Annual, so no seasonal distortion 4.2% Unemployment, June 2026. Every record above printed with no labor shock 28 years Since the 1997-98 autopsy prescribed dealer-level static-pool reporting **Data currency**Everything below on delinquency and recovery is verified through Q1 2026. Fitch's monthly subprime series is confirmable only through February 2026, and the S&P recovery series is annual and ends with full-year 2025. March through July 2026 is unobserved here. Statements about sector condition are dated to Q1 2026 throughout, and the historical case for cycles one and two does not depend on them. ## Bottom line Three cycles, three different macro regimes, and the same three causes of death every time: losses baked in at origination that reserving failed to recognize, funding concentration that turned a credit problem into a liquidity event, and collateral deterioration that no party outside the borrower could independently see. This cycle borrowed its plumbing from 1997. The money that can run sits in warehouse lines and private credit, which is where Tricolor and PrimaLend died and where Car-Mart's balance sheet came apart. The three outcomes are worth keeping separate: Tricolor liquidated under Chapter 7 with its founder facing a life-maximum charge, PrimaLend liquidated under a confirmed Chapter 11 plan, and Car-Mart is still open, still selling cars, and carrying a going-concern qualification with 60 of its 154 dealerships closed. But the recoveries tell a 2008 story the auction index hides. Lenders got back 37.74 cents on the repossessed dollar in 2025, the worst year in S&P's data, while wholesale prices held steady. Through the last data anyone can confirm, which is Q1 2026, unemployment was the only thing that had not moved. Underneath all three cycles is the same hole. In specialty finance, every downstream party reads a tape that one party writes, and no lender's mandate reaches past its own book. That hole killed lenders in a boom, in a crash, and now in an economy that looks fine. ## The read For a household in the bottom half, the car payment is the last bill to go unpaid. You need the car to get to work. TransUnion has measured that hierarchy since 2004 and found only two brief inversions: mortgage-first during pandemic accommodation programs, and card-before-mortgage in Q3 2008 when home equity went negative. So the last-paid bill going unpaid at a 32-year record, while prime delinquency sits at an eight-year low and unemployment sits at 4.2%, narrows the field considerably. Whatever is happening is happening to one cohort, and it is happening while that cohort is employed. Hold that against the other two cycles. In 1997-98, twelve subprime auto lenders filed for bankruptcy while GDP grew almost 4%. They died of funding withdrawal and losses that were wrong from origination. In 2008-09, the whole economy broke, and per asset-manager reporting no senior auto ABS tranche is documented to have taken a principal loss. The structures held while individual lenders nearly died of the same funding withdrawal. In 2025-26, Tricolor, PrimaLend, and Car-Mart failed or nearly failed on funding, fraud, and liquidity while the labor market stayed intact. The read The economy is the variable that keeps changing. The cause of death is the constant. A boom, a crash, and a soft landing produced the same three obituaries, which makes the economy the wrong suspect. **What holds steady is that nobody outside the borrower can independently verify what a lender holds, and no institution's mandate covers looking across facilities. That has survived all three cycles untouched.** ## What Issue 5 established [Issue 5](https://lendriskanalytics.com/insights/three-failures-one-blind-spot.html) read the three 2025-26 failures side by side. That finding sets the terms for everything below, so it goes first. Detection did not fail three times. It failed once. PrimaLend's lenders caught the deterioration fourteen months before the bankruptcy: CIBC's borrowing-base mechanics fired the quarter the first over-advance appeared, and the bank acted on it. Silver Point priced America's Car-Mart's distress accurately in advance, from audited public filings, and structured to take control when the default it expected arrived. Monitoring worked in both cases. What monitoring could not do was manufacture credit quality that had already deteriorated, or conjure a funding market that had repriced on someone else's fraud. Tricolor is the detection failure, and the shape of that failure is precise. Each warehouse lender checked VINs against its own portfolio. No party could see across facilities, so a scheme that double-pledged roughly $800 million in bogus collateral survived seven years against three sophisticated banks, a Big Four agreed-upon-procedures review, a rating agency, and a trustee. It was unwound by one junior analyst at a mezzanine lender who noticed that loans reported as current were not paying down principal. The gap was never intelligence or resources. It was mandate: every institution in the chain did roughly what its role required, and the job of looking across facilities belonged to nobody. PrimaLend's funders had analytics and used them well, so the missing piece was something else: *cross-facility visibility and independent verification of borrower-reported data*. The sector has gone without both through three complete macro regimes. The 1997-98 autopsy prescribed the fix explicitly: "the ability to reliably track the performance of loans purchased from each dealer, on a static pool basis." The National Auto Finance Association was, in 1998, "working on creating standardized financial performance reporting guidelines." Twenty-eight years later, banks hold more than $2 billion in identified commitments to roughly a dozen BHPH lenders against self-reported borrowing bases, the Fed counts roughly $32 billion in BHPH receivables with no census of the finance companies that fund the long tail, and bank-reported probability of default on BHPH lending repriced nearly 150% in a single quarter. After Tricolor. Not before. The 28-year gap The fix was prescribed in the first cycle's autopsy. The third cycle's failures happened without it. Schematic timeline, not measured data. 1998 prescription per the ABI Journal (May 1998); 2026 figures per the Federal Reserve FEDS Note of May 8, 2026. Intermediate failures per contemporaneous coverage. The three cycles below are the evidence. ## Cycle 1 · 1997-98: twelve bankruptcies in a boom The macro backdrop was as strong as any in modern history. Real GDP grew 3.9% in 1997. Unemployment averaged 4.9%, the lowest since 1973. CPI inflation eased to 1.6%. There was no recession anywhere near this crisis. And the sector collapsed anyway. In January 1997, Mercury Finance Co., the largest independent subprime auto lender at roughly $1 billion in assets, disclosed accounting irregularities forcing a four-year earnings restatement: 1996 net income restated from $120.7M to $56.7M, shareholders' equity from $353M to $263M (Mercury 8-K, SEC, January 29, 1997). The principal accounting officer disappeared. The treasurer was later indicted. CEO John Brincat ultimately received ten years for wire fraud (DOJ, 2002). Mercury's stock fell 93.9% in 1997 and the company was gone. Then the dominoes: Jayhawk Acceptance (Chapter 11, February 1997, after a $15.5M charge for unexpected credit losses), First Merchants Acceptance (Chapter 11, July 1997, after defaulting on its credit agreement; stock down 99.8%), Western Fidelity Funding (Chapter 11, August 1997), then Reliance Acceptance Group, Search Financial, First Enterprise Financial, and Keller Financial through early 1998 (ABI Journal, May 1998). Moody's counted 12 subprime auto originators filing for bankruptcy across 1997-1999, with 11 exits and 18 acquisitions; the ABI autopsy names nine through early 1998, a narrower window rather than a discrepancy. Net losses ran from under 3% in January 1995 to more than 10% by December 1997. ### What actually killed them The ABI Journal's contemporaneous autopsy names the sequence. Twenty-plus subprime auto IPOs in 1991-94 flooded the space with capital chasing volume, and underwriting standards deteriorated to win paper. Under gain-on-sale accounting, lenders securitized loans, kept the subordinated residual tranches valued on their own optimistic cash-flow models, then pledged those residuals as collateral for warehouse lines to buy still more loans. When losses ran above projection, residual values collapsed, advance rates broke, covenants tripped, and liquidity vanished. Commercial paper first, warehouse lines behind it. The Moody's special report of January 16, 1998, quoted in the ABI autopsy, made a pointed observation about the provision spikes: *"Ironically, the boost to the loan-loss provision was not driven by a sudden deterioration in portfolio performance, but was due to inadequate reserving at loan inception."* Read carefully, that is a finding about accounting timing. Borrower behavior was doing plenty: losses more than tripled across the same window. The underwriting of the land-grab years had already decided the shape of the loss curve. What broke was recognition. Reserves did not catch up until the cash flows made denial impossible, and by then the funding structure had turned a credit problem into a liquidity death. **Inference**The operative lesson is narrower than "reserve more." The origination-date loss expectation is the number that decides survival, and it has to be honest before the tape forces it to be. A reserve trued up after the cash flows break is a historical record, not a control. ### The survivor When the Russia default and LTCM crisis froze capital markets in autumn 1998, BB-rated AmeriCredit was running cash-flow deficits and dependent on securitization. Its guarantor, Financial Security Assurance, withdrew reinsurance, forcing AmeriCredit to fund the full 8% cash reserve on each securitization itself, roughly $50M more per $1B securitized (CFO.com). CFO Daniel Berce deliberately shrank originations to preserve liquidity. AmeriCredit lived; only four of the roughly thirty specialty finance companies that IPO'd in the 1990s remained listed. Protect the funding lifeline first. Everything else is second. Dying in a boom The economy and the sector, same 36 months. Loss path schematic between the two anchors; anchors and bankruptcy dates per Moody's via the ABI Journal (May 1998), trade-press-sourced, see Limits. Macro per BEA and BLS. ## Cycle 2 · 2008-09: the structures held, the lenders nearly didn't The opposite regime. Unemployment peaked at 10.0% in October 2009. The Manheim Used Vehicle Value Index fell 12.6% over three months in 2008 and took seven months to recover (Cox Automotive, via trade press; see Limits). Defaults surged at exactly the moment each repossession recovered less. Both barrels at once, which is what separates a real recession from everything else in this file. **Subprime auto dramatically outperformed subprime mortgage.** Four reasons. Payment priority: borrowers pay the car first because they need it for work. Repossession is fast and certain, against slow judicial foreclosure. Short duration means pools deleverage quickly. And most fundamentally, auto finance never assumed collateral appreciation. Cars were underwritten as depreciating assets; houses in 2006 were underwritten as appreciating ones. That single modeling assumption is why auto ABS held and mortgage ABS detonated. Fitch's subprime auto annualized net loss index peaked around 7.45% in August 2008 and roughly 8.78% in Q1 2009 (via trade press; see Limits). Elevated, and nothing like the RMBS wipeouts. S&P later designated the 2007-08 subprime auto vintages the "recessionary peak" benchmark. Per asset-manager reporting (Western Asset, November 2019), senior highly-rated auto ABS tranches came through the crisis with no documented principal losses and few downgrades. No verbatim agency statement of the same claim has been located, so the attribution stays with the asset manager everywhere it appears here. **Inference**The main exception was ratings downgrades on bonds whose AAA depended on monoline insurance wraps, when the monolines themselves were downgraded. A wrap-dependency ratings event, not evidence of a senior-tranche principal loss. **The lenders were a different story.** AmeriCredit's annualized net charge-offs hit 9.5% in the December 2008 quarter, and fiscal-2009 charge-offs ran 7.9% against 6.2% the prior year (AmeriCredit 8-Ks, SEC EDGAR; figures pending primary re-confirmation, see Limits). Its 2008-2 securitization prospectus carried going-concern language: "substantial doubt about AmeriCredit's ability to continue as a going concern" absent completing the deal and renegotiating warehouse lines (SEC 424B5, 2008). It was rescued partly by a significant-shareholder investment before GM bought it for $3.5B in 2010. The same lesson as 1998, ten years apart: the thing that nearly killed the strongest survivor was the funding line, not the loan book. **The structural counter-example.** Credit Acceptance's advance and holdback model transfers first loss: on Portfolio-program paper (72.1% of net receivables today), CACC advances dealers a fraction of expected collections and pays dealer holdback only after it has been made whole, so a vintage miss is absorbed by forfeited holdback, which is dealer money, before it touches CACC equity. Its 2009 vintage materially outperformed its initial forecast through the worst recession in eighty years. Structural loss absorption beats equity absorption, because equity absorbs losses only until it runs out. The double-hit Frequency and severity, simultaneously. The mechanism 1997 did not have, and one that 2022-26 had not shown through Q1 2026. Unemployment per BLS; anchors exact, path smoothed. Manheim path schematic around the disclosed 12.6% three-month decline and seven-month recovery, per Cox Automotive via trade press, see Limits. ## Cycle 3 · 2022-26: a record set in a good economy Unemployment has stayed inside 3.4 to 4.3% across the entire 2022-2025 window and printed 4.2% in June 2026 (BLS, July 2, 2026). And subprime auto delinquency set an all-time record: Fitch's 60+ index printed 6.56% in January 2025, then a record 6.90% in January 2026, up 34 basis points year over year and a 385-month high back to January 1994 (Fitch via Wolf Street; independent analysis by Bill Ploog). S&P's full-year 2025 subprime delinquency read 6.18% against 5.78% in 2024, described as an all-time high. Prime held at 0.4%, equal to January 2018 and less than half the 0.9% Great Recession peak. The bifurcation is the story: a tenfold-plus gap between prime and subprime, sustained for a year. That index is seasonal. The records print in January, when holiday spending and pre-refund cash gaps collide, and the rate eases through the spring as tax refunds land. February 2026 came in at 6.80%, and no monthly print after February is confirmable from a named source as of this writing, so the seasonal path through summer 2026 is unknown here. Which is why the annual figure carries the argument: S&P's full-year 2025 reading of 6.18% against 5.78% in 2024 has no seasonality in it at all. January against January, and year against year, both point the same direction. Quoting a January peak in August and calling it current is the error, and the honest version is that the last confirmable monthly reading is five months old. The 4.2% needs an asterisk too. June's rate came with payrolls up just 57,000 and labor-force participation down 0.3 points to 61.5%. Some of that stability is people leaving the labor force rather than finding work. "The labor market is fine" carries a lot of weight in this cycle's story, here included. ### What got more expensive Everything about owning a car got more expensive at once. New-vehicle average transaction price crossed $50,000 for the first time in September 2025 and set a record $50,326 in December 2025 (Kelley Blue Book / Cox Automotive), roughly a third above February 2020. The average new-vehicle payment reached $767 a month, with the average amount financed up $1,882 year over year to $43,582 and the average used payment at $537 (Experian, Q4 2025). Insurance rose about 54% from 2020 to 2024 (USAFacts), with BLS data putting premiums up 55% since February 2020, almost all of it between 2022 and 2024. Cumulative CPI of roughly 22 to 25% since 2020 landed on stagnant real wages for bottom-half earners, student-loan 90+ delinquency reached 10.3% of balances (NY Fed, Q1 2026), and pandemic savings are gone for the lower quintiles. The cleanest evidence that this lands on one cohort is in the credit data itself. Prime 60+ delinquency sat at 0.4% in January 2026, unchanged from January 2018 and less than half its Great Recession peak, while subprime set a 32-year record in the same month. Same economy, same month, same asset class, a tenfold gap sustained for a year. Fitch reads it the same way, describing inflation stress offsetting solid aggregate balance sheets "especially for lower-income households." The widely-cited Moody's estimate that the top 10% of earners drove 49.2% of consumer spending in Q2 2025 points in the same direction, and its methodology is contested by the Minneapolis Fed and others, so nothing above rests on it. The affordability stack, Feb 2020 → Dec 2025 The borrower's costs moved. The borrower's income didn't. ATP per Kelley Blue Book / Cox Automotive (Dec 2025 report); insurance per BLS via NPR and USAFacts; payments and amounts financed per Experian Q4 2025. Wage bar directional, not a plotted series. ### The severity read most people have is wrong The standard read, and the one an earlier draft of this series carried, is that severity has gone quiet because wholesale values stabilized. Manheim peaked at a record 257.7 around January 2022, fell 14.9% in 2022, the largest one-year decline in series history, bottomed at 196.1 in June 2024, and closed 2025 at 205.5, still far above the pre-pandemic norm. The recovery data says otherwise. S&P reports subprime auto ABS recoveries averaged **37.74% in 2025, the lowest annual level in data back to 2007**. Worse than any Great Recession year in the series. Wholesale prices measure the general fleet at auction. Subprime repossession recovery is a different animal: older units, higher mileage, worse condition, rising repossession and transport and reconditioning costs, in a segment the Fed just documented running a 16.63x repossession rate against traditional auto lending. A stable wholesale index with collapsing realized recoveries means severity is deteriorating through cost and mix, invisible to anyone reading Manheim off a chart. The same caveat that applies to the delinquency record applies here. The recovery average is a pool-level figure, and issuer mix, vehicle age, and deep-subprime share all drift over eighteen years, so deterioration and composition cannot be separated from public data on either number. What survives the caveat is narrower and still holds. Whatever the mix, the cash coming back per repossessed dollar is the lowest S&P has recorded since 2007, and the wholesale index would never tell you. And the deterioration is now being built in at origination. In Q4 2025, 29.3% of new-vehicle trade-ins carried negative equity, the highest share since Q1 2021, and the average amount owed on underwater trade-ins hit a record $7,214, with 27% carrying $10,000 or more, also a record (Edmunds). 40.7% of negative-equity purchases were financed on 84-month terms, at an average $916 a month. **Inference**A loan that starts above 100% LTV and amortizes slower than the collateral depreciates has impaired recovery from day one, regardless of what the wholesale index does. Severity no longer needs a price crash to deteriorate; it is being originated. And the wholesale stability itself has a scheduled expiry: thin 3-to-6-year-old supply from the 2020-22 new-sales and lease collapse is propping values, and that cohort normalizes over the next two years. The severity divergence Stable auction prices, collapsing realized recoveries. The two series most readers assume move together. Both end at full-year 2025, the latest annual data available. Manheim anchors per Cox Automotive / WardsAuto; path smoothed between anchors; pre-pandemic average trade-press-sourced, see Limits. Recovery figure and the 2007-2024 comparison per S&P via Auto Remarketing; the shaded zone represents the licensed claim that 2025 is the series low, not per-year values, which sit in paywalled S&P data. Both series are pool-level and subject to composition drift. ### How the 2025-26 lenders actually failed [Issue 5](https://lendriskanalytics.com/insights/three-failures-one-blind-spot.html) covers these cases in depth. The short version, with the numbers current to August: **Automotive Credit Corp** (Southfield, MI, 33 years old) paused all originations indefinitely on August 7, 2025, citing "internal and external financial conditions." **Tricolor Holdings** filed Chapter 7, a liquidation rather than a reorganization, on September 10, 2025. Per the DOJ indictment unsealed December 17, 2025: approximately $2.2B pledged against approximately $1.4B of real collateral, roughly $800M in bogus collateral, double-pledged and fabricated across three warehouse lenders over seven years. JPMorgan charged off roughly $170M; Fifth Third disclosed $178M; $348M combined. CFO Jerome Kollar and finance executive Ameryn Seibold pleaded guilty December 16, 2025. COO David Goodgame pleaded guilty to six counts on June 24, 2026 and is cooperating. The same day, a superseding eight-count indictment against founder and CEO Daniel Chu added a Continuing Financial Crimes Enterprise charge, the rarely-used financial kingpin statute, which carries a maximum of life. Chu pleaded not guilty; trial is set for October 19, 2026. Allegations are unproven as to Chu. **PrimaLend Capital Partners** (BHPH warehouse and floor-plan lender, roughly $280M in dealer loans) filed Chapter 11 on October 22, 2025, after dealer-borrower defaults pushed it into over-advance beginning August 2024. Its lenders caught the deterioration fourteen months before the filing and acted on it, the monitoring counter-example per Issue 5. Plan confirmed as a liquidation February 20, 2026. **America's Car-Mart** is further along than most coverage reflects. The FY2026 10-K, filed July 14, 2026, shows revenue of $1.281B, down 7.9%; a net loss of $139.1M against prior-year net income of $17.9M; full-year EPS of −$16.79; the dealership footprint cut from 154 to 94, a 40% reduction; explicit going-concern language; and, critically, no revolving or warehouse facility at all. The company depends on operations and securitizations. The October 2025 Silver Point term loan ($300M, SOFR+7.50%, warrants for up to 10% of shares) replaced the revolver; by June 2026 Car-Mart was in default, Silver Point had board representation, inventory was down 52%, the stock had touched $1.67, its lowest since the 1992 IPO, and lenders had extended the runway only into early September 2026. **Inference**A public company carrying a going-concern qualification with no committed revolving facility is running the Mercury Finance liability structure in 2026. **The repricing.** Post-Tricolor, warehouse lenders pulled back from subprime broadly (Drive Now Acceptance's CFO confirmed capital providers retreating; Auto Finance News, July 2026) and private credit stepped in at distressed pricing. The Fed's May 8, 2026 FEDS Note on BHPH lending (Chyruk, Cox, Liu, Wang, Zoulalian) is the best structural document of this cycle. BHPH loans are 78% subprime against 27% for traditional auto; BHPH balances grew 214% since 2018 against 34%; BHPH loans are 16.63x more likely to be in active repossession, with roughly 5% of balances in active repossession in Q3 2025; weighted-average BHPH subprime APR of 25.39% against 14.60%; banks holding more than $2B in commitments to roughly a dozen BHPH lenders; and bank-reported probability of default on BHPH lending up nearly 150% from Q2 to Q3 2025. A repricing that happened after Tricolor, reactively, against borrowing bases the borrowers themselves report. The note observes that Tricolor's advances ran at only 60 to 80% of stated value through SPE structures, and the fraud defeated those protections anyway, which is worth sitting with. The banks had the legal structure right and still could not see what they held. One point from Issue 5 is worth repeating here, because it is the cleanest evidence that the credit line and the funding line are separate systems. Car-Mart's loan book was improving through this entire period, charge-offs declining and newer vintages cleaner, while its funding repriced on another company's fraud. No amount of underwriting discipline reaches that risk. The only thing that does is knowing what your next-best facility costs before the week you need it. ## What fired, and when | Channel | 1997-98 | 2008-09 | 2022-26 | |---|---|---|---| | Unemployment → frequency | Did not fire (4.9%, GDP +3.9%) | Fired hard (to 10.0%) | Not firing (4.2%, participation caveat) | | Collateral → severity | Not central | Fired via price (−12.6% in 3 months) | Firing via cost and mix (37.74% recoveries, record day-one negative equity) while prices hold | | Rates → funding cost | Fired (CP froze; LTCM) | Fired (markets froze) | Fired (spreads wider; Fed at 3.50-3.75% with markets pricing possible hikes) | | Inflation → payment capacity | Muted (CPI 1.6%) | Muted | Fired, the defining channel (prices, insurance, flat real wages) | | Funding confidence → death spiral | Fired, the killer (Mercury CP loss) | Fired (AmeriCredit going-concern) | Fired, the killer (post-Tricolor repricing: ACC, PrimaLend, Car-Mart) | Which channels killed lenders: 1997, channel five on losses baked in at origination. 2008, channels one and two simultaneously, with five nearly finishing the survivors. 2022-26, channels four and five, with two partially engaged through cost and mix. Five channels, three cycles Through Q1 2026, four of five were on. The one still off did the most damage in 2008. Qualitative assessment, not measured data; sourced per the table above. Dark red marks the channel that killed lenders in that cycle. The 2022-26 column is read through Q1 2026, the last period with confirmable delinquency and recovery data. March to July 2026 is unobserved here, so the column is a state as of Q1 rather than as of publication. "This rhymes with 1997" is close, and too loose. What this cycle borrowed from 1997 is the plumbing. The money that can run sits in the pre-securitization layer, warehouse lines and private credit, which is where all three 2025-26 failures happened. Term ABS does not run, and that layer has been structurally sound since the residual-pledging era ended. What it borrowed from 2008 is the severity, hidden in cost and mix rather than price. Four of five channels are on. The one still off did the most damage in 2008. ## What happens if unemployment rises Two published estimates get quoted in this debate, and they are routinely treated as contradicting each other. They do not, because they measure different quantities. A Federal Reserve working paper, pointedly titled *Macroeconomic Parameter Instability in Auto Loan Loss Models* (Fritsch and Prescott), puts the *relative* sensitivity of default probability to a one-point unemployment rise at roughly 16% in 2006 and roughly 3% post-2020: a proportional change inside an auto loss model. The Richmond Fed's 2020 work (Zhu Wang) puts the *absolute* move at roughly 0.54 percentage points of default rate per point of unemployment, with stress projections near 14% default rates at 20% unemployment. One is a percentage change in a probability, the other is percentage points on a rate. Stacking them side by side as a range is a units error. The finding that survives is the one in the Fed paper's title. The sensitivity itself has fallen since 2006, which is consistent with a default cycle now driven by prices and payment capacity instead of by job loss: when defaults are already being generated by something other than unemployment, unemployment explains less of the variance. Read that way, the low number describes the recent past accurately and says almost nothing about a labor break. **Inference**A dulled historical sensitivity, estimated across a period when joblessness was not the binding constraint, tells you little about what happens when joblessness becomes the binding constraint. An unstable parameter sitting on top of a record delinquency base is itself the risk. The estimates are used here as evidence of instability, never as point forecasts. What can be said without a model is narrower, and it comes with a date attached. Through Q1 2026, this cycle's record delinquency and worst-since-2007 recoveries had been produced with the unemployment channel off. The sector spent its good-economy cushion against a benign labor market. A labor break arriving on top of that base would be the first time since 2008 that all five channels fired together. S&P's stress framework, which replays a 10%-unemployment 2008 scenario, argues most senior subprime tranches remain resilient even then. How much cushion is left as of August is genuinely unknown here. The monthly delinquency series runs cold after February 2026 and the recovery series is annual, so March through July 2026 is unobserved. If delinquency normalized sharply across the spring, the base a labor shock would land on is lower than the one described above, and the asymmetry below is correspondingly smaller. The historical argument does not move either way. What 1997 and 2008 show about how lenders die, and what the 28-year measurement gap has cost, does not depend on where the tape printed in June. **Inference**That is a ratings-resilience statement about bondholders, not about lender-equity survival. 1997 and 2008 both showed that the bonds surviving and the lenders surviving are different questions. Nothing in the current tape predicts a labor break: jobless claims are low, and Fitch's own 2026 outlook expects deterioration against 2025 on a "cooling labor market" rather than a rupture. But the asymmetry is what matters. The upside case is that weak vintages amortize out and the record proves partly compositional. The downside case is 2008's mechanism arriving on top of a base 2008 never had. **Thresholds, and what they are not.** Sustained unemployment above roughly 4.7 to 5.0% for two quarters, or Manheim decisively below roughly 190, would be the points at which to shift from monitor-and-reserve to active de-risking: cut volume, raise down-payment and LTV floors, pre-negotiate incremental liquidity while it exists. Conversely, if the 2022-2023 vintages keep amortizing out and later cohorts keep tracking to plan, treat the elevated headline as partly a composition artifact, which per the Limits below is partly what it already is. **Inference**Those numbers are monitoring triggers set by judgment, and calling them anything more would contradict the paragraph above. If the published elasticities do not transfer across regimes, no threshold derived from them is calibrated either, including these. They mark where an operator should stop assuming and start acting, and they carry no claim about how much loss follows. Anyone presenting a specific unemployment level as a modeled loss trigger for this cycle is overstating what the literature currently supports. ## The functioning-sector case The counter-arguments have real weight, and two of them are strong enough to change how you read everything above. **The structures are genuinely stronger.** Per asset-manager reporting, senior auto ABS came through the worst recession in eighty years with no documented principal losses. Excess spread, overcollateralization, and sequential-pay subordination work; deep-subprime deals routinely carry 20%+ expected cumulative losses and pass them through without senior investors taking a hit. This market bears no resemblance to 1997's residual-pledging loop. Credit Acceptance discloses that its securitizations are structured to withstand a 35% decline in forecasted collection rates before the most junior bond is at risk. Its worst vintage miss in a decade, the 2022 book, is 8.2 points against a 67.5% initial forecast, a relative decline of about 12%. Measured the same way the cushion is measured, the worst miss in a decade ate about a third of it. **Recent vintages are stabilizing.** CACC's 2025 vintage is tracking +0.2% against initial forecast (67.2% vs 67.0%, per the Q4 2025 vintage table); initial spread on new assignments held roughly flat at about 22.0% against 22.1% in 2024, pricing discipline while ceding volume. Its quarterly downward forecast revision shrank from $189.3M in Q2 2024 to $9.1M in Q1 2026, a 95% deceleration. S&P found the 2023 and Q1 2024 subprime cohorts improving against 2022's highs. The full CACC read, including the live NY AG matter and the $82.6M in contingent losses booked against a potential $75.5M settlement, is in [Issue 4](https://lendriskanalytics.com/insights/cacc-stress-signals.html). The model works, and it carries an unresolved regulatory tail that any operator pointing at it should price separately. **The record is real, narrow, and not decomposable from public data.** The Fitch index tracks the securitized subprime slice, a minority of outstanding auto loans, and its composition drifts as issuers enter, exit, and shift mix. For scale, the Fed's BHPH note puts BHPH alone at roughly 2% of the $1.6 trillion auto market and 5% of the subprime market. Fitch itself attributes part of the elevation to composition effects as the weak 2022-2023 vintages age through. The aggregate picture is far calmer: NY Fed data shows 4.8% of all household debt in some stage of delinquency, a household-debt figure rather than an auto figure, with auto transition rates holding steady in Q1 2026 against $1.69T in balances. How much of the 6.90% record was borrower deterioration and how much was issuer-mix drift cannot be separated from public data. The record says something true about the securitized subprime borrower, on a narrow slice, and the February easing to 6.80% belongs in the same picture. **Rate relief is genuinely uncertain in both directions.** The Fed cut three times in 2025 to 3.50-3.75%; by mid-2026 markets had swung to pricing possible hikes on renewed inflation. The funding-cost channel could ease or tighten from here, which cuts against both the bear and bull cases. ## Limits: what the record does not show **The 2008-column anchors are trade-press-sourced.** Manheim's 12.6% three-month 2008 decline, the 14.9% 2022 full-year record, the 152.9 pre-pandemic average, Fitch's 7.45% and 8.78% loss-index peaks, Moody's under-3%-to-over-10% 1997 loss path, and AmeriCredit's 9.5%, 7.9%, and 6.2% charge-off figures all trace to trade coverage of paywalled or archival primary reports. They are directionally solid and multiply corroborated, and they are flagged here rather than silently trusted. Any that fail primary confirmation in a future pass get removed rather than softened. **The recovery figure carries the same composition caveat as the delinquency record.** The 37.74% is a pool-level average; issuer mix, vehicle age, and deep-subprime share drift over the comparison window, and deterioration cannot be separated from mix using public data. **No exact 2007-08 subprime cumulative-net-loss figure** from a free named source; the "recessionary peak" framing is S&P's, and the precise CNL sits behind paywalls. **The "no senior principal loss" claim** rests on asset-manager language (Western Asset, Janus Henderson) rather than a verbatim agency statement, and is attributed accordingly everywhere it appears in this study. **Income-cohort auto delinquency splits are inferred** via the prime and subprime ABS split and TransUnion's K-shaped framing, and are not directly published at the granularity the argument wants. The Moody's 49.2% spending-share figure is contested (Minneapolis Fed; UC Berkeley), is presented as a widely-cited estimate, and carries none of the argument; the prime-against-subprime delinquency gap does that work instead. **The unemployment elasticities (16%, 3%, 0.54%)** come from single working-paper models and are used only as evidence of parameter instability, never as point estimates. The Fritsch and Prescott figures are relative sensitivities of default probability; the Richmond Fed figure is percentage points of default rate. They are not directly comparable and are not presented here as a range. **The monthly delinquency series runs cold after February 2026.** Fitch's January 2026 record of 6.90% and the February easing to 6.80% are the last readings confirmable from a named source as of August 7, 2026. Any monthly print between March and July 2026 is unknown here, so the seasonal path through summer is not characterized. The trade sources also disagree on the January 2025 baseline: Wolf Street reports 6.56%, which is arithmetically consistent with the stated 34-basis-point year-over-year move, and Auto Remarketing reports 6.45%. The 6.56% is used above, and a 6.31% June figure that circulates in coverage belongs to June 2025, not 2026. **America's Car-Mart's ultimate outcome is unresolved.** No Chapter 11 as of August 7, 2026; the lender runway extends into early September. The going-concern language and facility structure are from the FY2026 10-K, filed July 14, 2026. **Tricolor allegations are unproven as to Chu**, who has pleaded not guilty. The double-pledged loan counts circulating in coverage (roughly 31,000 loans) derive from a dismissed noteholder complaint and are treated as alleged, not established. Nothing here asserts wrongdoing by any lender. **Timing against credit.** Part of the residual forecast drag at disclosed lenders reflects slower prepayment, which is timing, rather than default, which is credit; public tables do not cleanly separate the two. ## Three questions for operators and the lenders who fund them 1 **Who can see across all your funding facilities at once?** If the answer is only you, the answer is nobody independent. Tricolor's fraud survived seven years specifically because every lender's mandate ended at its own collateral schedule, and it was unwound by a single cross-facility observation: current loans that weren't paying down principal. Run that reconciliation on yourself, remittance file against status file, monthly, and be able to hand a lender the result before they ask. 2 **Is your origination-date loss number honest before the tape forces it to be?** The 1997 lenders died in a boom because their losses were real at origination and unrecognized until cash flows made denial impossible. Under CECL the mechanics differ; the discipline doesn't. Monitor at vintage and static-pool level rather than aggregate, treat a later cohort crossing above an earlier one at the same seasoning as an alarm rather than a footnote, and run your severity assumption against realized recoveries rather than the wholesale index. The 2025 gap between the two is the widest in the modern record. 3 **What does your funding cost the day your sector has a bad quarter you had nothing to do with?** Car-Mart's book was improving when its funding repriced on another company's fraud, and the replacement structure, one term lender and no revolver, is the structure that ended Mercury Finance. Multiple facilities, matched duration, and a priced next-best option before distress. The credit line and the funding line are separate systems, and only one of them takes your underwriting into account. **Sources & notes** **Cycle 1 · 1997-98.** ABI Journal, "Subprime Auto Finance: The Year of the Bankruptcies" (May 1998, Buenzow, Pate, Sadarangani), for the failure roster, the causes sequence, the static-pool prescription, the NAFA reporting-guidelines status, and the quoted Moody's special report of January 16, 1998. Mercury Finance Form 8-K (SEC, January 29, 1997) for the restatement and equity figures. DOJ (2002) and CFO.com (2007) for the treasurer indictment and Brincat's sentence. CFO.com, "Less Business Wanted," for the AmeriCredit and Financial Security Assurance episode, the roughly $50M per $1B figure, and the survivor count. Moody's sector counts (12 filings, 11 exits, 18 acquisitions, 1997-1999; net losses under 3% to over 10%) via American Banker and the ABI Journal, trade-press-sourced, see Limits. 1997 macro (GDP +3.9%, unemployment 4.9%, CPI 1.6%) per BEA and BLS. **Cycle 2 · 2008-09.** BLS for the 10.0% October 2009 unemployment peak. Cox Automotive via trade press for the Manheim 12.6% three-month decline and seven-month recovery. Fitch loss-index readings (about 7.45% August 2008, about 8.78% Q1 2009) via Auto Remarketing and F&I Magazine. S&P's "recessionary peak" designation via Auto Remarketing (2018). Western Asset Management (November 2019) for the senior-tranche performance claim, attributed as asset-manager reporting throughout. AmeriCredit 8-Ks (January 29 and August 5, 2009) and the 2008-2 424B5 going-concern language, SEC EDGAR, pending primary re-confirmation per Limits. GM's $3.5B acquisition (announced July 2010) per SEC filings and contemporaneous coverage. Credit Acceptance 2009-vintage outperformance per then-CEO Brett Roberts via Auto Remarketing. CNN Business (September 2025), quoting Prof. Pamela Foohey, on payment priority. **Cycle 3 · 2022-26.** Fitch subprime and prime 60+ indices: the 6.90% January 2026 record, 6.56% January 2025, the 2.58% May 2021 trough, and prime at 0.4% per Fitch via Wolf Street (February 17, 2026); the 385-month framing per independent analysis of Fitch data by Bill Ploog and Auto Finance News; the February 2026 easing to 6.80% per Wolf Street. Auto Remarketing (May 21, 2026) reports the January 2025 baseline as 6.45% against Wolf Street's 6.56%, noted in Limits. S&P full-year subprime delinquency (6.18% against 5.78%) and the 37.74% recovery average, lowest since 2007, via Auto Remarketing. BLS Employment Situation for June 2026 (released July 2, 2026): unemployment 4.2%, payrolls +57,000, participation 61.5%. Federal Reserve FEDS Note, "Subprime Auto Lending: Trends in Buy Here Pay Here Auto Lending" (Chyruk, Cox, Liu, Wang, Zoulalian, May 8, 2026), for every BHPH figure cited. NY Fed Household Debt and Credit, Q1 2026: $1.69T auto balances, 10.3% student-loan 90+, 4.8% aggregate household-debt delinquency, transition rates. Edmunds Q4 2025 Insights for the negative-equity records. Experian State of the Automotive Finance Market, Q4 2025, for payments and amounts financed. Kelley Blue Book / Cox Automotive ATP reports (September and December 2025) for the $50,000 crossing and the $50,326 record. USAFacts and BLS via NPR (October 30, 2025) for insurance. Cox Automotive and WardsAuto for Manheim (257.7 peak, 196.1 trough, 205.5 December 2025). Moody's Analytics / Zandi via Bloomberg (September 16, 2025) for the 49.2% spending share; Minneapolis Fed (2026) for the methodological review. TransUnion payment-hierarchy research (2012, 2019) and Q1 2026 K-shaped framing. Fritsch and Prescott, "Macroeconomic Parameter Instability in Auto Loan Loss Models," Federal Reserve, for the 16% and 3% relative sensitivities. Richmond Fed (Zhu Wang, April 16, 2020) for the 0.54-percentage-point estimate and the 20%-unemployment stress projection. **The 2025-26 failures.** Tricolor: DOJ indictment unsealed December 17, 2025 (SDNY) for the collateral figures and duration; guilty pleas per DOJ releases and Reuters and Bloomberg (June 24, 2026); the superseding indictment and Continuing Financial Crimes Enterprise charge per the same; the Rakoff dismissal per Reuters (June 10, 2026); Deloitte agreed-upon-procedures scope per SEC Form ABS-15G; JPMorgan's roughly $170M charge-off per the October 14, 2025 earnings call via Banking Dive; Fifth Third's $178M per Reuters (October 17, 2025) and its Q3 2025 call. PrimaLend: First Day Declaration and case coverage; plan confirmed February 20, 2026. America's Car-Mart: FY2026 Form 10-K (filed July 14, 2026) for revenue, net loss, EPS, footprint, going-concern language, and facility structure; Silver Point term loan 8-K (October 30, 2025); First Amendment and Limited Waiver 8-K (filed June 22, 2026); Bloomberg via TT News for the June 10, 2026 decline to $1.67; subsequent Bloomberg and press coverage (July 2026) for board representation, the Houlihan Lokey process, and the September runway. ACC pause: Non-Prime Times and Auto Finance News (August 2025). Sector repricing: Auto Finance News (July 2026). **Companion issues.** [Issue 4](https://lendriskanalytics.com/insights/cacc-stress-signals.html) (Credit Acceptance) for the vintage table, funding-cost series, structural cushion, and regulatory status. [Issue 5](https://lendriskanalytics.com/insights/three-failures-one-blind-spot.html) (Tricolor, PrimaLend, Car-Mart) for the case-level detection analysis the argument above rests on. Where this study reasons beyond what a document literally states, it is labeled as an inference. Charts 1, 2, 3, and 6 are schematic or qualitative and are captioned as such; Chart 5's shaded zone represents a licensed comparative claim rather than per-year values. Figures flagged in Limits are trade-press-sourced pending primary confirmation and are subject to removal rather than revision if they fail. Allegations in the Tricolor indictment are unproven as to Chu, who has pleaded not guilty; nothing here asserts wrongdoing by any lender. Point-in-time reading of the public record through August 7, 2026. LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned, and this is not investment, legal, or accounting advice. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). LR LendRisk Analytics Independent market research Continue reading [Issue · 05 · Comparative postmortem Three failures, *one blind spot.*](https://lendriskanalytics.com/insights/three-failures-one-blind-spot.html) [Issue · 04 · Stress-signal brief What the tape said: *Credit Acceptance.*](https://lendriskanalytics.com/insights/cacc-stress-signals.html) --- title: "What the tape said: Tricolor, PrimaLend, Car-Mart" url: https://lendriskanalytics.com/insights/three-failures-one-blind-spot.html publisher: LendRisk Analytics series: What the Tape Said issue: 5 published: 2026-08-03 kind: Comparative postmortem description: "Three subprime auto lenders failed or nearly failed in nine months, funded by JPMorgan, Fifth Third, Barclays, CIBC and Silver Point. A comparative postmortem on what each institution actually missed, and why the answer is different in all three cases. Every figure sourced; inferences labeled." html: https://lendriskanalytics.com/insights/three-failures-one-blind-spot.html --- # What the tape said: Tricolor, PrimaLend, Car-Mart [← All insights](https://lendriskanalytics.com/articles.html) What the Tape Said · Issue 5 Comparative postmortem · 16 min read Tricolor · PrimaLend · America's Car-Mart # Three failures, one blind spot. Public record through August 3, 2026 · LendRisk Analytics Three subprime auto lenders failed or nearly failed in nine months. Every one of them was funded by institutions with more analysts, more capital, and more surveillance infrastructure than anyone reading this will ever have. JPMorgan. Fifth Third. Barclays. Silver Point. CIBC. These are not naive lenders, and they still lost, collectively, more than half a billion dollars. The useful question is how the giants funding them failed to see it coming, and the answer turns out to be different in each case. That is the reason to read all three together. ~$800M Gap between what Tricolor pledged (~$2.2B) and what existed (~$1.4B), per the DOJ indictment ~7 years How long the scheme ran undetected across three warehouse lenders, 2018 to 2025 $340M+ Combined JPMorgan and Fifth Third disclosed losses on a single borrower Dismissed Noteholder fraud suit against the three banks, June 10, 2026 ## Bottom line These three failures look identical from a distance and share almost nothing up close. Tricolor was criminal fraud, an indicted scheme that fabricated loans and pledged the same collateral to multiple lenders simultaneously for seven years. PrimaLend was a credit failure at a lender whose own funders caught the deterioration early and acted on it. Car-Mart was a liquidity failure that sat in plain view in its own audited public filings before its final lender ever wrote the check. What connects them is structural rather than a matter of competence. In specialty finance, the lender's picture of the collateral is assembled from data the borrower controls. When the borrower originates the loans, services the loans, and reports on the loans, every downstream party, the warehouse lender, the rating agency, the ABS investor, the trustee, is reading a tape that one party writes. Tricolor showed what happens when that party writes fiction. PrimaLend and Car-Mart showed that the same structure gives a lender too little warning to act in time even when nobody lies. The read The most uncomfortable fact in this file is who actually caught Tricolor. Not JPMorgan's credit committee. Not Fifth Third's field examiners. Not KBRA, not Deloitte, not Wilmington Trust as indenture trustee. A junior analyst at a mezzanine lender noticed that loans reported as current were not paying down principal the way current loans do. **The cash was not behaving the way the tape said it should.** That is one observation, available to anyone holding the remittance data, and it unwound a seven-year, $800 million scheme in a matter of weeks. Everything expensive missed it. One person reading the numbers carefully did not. ## Case 1 · Tricolor, and the verification vacuum that let fraud run seven years Tricolor Holdings was an Irving, Texas subprime auto retailer and lender serving largely Hispanic, thin-file borrowers. CDFI-certified. Seventh-largest independent used car chain in the country. It originated its own loans and serviced them in-house. The DOJ indictment unsealed December 17, 2025 alleges that at founder and CEO Daniel Chu's direction, executives "repeatedly double-pledged collateral to multiple lenders and manipulated the characteristics of collateral to make ineligible, near-worthless assets appear to meet lender requirements." By August 2025, Tricolor had pledged approximately $2.2 billion of collateral against approximately $1.4 billion of real collateral. The difference, per the Justice Department, was "approximately $800 million in bogus collateral." ### The mechanism Double-pledging means selling the same asset twice. Pledge loan #1234 to Lender A as collateral for an advance, then pledge the same loan #1234 to Lender B for a second advance. Both lenders believe they hold a first-priority security interest in the same car. Neither can see the other's collateral schedule. Tricolor also fabricated loans that never existed and falsified payment records to make delinquent accounts appear current. What double-pledging looks like mechanically One loan, two lenders, neither able to see the other's collateral schedule. Schematic, not measured data. Per the DOJ indictment unsealed December 17, 2025, Tricolor repeatedly pledged the same collateral to multiple lenders between 2018 and 2025. ### Why nobody caught it for seven years Each warehouse lender checked VINs against its own portfolio. There was no cross-lender registry, no shared collateral database, and no mechanism by which JPMorgan could learn that a VIN it had just financed was already sitting on Fifth Third's borrowing base. The fraud was structurally invisible to any single lender acting alone, and would have been trivially visible to anyone able to compare across facilities. How long the scheme ran before anyone looked across facilities Tricolor pledged the same collateral to multiple lenders from 2018 until detection in August 2025. Schematic timeline. Duration of the alleged scheme per the DOJ indictment unsealed December 17, 2025. Detection point per contemporaneous reporting. **Inference**The control that would have caught this is neither exotic nor expensive. Matching UCC-1 filings across secured parties, plus VIN-level de-duplication across facilities, is a data-matching exercise costing hundreds of dollars per entity rather than millions. Cost was never the obstacle. The obstacle was that no institution owned the job of looking across facilities, because no single lender's mandate extends past its own book. ### What the diligence actually covered A Deloitte agreed-upon-procedures review was performed and disclosed on the rated ABS via SEC Form ABS-15G. It sampled a small fraction of the pool, relied on Tricolor's own data, found two FICO discrepancies, and expressly disclaimed any representation as to the existence or ownership of the receivables. KBRA relied on static pool data, an operational review, and periodic update calls with the company. The verification chain existed on paper. Every link in it terminated at data Tricolor supplied. ### Where it stands now CFO Jerome Kollar and finance executive Ameryn Seibold pleaded guilty December 16, 2025 and are cooperating. COO David Goodgame pleaded guilty June 24, 2026 to six counts, covering bank fraud, securities fraud, wire fraud, conspiracy, and false statements, and agreed to cooperate. "I knew that Tricolor was deceiving and defrauding the banks," Goodgame told Judge Castel. The same day, prosecutors filed a superseding eight-count indictment against Chu adding a Continuing Financial Crimes Enterprise charge, the rarely-used financial kingpin statute, which carries a maximum of life. Chu pleaded not guilty June 30. Trial is set for October 19, 2026. ## Case 2 · PrimaLend, the case where monitoring worked PrimaLend Capital Partners was an asset-based lender to BHPH dealers rather than an operator itself: revolving lines, floor plan, sub-debt, roughly $280 million in loans across twelve states, about two-thirds of it in Texas. Plano, Texas. About 35 employees. There is no fraud allegation here. What happened was the subprime credit cycle arriving on schedule. Post-2022 inflation and rate increases hit BHPH dealers, those dealers lost the consumer payment flow that serviced PrimaLend's lines, and PrimaLend's own collateral began failing. ### The timeline that matters | Date | Event | |---|---| | Aug 2024 | Losses from dealer defaults push PrimaLend into over-advance on its CIBC facility. The loan balance exceeds the collateral value supporting it. | | Late 2024 | Roughly $34 million in loan participations sold to cure the shortfall. | | Jan 2025 | Another over-advance, this time on both the CIBC and ANB facilities. | | Feb 2025 | CIBC sends default notices and demands a financial advisor. PrimaLend engages FTI and Spencer Fane. | | Oct 22, 2025 | A facility maturing two days later forces the Chapter 11 filing. | | Feb 20, 2026 | Plan confirmed as a liquidation, with credit-bid sales to CIBC and Amarillo National Bank. | Fourteen months elapsed between the first over-advance and the bankruptcy filing. The lender was on notice for every one of them. **Inference**This is the counter-example in the set, and it deserves to be stated as clearly as the failures. PrimaLend's lenders caught the problem fourteen months before the bankruptcy. The borrowing-base mechanics did exactly what they are designed to do: an over-advance is a mathematical signal that fires the moment collateral value drops below the loan balance, and CIBC acted on it in the same quarter it appeared. Continuous monitoring worked. What it could not do was manufacture credit quality that had already deteriorated. ### On contagion PrimaLend filed weeks after Tricolor, and the two are frequently lumped together. The over-advances predate Tricolor's collapse by more than a year. The honest read is shared macro cause rather than direct contagion. ## Case 3 · America's Car-Mart, where nothing was hidden and it still ended here Car-Mart is a public company. Every number in its deterioration was in an audited SEC filing before Silver Point wrote a check. Rising net charge-offs. An allowance climbing through the 23% to 25% range. Repeated securitizations needed to pay down the revolver. None of it concealed. On October 30, 2025, Car-Mart closed a $300 million, five-year term loan from Silver Point Capital at SOFR plus 7.50%, with warrants for up to 10% of fully diluted shares. Proceeds repaid $162.9 million outstanding on a $350 million ABL revolver from BMO Harris Bank and terminated that facility. ### What that pricing means A lender charging SOFR+7.50% and taking 10% equity warrants is not underwriting a healthy borrower. Reporting since has indicated Silver Point viewed a Car-Mart default as close to inevitable at the time of closing, and structured the covenants and triggers to take control quickly when it happened. Seven and a half months later, Car-Mart was in forbearance on that same agreement. ### The forbearance The June 19, 2026 First Amendment and Limited Waiver covered five simultaneous defaults: minimum liquidity, the Collateral Coverage Ratio, borrowing-base reporting, additional liquidity reporting, and the anticipated inability to deliver an unqualified audit opinion for the fiscal year ended April 30, 2026. The waiver period runs to September 7, 2026, extendable to November 6 if milestones are met. Fees: up to $18 million. ### Where it stands now Car-Mart defaulted, and Silver Point took board representation. Inventory is down 52% year over year. The company is considering asset sales and a possible wind-down, inside or outside bankruptcy. Houlihan Lokey has been seeking at least $500 million in rescue capital. The stock fell 68% on June 10, 2026 to $1.67, the lowest since the company went public in 1992. **Inference**Detection never failed here. A lender priced a distressed borrower accurately and structured to convert to control. The distance from Tricolor is total: there, the lenders were deceived about what they held. Here, the lender knew exactly what it held and charged accordingly. ## What actually went wrong, side by side The same headline, three unrelated mechanisms Where each failure actually originated. Fraud allegations pertain to Tricolor only. Qualitative assessment, not measured data. Based on court filings, SEC disclosures, and contemporaneous reporting. Fraud allegations pertain to Tricolor only and are unproven as to Chu. | | Tricolor | PrimaLend | Car-Mart | |---|---|---|---| | Fraud alleged | Yes, indicted, guilty pleas | No | No | | Reliance on unverified borrower data | High, sole originator and servicer | Moderate, and caught | Low, public audited filer | | Did monitoring catch it | No, seven years | Yes, 14 months ahead | Yes, priced in advance | | Who first detected the problem | Junior analyst at a mezzanine lender | CIBC borrowing-base mechanics | Public filings, read by the lender | | Failure type | Detection | Credit cycle | Liquidity and structure | ## The contagion nobody priced Here is the connection between the three that is easy to miss. Tricolor's collapse did no damage to Car-Mart's loan book. It damaged Car-Mart's access to capital. Reporting indicates that after Tricolor filed, the private credit market repriced the entire subprime auto sector, banks providing working capital pulled back, and Car-Mart's planned financing became substantially more expensive than it would have been three months earlier. The company then ran at that more expensive structure for six months and hit the wall. That is the mechanism this series keeps returning to. Car-Mart's credit book was improving through this entire period. Charge-offs were declining. The newer vintages were cleaner. None of it mattered, because the funding market repriced on someone else's fraud. Car-Mart's credit improved while its funding repriced The two things that were supposed to move together, and did not. The credit-quality arrow is directional only, drawn from management-disclosed charge-off and credit-tier trends. It is not a plotted series and carries no scale. The funding-distress markers are dated events from Car-Mart SEC 8-K filings and Bloomberg reporting. **Inference**No amount of underwriting discipline gets you out of a sector-wide funding repricing. The credit line and the funding line are separate systems, and the funding line responds to events that have nothing to do with your book's performance. Originating better paper does not manage that risk. Knowing, before you need it, what your next-best funding option costs does. ## Why scale did not help Four structural reasons the giants missed what a careful reader caught. | Reason | What it means in practice | |---|---| | Incentives | Volume and fee economics run against verification. A lender that demands more documentation loses deals to one that demands less. Until fraud losses exceed the profit from moving faster, speed wins. In Tricolor's case the banks earned warehouse spread and securitization fees on the same relationship. | | Materiality | Small exposures get generalist attention. Fifth Third's non-depository financial institution lending was roughly 8% of its book. JPMorgan's CFO said plainly that the bank does not typically call out individual borrower exposures "for amounts immaterial." A $200 million line is a rounding error at a trillion-dollar institution, and gets monitored accordingly. | | Silos | Siloed lenders cannot see across facilities. This is the specific structural hole Tricolor exploited. Each lender's mandate ends at its own collateral schedule. Nobody is paid to look at the whole picture, so nobody does. | | Sampling | Periodic sampling misses systematic fraud. Field exams and agreed-upon-procedures reviews test a sample at a point in time. A scheme running continuously across every facility survives a 1% sample almost by definition. | **Inference**Every one of these is a failure of mandate rather than of intelligence or resources. Each institution in the Tricolor chain did roughly what its role required. The gap sat between the roles, and nobody owned it. ## What was not clearly visible in the public record **The double-pledged loan counts.** The figures circulating in coverage, roughly 31,000 double-pledged loans and roughly 6,960 fabricated loans, come from the noteholder complaint's characterization of the Chapter 7 trustee's analysis. That complaint was dismissed. Treat those numbers as alleged rather than established. **The abandoned IPO reporting.** The claim that JPMorgan's equity capital markets team found irregularities during an abandoned Tricolor IPO process, and that the CFO had a prior association with an accounting fraud, is single-sourced to trade reporting and repeated in the dismissed complaint. It is plausible and consistent with the rest of the file, and it is not established fact. **Warehouse field exams.** Whether JPMorgan or Barclays conducted independent field examinations on the warehouse collateral is not disclosed anywhere I can find. The absence of public evidence is not evidence of absence. **The Rakoff dismissal, and how this is framed.** On June 10, 2026, Judge Jed Rakoff dismissed the investor suit against the three banks, finding the plaintiffs had at most alleged negligence rather than the intentional misconduct required for a securities fraud claim. The banks argued the deals were private Rule 144A placements with no underwriter diligence obligation, and that they were themselves fraud victims. Nothing in this brief asserts the banks committed wrongdoing. The argument here is about process gaps, not culpability. ## What this means for your book 1 **Who can see across all your funding facilities at once?** If the answer is only you, then the answer is nobody independent. Tricolor's fraud survived seven years specifically because no party had visibility across facilities. Run the reconciliation yourself before someone else runs it for you, and be able to hand a lender the result. 2 **Does the cash behave the way your tape says it should?** The single observation that unwound Tricolor was that loans reported as current were not paying down principal. That is a comparison anyone can run monthly between the remittance file and the status file. When those two disagree, something is wrong with the data, the servicing, or the loans. 3 **What does your funding cost if your sector has a bad quarter you had nothing to do with?** Car-Mart's book was improving when its funding repriced, and the repricing came from another company's fraud. Know your next-best facility and its cost before the moment you need it, because at that moment it is always more expensive, and sometimes the difference is the whole business. **Sources & notes** **Tricolor.** The $2.2 billion pledged, $1.4 billion real, and approximately $800 million in bogus collateral figures, the double-pledging and collateral-manipulation language, and the 2018 to 2025 duration are from the DOJ indictment unsealed December 17, 2025 (SDNY), as reported by the U.S. Attorney's Office and corroborated by Auto Remarketing, AP, and Business Insider. Kollar and Seibold's December 16, 2025 guilty pleas before Judge Liman are from the DOJ release. Goodgame's June 24, 2026 guilty plea to six counts, his statement to Judge Castel, and the superseding eight-count indictment against Chu including the Continuing Financial Crimes Enterprise charge are from Reuters, Bloomberg, and TT News, June 24, 2026. Chu's June 30 not-guilty plea and the October 19, 2026 trial date are from court transcripts as reported by National Law Review. The Rakoff dismissal is from Reuters, June 10, 2026. The Deloitte agreed-upon-procedures scope and disclaimers are from the SEC Form ABS-15G filing; KBRA methodology is from its presale and surveillance releases. **PrimaLend.** Business model, portfolio size, over-advance timeline, the CIBC and ANB facilities, the FTI and Spencer Fane engagement, the October 22, 2025 Chapter 11 filing, and the February 20, 2026 plan confirmation are from the First Day Declaration and case coverage. **America's Car-Mart.** The Silver Point term loan terms ($300M, SOFR+7.50%, 10% warrants, October 30, 2025, repaying $162.9M on the BMO Harris ABL) are from the company's 8-K and press release. The June 19, 2026 First Amendment and Limited Waiver terms are from the 8-K filed June 22, 2026. The 68% single-day decline to $1.67 on June 10, 2026 is from Bloomberg via TT News. Post-default status, covering Silver Point board representation, inventory down 52%, asset sales under consideration, and Houlihan Lokey seeking $500M, is from Bloomberg and subsequent coverage, July 2026. **Bank loss disclosures.** JPMorgan's approximately $170M Q3 2025 charge-off and the Barnum and Dimon quotes are from the October 14, 2025 earnings calls via Banking Dive and American Banker. Fifth Third's disclosure is from Reuters, October 17, 2025 and its Q3 2025 earnings call. Where this brief reasons beyond what a document literally states, it is labeled as an inference. Figures 1 through 3 are schematic or qualitative and are captioned as such. Figure 4's credit-quality arrow is directional and carries no scale. Allegations in the indictment are unproven as to Chu, who has pleaded not guilty. Nothing here asserts wrongdoing by any lender; the dismissed noteholder suit is cited as context, not as a finding. Point-in-time reading of the public record through August 3, 2026. LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned, and this is not investment, legal, or accounting advice. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). LR LendRisk Analytics Independent market research Continue reading [Issue · 04 · Stress-signal brief What the tape said: *Credit Acceptance.*](https://lendriskanalytics.com/insights/cacc-stress-signals.html) [Issue · 03 · Stress-signal brief What the tape said: *America's Car-Mart.*](https://lendriskanalytics.com/insights/carmart-stress-signals.html) --- title: "What the tape said: Credit Acceptance (CACC)" url: https://lendriskanalytics.com/insights/cacc-stress-signals.html publisher: LendRisk Analytics series: What the Tape Said issue: 4 published: 2026-05-05 kind: Stress signal description: "A stress-signal snapshot on Credit Acceptance: an 8.2-point forecast miss on the 2022 vintage, the worst in a decade, against ABS funding costs that fell from 8.6% to 5.1% over the same two years. Deterioration and stabilization, read from the same tape. Every figure sourced; inferences labeled." html: https://lendriskanalytics.com/insights/cacc-stress-signals.html --- # What the tape said: Credit Acceptance (CACC) [← All insights](https://lendriskanalytics.com/articles.html) What the Tape Said · Issue 4 Stress-signal brief · 14 min read Stress signal · Credit Acceptance (CACC) # What the tape said: Credit Acceptance. Public record through Q1 2026 earnings (May 5, 2026) · LendRisk Analytics Credit Acceptance is not breaking. It is a profitable, cash-generating subprime auto lender that publishes its own forecast misses every quarter, and the recent vintages in that table are telling a coherent story. The tape shows deterioration and stabilization at the same time. This snapshot reads both halves. Every figure is sourced; every forward-looking or interpretive statement is labeled as inference. −8.2% 2022 vintage forecast miss vs. initial, worst in a decade ~$492.6M Cumulative downward revision, Q1 2024, Q1 2026 ~5.1% Nov 2025 ABS cost, down from ~8.6% in Nov 2023 −11.4% Shares outstanding, one year, on active buybacks ## Bottom line Credit Acceptance is not breaking. It is a profitable, cash-generating subprime auto lender that publishes its own forecast misses every quarter, and the recent vintages in that table are telling a coherent story. The 2022 book is CACC's worst forecast miss in a decade at −8.2% versus initial. The 2023 book is −4.2%. The 2024 book is −1.9%. But the quarterly bleed has slowed sharply: the Q1 2026 downward revision of $9.1 million was the smallest in three years, and the 2025 vintage is currently running slightly above plan. Meanwhile funding got cheaper, ABS buyers kept showing up, and the company kept shrinking its share count. This is a book under strain, being actively managed, inside a structure explicitly built to survive strain. The read The single most interesting tension in this file is that CACC's disclosed collateral got worse for three straight years while its cost of funding got better. From late 2023 to late 2025, forecasted net cash flows were cut in all but one quarter, the 2022 vintage alone sliding from a 67.5% projected collection rate to 59.3%. Yet across the same window, the all-in cost of a CACC term securitization fell from roughly 8.6% to roughly 5.1%. The ABS market did not flinch. That divergence is the whole story: the advance-structure model insulates CACC's bondholders and equity in ways a conventional subprime lender's are not, so deteriorating collateral shows up as a smaller spread, not a solvency event. **The company tells you this itself**: its securitizations are structured to withstand a 35% decline in the forecasted collection rate before the most junior bond is at risk. The 2022 vintage missed by 8 points, not 35. That is the counter-argument to the short case, and it deserves to be stated plainly. ## How the model actually works CACC does not simply lend. Dealers assign retail installment contracts to CACC under two programs, and the split between them is the reason CACC can stay profitable at loss rates that would sink a conventional lender. | Program | Portfolio (72.1% of net receivables) | Purchase | |---|---|---| | Structure | CACC advances a set amount to the dealer up front, keeps collections to repay that advance plus its return | CACC buys the loan outright for a one-time payment; higher advance rate | | Dealer holdback | Paid only after CACC has been made whole | None, dealer paid in full at assignment | | First loss absorbed by | The dealer, via forfeited holdback | CACC directly | In both programs, CACC sets the advance so it earns an acceptable return even if collections come in below forecast. That is why its signature disclosure, the forecasted-collection-rate-by-vintage table, is the entire tell. CACC is a company whose disclosed history is a running scorecard of its own forecast accuracy. ## Signals, ranked by actionability ### Signal 1 · The 2022 vintage is CACC's worst forecast miss in a decade, and it is now baked in As of December 31, 2025, the 2022 assignment year is forecast to collect 59.3% of contractual amounts owed, against an initial forecast of 67.5%, a −8.2% variance. On dealer loans specifically the miss is −8.8% (58.5% vs. 67.3%). The 2022 vintage is 81.7% realized, so the miss is largely locked in. **Inference**With 81.7% of 2022 collections already in, further downside on that specific vintage is limited, the damage is done and provisioned. The risk that rhymes is not 2022 itself but whether 2023 (−4.2%, only 65.3% realized) and 2024 (−1.9%, only 43.5% realized) drift further as they season. ### Signal 2 · The quarterly forecast bleed has slowed to a three-year low Downward revisions to forecasted net cash flows have shrunk almost every quarter since mid-2024, from $189.3M in Q2 2024 to $9.1M in Q1 2026, the smallest quarterly change in three years. CEO Vinayak Hegde described the Q1 2026 result as reflecting "reduced volatility in loan forecast changes and moderation in unit volume declines." Downward revision to forecasted net cash flows, by quarter The size of each quarter's cut to CACC's own collection forecast. The bleed is not gone, but it has shrunk by 95% from its 2024 peak. Source: CACC quarterly earnings releases, forecast-change tables, Q2 2024 through Q1 2026. **Inference**The trajectory is consistent with stabilization, not acceleration. But management attributes part of the residual drag to slower cash-flow timing, consumers prepaying less and staying in loans longer, rather than pure credit loss, which is a softer problem than rising defaults. ### Signal 3 · Funding got cheaper even as collateral deteriorated The expected all-in annualized cost of CACC term ABS fell from roughly 8.6% in November 2023 to roughly 5.1% in November 2025, ticking up only slightly to ~5.2% by the May 2026 deal. In February 2025 CACC also priced $500M of 6.625% senior notes due 2030 to refinance $400M of 6.625% notes due 2026. Expected all-in cost, CACC term ABS financings Cost of funds fell by roughly a third over two years, the clearest market signal that ABS buyers do not see the collateral deterioration as structurally threatening. Source: CACC 8-K financing announcements, Nov 2023, May 2026. **Inference**Most of the improvement tracks the rate environment and tighter ABS spreads sector-wide, not a CACC-specific vote of confidence. But the continued, oversubscribed access on cheaper terms is the clearest market signal that sophisticated ABS buyers do not see CACC's collateral deterioration as structurally threatening. ### Signal 4 · The company is shrinking its equity base while revisions accumulate Shares outstanding fell from 12,048,151 to 10,680,143 over 2025, a decline of 11.4%. GAAP average shareholders' equity fell 9.8% year over year in Q4 2025, and retained earnings dropped from $1,414.7M to $1,119.2M. Buybacks continued into 2026: roughly 329,000 shares in Q1 2025, 530,000 in Q2 2025, 425,000 ($191.4M) in Q4 2025, and 365,258 ($178.9M) in Q1 2026. Shares outstanding, year over year Buybacks pulled the share count down 11.4% in a single year, alongside a 9.8% drop in average GAAP equity. Source: FY2025 10-K balance sheet. Bar length scaled to share count; −11.4% year over year. **Inference**CACC is returning capital aggressively into a period of elevated forecast uncertainty. This is either confidence, management believes intrinsic value exceeds price and the model absorbs losses, or a thinner cushion, less book equity beneath a strained loan book. The company's own structural disclosure, a 35% collection-rate decline tolerance before the most junior ABS bond is at risk, argues for the former, but the shrinking equity is a real trade-off to track. ### Signal 5 · Volume is contracting but the dealer network is at a record, a mix story Consumer Loan unit volume declines are moderating: −16.5% in Q3 2025, −9.1% in Q4 2025, −4.3% in Q1 2026. Yet CACC reported a record 10,977 active dealers in Q1 2026. Initial spread on new assignments held roughly flat, 22.1% in 2024 versus ~22.0% in 2025, so pricing discipline, not spread collapse, is driving the volume loss. **Inference**CACC is deliberately ceding volume to protect economics, management framed it as prioritizing economic profit over volume. More dealers writing fewer loans each suggests competitive pressure on per-dealer economics, but the flat initial spread says CACC is not buying growth by underpricing new paper. ## The securitization record Credit-enhancement context from the rating agencies backs up the "structurally protected" read. KBRA assigned CAALT 2024-3 ($500M) initial credit enhancement of 59.41% on the Class A notes, 43.20% on Class B, and 21.62% on Class C, and later affirmed those ratings through the August 2025 distribution date. S&P raised four ratings and affirmed two across two CAALT transactions on May 29, 2024, rating actions, not downgrades. | Deal | Size | Class A coupon | Expected all-in cost | |---|---|---|---| | Nov 2023 | $200.0M | 6.98% | ~8.6% | | Dec 2023 | $294.0M | — | ~7.0% | | Feb 2024 (3-class) | — | 6.95% | ~7.8% | | Dec 2024 | $300.0M | 5.79% | ~6.3% | | CAALT 2025-1 (Mar 2025) | ~$400.0M | 5.02% | — | | CAALT 2025-2 (Nov 2025) | $500.0M | 4.50% | ~5.1% | | CAALT 2026-1 (May 2026) | $450.0M | 4.65% | ~5.2% | Class A coupon fell by roughly 235 basis points from the November 2023 deal to the November 2025 deal, even as the forecast tables above were being revised down almost every quarter over that same window. **Inference**Exact S&P expected-cumulative-net-loss figures for the 2024-2026 CAALT deals were not confirmable from primary rating-agency documents for this brief. Directionally, deal-over-deal senior enhancement stayed high (~59% Class A on 2024-3) and rating actions were affirmations or upgrades, not downgrades, consistent with a market that views the structures as amply protected against the disclosed collateral softness. ## The vintage table: a smile, then a dip, then a recovery CACC's signature disclosure is the forecasted collection rate by assignment year, restated every quarter against the initial forecast made at origination. Ten years of that table tell a shape: the pandemic-era 2019-2020 books beat plan by 3 to 5 points, stimulus and used-car price spikes did the work. The 2022-2024 books, originated into peak used-car prices and then inflation-squeezed borrowers, missed. The 2025 book, priced after CACC's Q3 2024 scorecard change, is so far tracking to plan. Variance vs. initial forecast, by assignment year Ahead of plan through the pandemic vintages, a sharp miss centered on 2022, and an early recovery in 2025. Source: CACC Q4 2025 forecasted-collection-rate-by-vintage table, as of December 31, 2025. | Vintage | Current forecast | Initial forecast | Variance | % realized | |---|---|---|---|---| | 2016 | 63.9% | 65.4% | −1.5% | 99.7% | | 2017 | 64.8% | 64.0% | +0.8% | 99.6% | | 2018 | 65.5% | 63.6% | +1.9% | 99.3% | | 2019 | 67.2% | 64.0% | +3.2% | 98.6% | | 2020 | 68.0% | 63.4% | +4.6% | 96.9% | | 2021 | 63.8% | 66.3% | −2.5% | 92.5% | | 2022 | 59.3% | 67.5% | −8.2% | 81.7% | | 2023 | 63.3% | 67.5% | −4.2% | 65.3% | | 2024 | 65.3% | 67.2% | −1.9% | 43.5% | | 2025 | 67.2% | 67.0% | +0.2% | 15.2% | Forecasted collection rate by assignment year, as of December 31, 2025. Q1 2026 update: 2023 slipped further to 63.1% (−4.4% vs. initial); 2021 improved to 64.0%; 2025 held at 67.2% (+0.2%); the 2026 vintage opened at 66.3% vs. 66.6% initial (−0.3%), which the company attributes primarily to loan cancellations rather than performance. **Inference**If the 2025 and 2026 vintages hold to plan, the miss was a definable 2022-2024 cohort problem, tied to peak used-car prices and an inflation-squeezed borrower, not a permanent break in the underwriting model. ## Provision mechanics: why GAAP earnings swing and adjusted don't Under CECL, CACC recognizes a large provision at loan assignment for cash it never expected to collect, then books above-normal finance-charge revenue over the life of the loan. Forecast changes flow straight through the provision line in the quarter they occur. In FY2025 that line was $338.3M, down from $493.8M in FY2024; provision on new assignments was $277.8M. The mechanical result is that GAAP net income swings hard with revisions, Q2 2024 printed a −$47.1M GAAP net loss against +$127.2M adjusted net income the very same quarter, while CACC's non-GAAP "floating yield" adjusted figures smooth it. Neither number is truth: the GAAP number over-punishes the revision quarter, and the adjusted number relies on management's own expected-cash-flow assumptions. ## What the market is, and is not, pricing Pricing in Sector-wide subprime stress. Fitch's 60+-day subprime auto ABS delinquency rate hit a record 6.74% in December 2025 and 6.90% in January 2026, a 32-year high. Tricolor Holdings filed Chapter 7 on September 10, 2025 amid an alleged ~$800M double-pledged-collateral fraud; JPMorgan and Fifth Third disclosed $340M+ in combined losses. America's Car-Mart shares hit ~$11 in mid-May 2026, a low not seen since 2008-2009. Analysts on CACC are cautious-to-neutral: consensus Hold, price targets clustered $450,$575, short interest around 11% of the float. Not fully pricing *(inference)* The stabilization in the forecast-change table and the near-flat initial spread on new vintages. The market is reading CACC through the sector delinquency headline; the company's own tape shows the bleed decelerating and 2025 vintage economics intact. Citron Research, a longtime bear, reversed to bullish on March 4, 2026, assigning a $714 fair value and writing: "We were among those who believed the regulatory scrutiny posed existential risk. We were wrong." A CEO transition (Vinayak Hegde, ex-Amazon and T-Mobile, replaced Ken Booth effective November 13, 2025) and an April 2026 workforce cut of ~151 employees signal a cost-and-tech reset not yet in the numbers. ## Limits: what the public record does not show **Exact ECNL for recent CAALT deals.** S&P's expected-cumulative-net-loss assumptions for the 2024, 2025, and 2026 CAALT securitizations could not be confirmed from primary rating-agency documents for this brief. The direction, whether ECNL rose with 2022-2024 collateral stress, is therefore inference, not fact. **Loan-level ABS-EE data** exists but was not parsed here; deal-level static-pool curves would sharpen the vintage read. **The GAAP-vs-adjusted gap is real and mechanical** under CECL; the "right" earnings number is a judgment call, not a disclosed fact. **The settlement is not final.** CACC recognized cumulative contingent losses of $82.6M through Q4 2025 reflecting preliminary alignment on a potential $75.5M cash payment in the multi-state and NY AG matters, but preliminary alignment is not a signed agreement. **Timing vs. credit.** Management attributes much of the residual forecast drag to slower prepayments (timing), not higher defaults (credit). The public tables do not cleanly separate the two. ## Legal & regulatory status, as of mid-2026 **CFPB / NY AG** (filed Jan 4, 2023): the CFPB filed to withdraw April 24, 2025; the court granted withdrawal April 29, 2025. The New York Attorney General remains the sole plaintiff, and the case is now limited to New York consumers. CACC's motion to dismiss remains pending. **Multi-state + NY AG potential settlement:** cumulative contingent losses of $82.6M booked through Q4 2025; a potential cash payment of $75.5M, not yet finalized. **Massachusetts AG (2021):** $27.2M settlement, at the time the largest of its kind. **Mississippi AG (2021):** $325K to the state plus a $125K charitable donation. ## Three questions for operators running similar books 1 **Does your disclosure let the market watch you miss?** CACC publishes forecasted-vs-initial collection rates by vintage every quarter. That transparency is why an 8-point miss on 2022 was a spread story, not a run. If your book's misses are invisible until they are realized losses, you are trading short-term optics for tail risk. 2 **Is your loss-absorption structural or just capital?** CACC's advance/holdback model and 35%-collection-decline ABS cushion mean high loss rates compress profit rather than threaten solvency. If your model relies on equity to absorb the same shock, run the sensitivity at a 2022-style miss and see what survives. 3 **When collateral weakens, does your funding cost confirm or contradict it?** CACC's ABS cost fell while its collateral deteriorated. If your funding market is repricing you wider while your book softens, you have a very different problem than CACC does, that is the sequence that ended Tricolor and pressured others. **Sources & notes** Primary: CACC FY2025 Form 10-K; Q4 2025 and Q1 2026 earnings releases (GlobeNewswire / SEC 8-K); Q3 2025 and Q2 2025 10-Qs; CACC 8-K financing announcements (Nov 2023, Dec 2023, Feb 2024, Dec 2024, Nov 2025, May 2026); CACC senior notes 8-K (Feb 2025); FY2025 Annual Report. Rating agencies: S&P Global Ratings CAALT presales; KBRA CAALT press releases; Moody's. Regulatory: CFPB enforcement page; Mass.gov AG release; CACC litigation releases. Sector: Fitch subprime auto delinquency data via Wolf Street, Marketplace, Auto Finance News, Motley Fool; Tricolor bankruptcy via Bloomberg and court filings. Market: stockanalysis.com, GuruFocus, TD Cowen / Stephens via TipRanks, Citron Research (Mar 4, 2026). All forward-looking and interpretive statements are labeled "Inference." Figures not confirmable from a named source were excluded or flagged in Limits. This is a snapshot of a functioning, profitable company under sector stress, not a prediction of failure. Point-in-time reading of the public record through Q1 2026 earnings (May 5, 2026). LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned, and this is not investment, legal, or accounting advice. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). LR LendRisk Analytics Independent market research Continue reading [Issue · 03 · Stress-signal brief What the tape said: *America's Car-Mart.*](https://lendriskanalytics.com/insights/carmart-stress-signals.html) [Issue · 02 · Stress signal What the tape said: *CarMax.*](https://lendriskanalytics.com/insights/carmax-stress-signals.html) --- title: "What the tape said: America's Car-Mart" url: https://lendriskanalytics.com/insights/carmart-stress-signals.html publisher: LendRisk Analytics series: What the Tape Said issue: 3 published: 2026-07-03 kind: Stress signal description: "A read of what is publicly visible about America's Car-Mart right now: a $300M distressed-fund term loan, a June 2026 forbearance covering five simultaneous covenant defaults, an $18M waiver fee, and 66 days on the clock. The credit book was improving; the funding architecture is what broke. Straight from the filings." html: https://lendriskanalytics.com/insights/carmart-stress-signals.html --- # What the tape said: America's Car-Mart [← All insights](https://lendriskanalytics.com/articles.html) What the Tape Said · Issue 3 Stress-signal brief · 12 min read Stress-signal brief · America's Car-Mart # What the tape said: America's Car-Mart. Public record through July 3, 2026 · LendRisk Analytics Issue 1 was a company that had already failed. Issue 2 was a profitable one carrying a stressed vintage. This one sits between them, and it is the most instructive of the three. America's Car-Mart is still open, still originating, still issuing bonds. It is also *66 days* from a covenant-waiver cliff, paying up to $18 million for the extension, with its board running a strategic-alternatives process that has no disclosed outcome. Here is the part worth sitting with: the credit book is not what broke. The funding did. $18.0M Fees for the June 2026 covenant waiver 66 days Left on the waiver clock, to Sep 7, 2026 SOFR + 7.50% Rate on the $300M Silver Point term loan 136 → 94 Active dealerships, Jan to Apr 2026 ## Bottom line Let me say the counterintuitive part first, because it is the whole point. Car-Mart's credit book has been getting *better*. Charge-offs are declining. The top credit tier is now 66.7% of receivables. Collections are up year over year. The LOS V2 underwriting platform management has talked about for six quarters is producing cleaner paper, and the numbers back it up. None of that is in question. What broke is the funding architecture. In October 2025 Car-Mart replaced a multi-lender revolving line with a single $300 million term loan from Silver Point Capital, a fund that specializes in distressed situations, at SOFR plus 7.50% with 10% warrant coverage. Seven and a half months after that closed, the company was in forbearance on the same agreement, paying $18 million for a stay, anticipating it cannot deliver a clean audit, and running a board-level review that the company itself says may include restructuring. The read In subprime auto, the funding line kills faster than the credit line. That is the pattern at American Car Center, at U.S. Auto Sales, at Tricolor. Car-Mart has **not** lost its funding. But it has lost its funding **flexibility**, and the credit book ran out of runway to season into its own improvement before the structure was tested. The book was getting better. The structure was not. ## The sequence is the thing No single event here would end a company. Read in order, they describe one. The stretch from Jan to Jun 2026 is where the events stop being spaced out and start stacking. Eight months, one direction Each marker was disclosed publicly and, on its own, was manageable. Placed on a real timeline, they cluster into the first half of 2026 and run straight at the September 7 waiver cliff. Source: Car-Mart SEC 8-K filings and earnings releases, Oct 2025 through Jun 2026. Markers placed on a true time axis. ## The five signals Ranked the way I would read them as a counterparty: most actionable first. Where I reason past what the filing literally says, it is labeled an inference. ### Signal 1. Five defaults waived at once On June 19, 2026, Car-Mart signed the First Amendment and Limited Waiver to its Credit and Guaranty Agreement with Silver Point Finance. The waiver covers **five** actual or anticipated defaults simultaneously: a minimum-liquidity failure, a Collateral Coverage Ratio (CCR) failure, a borrowing-base reporting failure, an additional liquidity-reporting failure, and the one that does not show up in ordinary covenant stress, an **anticipated inability to deliver an unqualified audit opinion** for the fiscal year ended April 30, 2026. That last item is the tell, because of timing. A company does not learn it cannot deliver a clean audit the week before the waiver. The auditor says so weeks or months ahead. So management knew before the fiscal year even closed that going-concern language was coming, and built the forbearance around that knowledge. | The waiver, in one look | Term | |---|---| | Fees to agent and lenders | up to $18.0M | | Waiver period (Specified Period) | through Sep 7, 2026 | | Possible extensions if milestones met | Sep 21 or Nov 6 | | Minimum weekly liquidity | $7M Fri / $5M else | | Minimum CCR (steps down Jul 1) | 1.25 then 1.20 | | Board strategic review may include | restructuring | **Inference**A company paying $18 million for 66 remaining days of lender patience, while its board runs a strategic-alternatives process with no guaranteed outcome, has moved past the ordinary toolkit. What the outcome is, recapitalization, sale, or restructuring, is not visible in the public record. That it is uncertain is the company's own disclosure. ### Signal 2. The Silver Point refinancing was the first signal On October 30, 2025, Car-Mart closed a $300 million, five-year term loan with Silver Point at SOFR plus 7.50%, maturing October 30, 2030, with warrants for up to 10% of fully diluted shares. The proceeds fully repaid the revolving line. Silver Point specializes in special situations and distressed credit. Equity warrants on top of a SOFR+7.50% coupon is not how a lender prices a borrower it sees as investment grade. At roughly 4.3% SOFR when it closed, the all-in cost was about 11.8% on $300 million of first-lien secured debt. Here is the comparison that lands it. Two months earlier, Car-Mart's ACM Auto Trust 2025-3 priced at a 5.46% weighted-average coupon. So the company was funding its ABS investors at 5.46% while accepting term debt at about 11.8% at the same time. That gap is an operator whose primary liquidity source was priced as distressed, regardless of what the ABS market thought of the collateral. Two prices for the same book, same season In late 2025 the ABS market funded Car-Mart's collateral at 5.46%. Its primary term lender priced the same company at roughly 11.8% all-in. When those two numbers diverge this far, the market is telling you the risk sits in the borrower, not the paper. Source: ACM Auto Trust 2025-3 8-K (Aug 29, 2025); Silver Point term loan press release (Oct 30, 2025). All-in cost assumes ~4.3% SOFR at closing plus the 7.50% margin. **Inference**The June 2026 forbearance was visible in the October 2025 refinancing. Not the specific covenant failures, those needed more deterioration to surface. But the architecture was built under stress, and structures built under stress tend to reveal it when the first covenant is tested. Time from closing to forbearance: seven and a half months. ### Signal 3. The allowance and the DTA write-down say the same thing twice At January 31, 2026, the allowance for credit losses was $347.6 million, or 25.53% of the finance receivable principal balance. Net charge-offs that quarter were 6.5% of average receivables, down from 6.8% a year earlier. So the allowance is about 3.9 times the quarterly charge-off run rate. That is not an under-reserved book. It is a book reserved for a future that management has marked as worse than the present. Allowance for credit losses, % of receivables The reserve dipped through mid-2025 as recent vintages performed better, then climbed to a five-quarter high into January 2026, even as actual charge-offs fell. A book reserving up while losses come down is reserving for something it sees ahead. Source: Car-Mart quarterly earnings releases, Q2 FY2025 through Q3 FY2026. The $47.0 million non-cash valuation allowance recorded against the deferred tax assets of the finance subsidiary in Q3 FY2026 is the same story in a different accounting language. A DTA valuation allowance means management and auditors have concluded it is more likely than not the benefit will not be realized, because there will not be enough future taxable income to absorb it. It is not a cash charge. It is a statement that the subsidiary is not expected to be profitable enough, for the foreseeable future, to use the deduction. **Inference**The allowance and the DTA write-down are not two signals. They are one assessment in two languages, three months apart. The allowance says: we expect more losses on the book we have. The DTA write-down says: we do not expect to be profitable enough to recover taxes we already paid. Read together, they describe a finance subsidiary not expected to generate taxable income in the near term, and that gap is what the strategic review is trying to close. The write-down in March told you the audit qualification was coming; the qualification became a covenant default in June. ### Signal 4. In a term-loan structure, an origination slowdown feeds itself In Q3 FY2026, Car-Mart sold 10,275 retail units, down 22.1% year over year. Management named two causes: origination-capacity constraints from the capital-structure transition, and Winter Storm Fern disrupting late-January operations. The storm was real but temporary, 30+ DPD moved from 3.14% at October 31 to 4.4% at January 31, then normalized back toward 3.7 to 3.8% by mid-February. The storm was noise. The origination constraint is structural. A revolving line lets you draw as you originate and repay as collections arrive. A term loan does not revolve. Car-Mart's $300 million was fully drawn at closing, so every new loan after that had to be funded from collections, ABS proceeds, or cash on hand. That is a tighter constraint at every point in the cycle, and it squeezes originations exactly when holding volume matters most. The store-count collapse poured fuel on it. Active dealerships A 31% cut to the retail footprint in under three months, at the same moment origination capacity was already pinched by a non-revolving loan. Fewer stores, fewer originations, smaller book, fewer collections, less cash to originate. The loop tightens on itself. Source: Q3 FY2026 earnings release (Mar 12, 2026) for the 136 count; company press release (Apr 14, 2026) for 94. Prior-year count ~154. **Inference**The mechanism is simple and self-reinforcing. Fewer stores, fewer originations. Fewer originations, a smaller portfolio. A smaller portfolio, fewer collections. Fewer collections, less cash to fund new loans under a non-revolving facility. Management has said plainly that securing a revolving warehouse line is the single near-term priority. The forbearance gives them 66 days to get it. ### Signal 5. The personnel moves are the standard pre-restructuring pattern Jeff Williams ran Car-Mart as CEO for roughly eighteen years. Doug Campbell took over on October 1, 2023, as the company was launching its first ABS program, a real strategic shift for a firm that had historically held every loan on balance sheet. On June 23, 2026, four days after the forbearance was signed, two new independent directors joined the board and a reconstituted special committee. The same day, CFO Jonathan Collins announced his resignation effective July 31, with Marie Persichetti succeeding him August 1. Collins had led the ABS program from its start. The forbearance filings also disclosed cash retention awards, $1.2 million for the CEO and $563,000 for the CFO, contingent on staying through a defined date. **Inference**Cash retention for senior executives during a forbearance and strategic review is not a confidence signal. It is an acknowledgment that the people who understand the book and the structure need a financial reason to stay through an uncertain outcome. Board reconstitution, added independent directors, a CFO departure timed to the forbearance, retention for those who remain, these are the standard moves in a pre-restructuring context. They do not prove a restructuring is coming. They are consistent with a board preparing for one as a possibility. ## The ACM Auto Trust record Car-Mart entered the ABS market in April 2022 after decades of holding all paper on balance sheet. The program was a genuine achievement: the company cut its weighted-average coupon by 308 basis points across four straight deals from late 2024 into mid-2025. Then December 2025 reversed the trend. ACM Auto Trust weighted-average coupon Four consecutive deals of falling cost, real execution, bottoming at 5.46% in August 2025. Then the December deal reversed 156 basis points higher on a smaller size. The market saw something in the collateral, and the forbearance came six months later. Source: respective ACM Auto Trust SEC 8-K filings and GlobeNewswire releases. 2024-1 WA coupon (9.44%) implied from the 2024-2 release citing a 198bp improvement on 7.44%. | Deal | Completed | WA coupon | Size | |---|---|---|---| | ACM Auto Trust 2022-1 | Apr 27, 2022 | n/d | ~$400M | | ACM Auto Trust 2024-1 | Jan 31, 2024 | ~9.44% | n/d | | ACM Auto Trust 2024-2 | Oct 9, 2024 | 7.44% | $300M | | ACM Auto Trust 2025-1 | Feb 3, 2025 | 6.49% | $200M | | ACM Auto Trust 2025-2 | May 29, 2025 | 6.27% | $216M | | ACM Auto Trust 2025-3 | Aug 29, 2025 | 5.46% | $172M | | ACM Auto Trust 2025-4 | Dec 17, 2025 | 7.02% | $161.3M | The 2025-4 deal was smaller and 156 bps more expensive than 2025-3. Management described it as a more efficient residual cash-flow structure, which is accurate and does generate more near-term cash. What that does not explain is why the coupon reversed so sharply while the structure was supposedly adding value. ## What was not clearly visible in the public record The going-concern audit language is anticipated, not yet filed. The FY2026 10-K audit report is not public as of today. The June 22 8-K discloses that Car-Mart *anticipates* failing to deliver an unqualified opinion. What the audit actually says lands in the 10-K, due around September. If that filing comes before September 7, its language will matter for whether the waiver converts to a permanent amendment or terminates. The prior revolving-line margin is not in a public filing. The term-loan-versus-revolver cost comparison rests on the disclosed SOFR+7.50% against market reference for comparable facilities. An earlier draft of this analysis cited SOFR+3.50% as the prior margin; that figure is not confirmed in any SEC filing and has been removed. The structural argument, a term loan replacing a revolving facility at materially higher cost, does not need the exact prior margin to hold. The loan-level ABS vintage curves are not in this brief. The ACM Auto Trust ABS-EE filings on EDGAR carry loan-level monthly performance back to 2022. Building the curves, showing whether 2022 originations perform differently than 2024 originations at the same seasoning, means parsing Exhibit 102 XML across many distribution periods. This brief does not do that. Management's own disclosures on improving recent-vintage quality are cited where they appear, but the independent extraction is the logical next step for anyone who needs deal-level precision. ## What this means for your book Car-Mart is not a pure BHPH operator the way most readers of this series are. It securitizes, reports to public investors, and runs at a scale most dealers never reach. But the mechanics transfer cleanly to a smaller book. Three questions worth asking of your own operation this week. **1. What is your cost of funds, and what does your margin look like if it rises 200 basis points?** Car-Mart's move from a revolver to a Silver Point term loan pushed its funding cost up to roughly its own ABS coupon on a blended basis. If your primary facility is a revolving line, know what the next-best alternative looks like before you need it. At the moment of need it is always more expensive than the current facility, and sometimes the difference is your entire margin. **2. What does your portfolio look like if originations stop for sixty days?** A BHPH book that stops originating immediately starts amortizing. Car-Mart's constraint is structural because the term loan does not revolve, but any operator negotiating a covenant breach or facility amendment is in a version of the same spot. Model 90 days of flat originations before the scenario is real, not while it is happening. **3. If your auditor read your finance subsidiary's tax position today, what would they find?** A DTA valuation allowance gets recorded when the benefit more likely than not will not be realized. If your finance sub has run cumulative GAAP losses for two or more years, that conversation with your auditor is closer than it feels. Car-Mart's forbearance covers an *anticipated* qualification, meaning management knew the outcome before the year closed. That knowledge lived in the building for weeks before it hit a public filing. None of this is a prediction that Car-Mart fails. It is still open, still originating, and the credit book is genuinely improving. It is a reminder that in this business the structure can run out of road before the book does, and that the place to catch it is your own funding stack, read the same way, before someone else reads it for you. **Sources & notes** Forbearance terms (up to $18M fees, CCR and minimum-liquidity failures, anticipated audit qualification, waiver through September 7, 2026 with extensions to November 6, new CCR and liquidity thresholds, strategic review) are from the SEC 8-K filed June 22, 2026 (EDGAR accession 000117184326004311), full amendment at [sec.gov](https://www.sec.gov/Archives/edgar/data/0000799850/000117184326004311/exh_101.htm); the initial forbearance and retention awards are from the 8-Ks filed June 5, 2026. The Silver Point term loan ($300M, SOFR+7.50%, 10% warrants, matures Oct 30, 2030, repaid the revolving line) is from the Car-Mart press release October 30, 2025 and the Q2 FY2026 release. Q3 FY2026 results (net loss $76.7M, NCO 6.5%, allowance 25.53%, 30+ DPD 4.4% at Jan 31, 2026, retail units 10,275 down 22.1%, 136 dealerships, $47M DTA write-down) are from the Q3 FY2026 earnings release March 12, 2026. Quarterly allowance figures are from the respective Car-Mart earnings releases, Q2 FY2025 through Q3 FY2026. ACM Auto Trust deal data (coupons, sizes, dates) are from the respective SEC 8-K filings and GlobeNewswire releases; the 2024-1 WA coupon is implied from the 2024-2 release citing a 198bp improvement on 7.44%. Management changes (Campbell succession Oct 1, 2023; Collins resignation effective Jul 31, 2026; Persichetti effective Aug 1, 2026; Nathan and Wartell added Jun 23, 2026; retention awards $1.2M and $563K) are from the respective 8-Ks. Store counts (136 at Jan 31, 2026; 94 by Apr 14, 2026) are from the Q3 FY2026 release and the April 14, 2026 press release. Figures from earlier drafts that could not be confirmed in a primary filing (a SOFR+3.50% prior margin, a 22.1% allowance, a 10.9% 30+ DPD, a 4.5% operating margin) have been removed. **Where this brief reasons beyond what a filing literally states, it is labeled as an inference.** Point-in-time reading of the public record through July 3, 2026. LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned, and this is not investment, legal, or accounting advice. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). LR LendRisk Analytics Independent market research Continue reading [What the Tape Said · Issue 2 What the tape said: *CarMax*](https://lendriskanalytics.com/insights/carmax-stress-signals.html) [What the Tape Said · Issue 1 What the tape said: *Tricolor Holdings*](https://lendriskanalytics.com/insights/tricolor-tape.html) --- title: "What the tape said: CarMax Auto Finance" url: https://lendriskanalytics.com/insights/carmax-stress-signals.html publisher: LendRisk Analytics series: What the Tape Said issue: 2 published: 2026-06-17 kind: Stress signal description: "A read of what is publicly visible about CarMax Auto Finance right now: a $71.3M lifetime-loss revision on 2022 and 2023 vintages, an allowance that climbed after management called the peak, and a nonprime shelf whose structure is quietly tightening. Straight from the filings, with every figure sourced." html: https://lendriskanalytics.com/insights/carmax-stress-signals.html --- # What the tape said: CarMax Auto Finance [← All insights](https://lendriskanalytics.com/articles.html) What the Tape Said · Issue 2 Stress-signal brief · 11 min read Stress-signal brief · CarMax Auto Finance # What the tape said: CarMax. Public record through June 17, 2026 · LendRisk Analytics Last issue I pulled apart a company that had already failed. This one is different, and I think it is the more useful exercise. CarMax is open, profitable, and still selling bonds every quarter. So the question is not what went wrong. It is quieter than that: *what is the public record already telling us, before anyone calls it a problem?* I went and read the filings so you do not have to, and I want you to be able to see what I saw. $71.3M Lifetime-loss revision on 2022 and 2023 vintages, in one quarter 2.61 → 3.02% CAF allowance, Feb to Aug 2025 603 → 613 WA FICO, first to latest nonprime deal 9.85 → 8.42% Excess spread, consecutive nonprime deals ## Bottom line Here is the thing I want you to take away before any of the detail: the CarMax prime book is not broken. If you came here for a collapse story, this is not one. What the filings show is narrower and, honestly, more relevant to your own book. The stress is sitting in two specific places, and both of them are things you can go check in your own portfolio this week. First, the loans CarMax wrote in 2022 and 2023, when used cars were near their peak price, are aging worse than the company expected, and management said so on the record. Second, CarMax has been deliberately walking down the credit spectrum since 2024, and the *structure* of its newer nonprime deals, more cushion on every tranche while the built-in margin shrinks, tells you what the rating agencies quietly think is coming. None of that is a siren. It is drift, and the mechanism behind it, peak-price collateral meeting a used-car market that corrected, is not unique to CarMax at all. If you originated used-vehicle paper in 2022 and 2023, it is your story too. The read The single most interesting thing in this whole file is not a number. It is a sequence. In June 2025 management told the market the worst of the provisioning was behind them. **One quarter later they added $71.3 million in lifetime losses on the exact vintages they said were handled.** That gap is twelve weeks, and it is the cleanest reminder I can give you that even a team with full access to its own tape could not see the seasoning curve coming. If they could not, you should assume you cannot either, and reserve like it. ## The four signals I ranked these the way I would if I were reading CarMax as a counterparty: most actionable first. Where I am reasoning past what the filing literally says, I label it as an inference, so you always know which is which. ### Signal 1. Management called the peak, then missed it On the Q1 FY2026 earnings call in June 2025, the head of CarMax Auto Finance told analysts that Q1 would be the **"high watermark"** for provisioning, that the adjustment on the older vintages had been made, and that they felt good about the reserve. That phrasing is straight from the call. Three months later, the Q2 FY2026 release on September 25, 2025 disclosed a **$71.3 million increase in lifetime-loss estimates,** pinned specifically to the 2022 and 2023 vintages. The allowance went from $474.2 million (2.76% of auto loans held for investment) at the end of May to $507.3 million (3.02%) at the end of August. I am not putting words in anyone's mouth here. The "high watermark" line is from a public transcript. The $71.3 million is from a public SEC filing. The market noticed the gap: the stock fell roughly 20% on the Q2 print, and by early November a securities class action had been filed over the provision miss. I am citing that as context, not as a verdict. CarMax also said in the same release that these vintages "remain highly profitable," which is true and which matters. This is a reserve-adequacy story, not a solvency one. **Inference**A lender that tightened underwriting in April 2024 because the older vintages were already turning, then told the market in June 2025 the reserve was set, then revised lifetime losses on those same vintages in September, is telling you the original number was light. Not dishonest. Light. The honest takeaway is humility about how late the loss curve on 2022 and 2023 paper actually shows up. ### Signal 2. Watch the allowance arc, not the single quarter Any one quarter's allowance number is noise. The shape across six quarters is the signal, and the shape is what I want you to look at here. It climbed into the August 2025 spike, eased, and is climbing again, but for a different reason the second time. CAF allowance, % of auto loans held for investment Management's "high watermark" call landed at the May 2025 reading. The line went up anyway, then turned over, and is now rising again on new lower-tier growth rather than the old vintages. Source: CarMax quarterly earnings releases (8-Ks on SEC EDGAR), Q4 FY2025 through Q1 FY2027. | Quarter end | Allowance | % of HFI | Source release | |---|---|---|---| | Feb 28, 2025 | $458.7M | 2.61% | Q4 FY2025 | | May 31, 2025 | $474.2M | 2.76% | Q1 FY2026 | | Aug 31, 2025 | $507.3M | 3.02% | Q2 FY2026 | | Nov 30, 2025 | $474.8M | 2.87% | Q3 FY2026 | | Feb 28, 2026 | $453.0M | 2.78% | Q4 FY2026 | | May 31, 2026 | $475.0M | 2.95% | Q1 FY2027 | One honest caveat: the Feb 2025 figure was reported against "ending managed receivables," and from Q1 FY2026 on, the denominator is "auto loans held for investment," because some nonprime loans moved to held-for-sale. So do not read this as one perfectly clean line. The direction is real; the exact slope is not apples-to-apples end to end. The second climb, from 2.78% back up to 2.95% in early 2026, is the more forward-looking part. By the company's own Q1 FY2027 disclosure, that move is being driven by growth into the lower-credit Tier 2 space, not by the old 2022 to 2023 vintages. In plain terms: the first hump was the past catching up. The second is the future being priced in as CarMax writes more thin-credit paper on purpose. ### Signal 3. The nonprime shelf is telling on itself In June 2024 CarMax did something it had never done in 76 prior prime securitizations: it launched a dedicated nonprime shelf, the CarMax Select Receivables Trust. The first deal, CMXS 2024-A, was $666.6 million of loans to borrowers with a weighted-average FICO of 603, an average APR of 16.06%, and 31% of the pool in older, higher-mileage "ValuMax" cars. That is a real subprime book, and ring-fencing it protects the prime investors. Good structural hygiene. But it also creates a new surface to watch, and the way that surface has evolved across deals is the signal. Where the cushion is going on the nonprime shelf Between two consecutive deals, the arranger demanded more hard credit enhancement on the senior class while the deal's own built-in margin, its excess spread, shrank. When protection has to be added and earned margin falls at the same time, that is the market telling you what it expects from the collateral. **Class A credit enhancement (left) **Excess spread (right) Source: S&P Global Ratings presale commentary via Asset Securitization Report, June 2026. Class B, D and E enhancement rose on the same deal (21.10 to 25.25, 5.60 to 8.50, 3.50 to 4.50). **Inference**More credit enhancement and less excess spread at the same time is a one-way signal. It means expected losses on the collateral are rising faster than the deal's own yield can absorb. These bonds are still printing at investment grade, so this is not the market panicking. It is the market quietly repricing the same risk you are carrying if you have moved down-spectrum to keep volume up. ### Signal 4. Some of the risk is now off the balance sheet In September 2025 CarMax upsized its second nonprime deal to **$900 million** and, for the first time, sold most of the residual interest to outside investors. That earns off-balance-sheet treatment: CarMax booked a $27.0 million gain on sale, and those loans, and their losses, no longer sit on its held-for-investment book. The Q1 FY2027 release in June 2026 confirmed CAF income was down slightly precisely because of that, fewer loans outstanding after the residual sale. This is completely standard, and CarMax disclosed it plainly. I am flagging it for one reason only, and it is a reason that matters to you if you benchmark yourself against CarMax: the allowance percentage everyone quotes, 2.95% as of May 2026, is calculated only on the loans still on the balance sheet. The growing nonprime pool that moved off-sheet is serviced by CarMax but is not in that denominator. So the headline reserve ratio will keep looking cleaner than the total book of credit CarMax is actually managing. If you are using their number as a yardstick, add the off-sheet book back in your head before you do. ## The securitization record CarMax is not improvising here. It has 30 years of issuance history. What is new is the second shelf, and the recent shape of it is worth a look on its own, because the nonprime program scaled hard and then pulled back. Nonprime shelf, deal size by issuance The program launched at $666.6M, peaked at $900M in September 2025, then stepped down across the next two deals. The prime shelf, for scale, has run a steady $1.2B to $1.6B per deal the whole time. **Nonprime deals **Most recent (CMXS 2026-B) Source: SEC EDGAR FWP filings and Asset Securitization Report. 25-B shown at its upsized $900M size with residual sold off-balance-sheet. | Deal | Issued | Size | Shelf | |---|---|---|---| | CAOT 2022-1 | Jan 19, 2022 | $1.60B | Prime | | CAOT 2022-4 | Oct 31, 2022 | $1.38B | Prime | | CAOT 2023-2 | Apr 19, 2023 | $1.50B | Prime | | CAOT 2023-3 | Jul 26, 2023 | $1.20B | Prime | | CAOT 2024-2 | Apr 24, 2024 | $1.60B | Prime | | CMXS 2024-A | Jun 26, 2024 | $666.6M | Nonprime · new shelf | | CMXS 2025-A | Mar 26, 2025 | $800M | Nonprime | | CAOT 2025-3 | Jul 23, 2025 | $1.45B | Prime | | CMXS 2025-B | Sep 24, 2025 | $900M | Nonprime · off-B/S | | CMXS 2026-A | Feb 18, 2026 | $750M | Nonprime | | CMXS 2026-B | Jun 2026 | $600M | Nonprime | The prime book is range-bound. All of the growth, and then the recent pullback, is on the nonprime side. Three straight smaller nonprime deals do not fit the 2024 to 2025 growth story, and I would not over-read it, but it is worth watching whether that is market conditions, an origination-volume call, or the shelf bumping a ceiling. ### The prime pool actually got better, and that is the fair part of the story I do not want to leave you thinking this is all one direction. The April 2024 underwriting tightening worked, and it shows up cleanly in the prime collateral. The weighted-average FICO on the prime deals climbed deal over deal, with the 2025 vintage landing far above where the troubled 2022 and 2023 paper was written. Prime shelf, weighted-average FICO at issuance The 2022 and 2023 deals (FICOs in the low 700s, written at peak car prices) are the cohorts that produced the $71.3M revision. After tightening, the 2025 pool came in materially stronger. Source: Fitch presale reports via Asset Securitization Report and SEC FWP filings. 2025-3 FICO is approximate. So the picture is honest in both directions. The new paper is cleaner. The old paper is still seasoning, and it is the old paper that surprised management. Both things are true at once, and you need to hold both to read the book correctly. ## What I could not nail down I would rather tell you the edges of what I know than pretend the file is airtight. Three things I deliberately left soft or out: The exact cumulative net-loss curves by vintage live in the loan-level ABS-EE filings on EDGAR, and building them means parsing XML across dozens of monthly reports. I did not do that here. This brief leans on management's own admission of vintage underperformance rather than my own curve extraction. If you need deal-level precision, that extraction is the logical next step. I dropped a recovery-rate figure that was floating around an earlier draft of this analysis (a roughly 38% cumulative recovery on one 2022 deal) because I could not source it to a primary filing cleanly. What I can stand behind: Fitch's subprime recovery index was running near 33% late in 2025, against a pre-pandemic average closer to 44%. That structural gap, not any single deal's number, is the point. Recoveries are simply worth less than your old loss model probably assumes. And a weighted-average LTV figure on the latest nonprime deal showed up in secondary commentary but not in a primary filing I could open, so I left it out. The Signal 3 argument does not need it; it rests on the credit-enhancement and excess-spread numbers, which are confirmed. ## What this means for your book CarMax is a $16 billion-plus managed book run by the biggest used-car retailer in the country. You are probably not that. Which is the whole point: if peak-vintage stress and recovery compression can surprise a team with that much scale and history, the same mechanics at your size, with less structural cushion and no off-balance-sheet release valve, will show up harder and with less warning. So here are the three questions I would actually go ask of my own portfolio this week. **1. What do my 2022 and 2023 vintages look like at 24 to 36 months of seasoning?** That is the window where used-vehicle losses tend to peak. Your blended delinquency rate can look calm while a specific cohort underneath it is accelerating. The aggregate hides exactly the thing you need to see. Pull the cohort, not the total. **2. What am I really recovering at auction versus 2021?** If your loss model still assumes 45 to 48% recovery and you are actually getting 35 to 40%, your net loss per default is materially higher than the model says, and your reserve is quietly short. The shortfall stays invisible until the book seasons enough to expose it. Go check the actual number. **3. If I am reaching down-spectrum to hold volume, is my reserving moving with me?** CarMax added cushion to every nonprime tranche while its excess spread fell. That is the market telling them the new collateral is riskier. The question for you is whether your internal reserve assumptions reflect that, or whether you are quietly assuming the new, weaker business performs like the old business did. It will not. None of this is a reason to panic about CarMax, and I have tried hard not to dress it up as one. It is a reason to go read your own tape the same way, before someone else reads it for you. **Sources & notes** The $71.3M lifetime-loss revision, the $507.3M / 3.02% allowance, the prior-quarter figures, the 2022 and 2023 vintage language, and the April 2024 tightening are all from CarMax's [Q2 FY2026 earnings release (September 25, 2025)](https://www.sec.gov/Archives/edgar/data/1170010/000117001025000115/q2fy26earningsrelease.htm), filed as an 8-K on SEC EDGAR and corroborated across BusinessWire and Morningstar. The "high watermark" management comment is from the CarMax Q1 FY2026 earnings call (June 2025), per published transcripts. Quarterly allowance figures (Feb 2025 through May 2026) are from the respective CarMax earnings releases, each filed as an 8-K. CMXS 2024-A collateral (FICO 603, APR 16.06%, LTV 97%, ValuMax 31%, $666.6M, 32,816 loans) is from Fitch via [Asset Securitization Report](https://asreport.americanbanker.com/news/carmax-select-receivables-trust-to-raise-625-million). CMXS 2026-B structure (FICO 613, excess spread 8.42% vs 9.85%, credit enhancement by class) is from S&P Global Ratings via [Asset Securitization Report (June 2026)](https://asreport.americanbanker.com/news/carmax-raises-600-million-from-used-vehicle-contracts). Deal sizes and dates are from SEC EDGAR FWP and 8-K filings. The Fitch subprime recovery index (~33% late 2025 vs ~44% pre-pandemic) is from Fitch Ratings commentary. The $900M upsized CMXS 2025-B, the $27.0M gain on sale, and off-balance-sheet treatment are from CarMax's Q2 and Q3 FY2026 and Q1 FY2027 releases. The securities class action reference is from public filings reported in October and November 2025 and is cited as context, not as any finding of wrongdoing. **Where this brief reasons beyond what a filing literally states, it is labeled as an inference.** LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned, and this is not investment, legal, or accounting advice. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). LR LendRisk Analytics Independent market research Continue reading [What the Tape Said · Issue 1 What the tape said: *Tricolor Holdings*](https://lendriskanalytics.com/insights/tricolor-tape.html) [Vol · 09 · Sector note Buy-here-pay-here *grew up*](https://lendriskanalytics.com/insights/bhph-institutionalized.html) --- title: "What the tape said: Tricolor Holdings" url: https://lendriskanalytics.com/insights/tricolor-tape.html publisher: LendRisk Analytics series: What the Tape Said issue: 1 published: 2025-09-10 kind: Post description: "A post-mortem on what was publicly visible about Tricolor Holdings before its September 2025 collapse: funding dependence, a thin-file collateral pool (62% no credit score in the final deal), and diligence blind spots, read straight from the public filings." html: https://lendriskanalytics.com/insights/tricolor-tape.html --- # What the tape said: Tricolor Holdings [← All insights](https://lendriskanalytics.com/articles.html) What the Tape Said · Issue 1 Post-mortem · 9 min read Post-mortem · Tricolor Holdings # What the tape said: Tricolor Holdings. Public record as of September 10, 2025 · LendRisk Analytics This memo answers one question only. What was publicly visible about Tricolor Holdings *before* September 2025, and what did those public filings actually show? Bankruptcy filings, DOJ charges, and post-collapse commentary are excluded from the signal analysis. Where an inference is made, it is labeled as such. 62.0% Borrowers with no credit score, final pre-collapse deal $2.2B / $1.4B Pledged vs. real collateral (later alleged by DOJ) down 34% Final deal vs. the deal three months earlier $1.9B Across 8 prior rated securitizations ## Bottom line The clean pre-September 2025 public signal was not that fraud was obvious in the filings. The cleaner read is this. Tricolor was structurally dependent on repeat securitization to fund its business. That program accelerated aggressively through early 2025, then abruptly shrank in June 2025. The borrower mix in the June deal looked materially weaker than the March deal. **62% of borrowers in the final pre-collapse deal had no credit score whatsoever.** Public due diligence explicitly excluded underwriting conformity, collateral value, legal compliance, and performance risk. And public market marks stayed relatively calm through August 2025, so senior bond pricing alone would not have flagged the problem. The read The pre-collapse warning was not a blatant public default signal. It was **funding dependence, plus diligence blind spots, plus a borrower base that depended entirely on Tricolor's own internal data integrity to validate.** When that data integrity failed, everything failed simultaneously. ## The four strongest pre-collapse signals Ranked by how actionable they were for a warehouse lender or investor reading in real time. ### Signal 1. 62% of final-deal borrowers had no credit score TAST 2025-2, the last Tricolor securitization before the collapse, disclosed that approximately **62.0% of borrowers did not have a credit score.** This is the sharpest single pre-September collateral signal in the public record. A pool with no external borrower-validation signal depends entirely on the originator's internal underwriting data being accurate and complete. Borrower validation, TAST 2025-2 Of the loans backing the final pre-collapse deal, nearly two in three had no external credit score at all. Source: KBRA pre-sale report, TAST 2025-2 (statistical cut-off Apr 30, 2025). **Inference**For a lender already operating in subprime auto, that share of thin-file borrowers meant the entire pool's performance depended on Tricolor's own data integrity. When the DOJ later alleged that data was fabricated, the absence of external credit validation meant there was nothing external to contradict it. ### Signal 2. The securitization machine accelerated, then abruptly shrank Tricolor scaled its ABS issuance aggressively, from $212M in 2022 to $328M by March 2025. Then the June 2025 deal came in at $217M, a **34% reduction** from the prior deal three months earlier. The non-zero weighted-average FICO also dropped from 614 in the March deal to 600 in the June deal, while the no-score share stayed extremely high. Rated securitization issuance by deal Three years of steady growth, then a sharp contraction in the final deal before the collapse. **Prior deals **Peak (TAST 2025-1) **Final deal (TAST 2025-2) Source: KBRA rating reports. TAST 2024-1 issuance size not disclosed (n/d). **Inference**A lender that needs structured-funding access to operate, and whose latest deal is materially smaller and weaker-collateralized than the prior one, is a lender under funding pressure. The question a warehouse lender should have been asking in July 2025: why did the June deal shrink, and what happened to the collateral that did not make it into that trust? ### Signal 3. Public due diligence explicitly excluded what later mattered most In the ABS-15G third-party due-diligence report for TAST 2025-2 (filed June 2025), the diligence provider sampled 150 receivables and checked a narrow set of fields. The report then explicitly stated it was *not* conducted for the purpose of testing conformity with underwriting or credit-extension guidelines, addressing collateral value, addressing compliance with federal, state, and local law, or addressing any factor material to whether investors would actually receive principal and interest. **Inference**The diligence package that went to investors in June 2025 explicitly excluded the exact categories where fraud was later alleged. This is not a criticism of the diligence firm. It was engaged for a narrow, standard scope. It is a structural observation about what public ABS diligence does and does not cover. ### Signal 4. The rating agency used no third-party diligence across 2024 to 2025 KBRA's information-disclosure forms for TAST 2024-1, 2024-2, 2024-3, 2025-1, and 2025-2 each carried the same language. In taking the rating action, KBRA did not use the due-diligence services of a third-party provider. The ratings were based on issuer-provided data, historical performance, company financials, servicer reports, and prior structure attributes. **Inference**This is standard practice in ABS, and the ratings may well have been appropriate at issuance. The analytical point is that the entire public verification chain for five consecutive Tricolor deals ran through issuer-provided data. Independent verification of the underlying collateral and underwriting was not part of the public process. ## The securitization record in full Tricolor was not a startup that stumbled into ABS. It was a repeat issuer with a long-running structured-funding program. By June 2025 it had completed five unrated securitizations totaling roughly $545 million and eight rated securitizations totaling roughly $1.9 billion. TAST 2025-2 was its ninth rated deal. | Rated deal | Published | Size | Direction | |---|---|---|---| | TAST 2022-1 | Apr 20, 2022 | $212.13M | flat | | TAST 2023-1 | Feb 2, 2023 | $223.97M | up | | TAST 2024-1 | Jan 31, 2024 | n/d | flat | | TAST 2024-2 | May 8, 2024 | $276.70M | up | | TAST 2024-3 | Oct 2, 2024 | $287.55M | up | | TAST 2025-1 | Mar 5, 2025 | $328.10M | up | | TAST 2025-2 | Jun 4, 2025 | $217.18M | down 34% | The program grew consistently for three years. The final pre-collapse deal was 34% smaller than the one three months prior. That inflection is the timeline anchor for any real-time signal analysis. ## The collateral pool: what changed deal to deal The borrower profile of each deal is publicly available in KBRA's pre-deal summaries. Comparing the last three deals shows a pool that was not improving into the collapse. Deals are labeled by issuance date; FICO, balance, and APR are stated as of each deal's statistical cut-off. Non-zero weighted-average FICO, last three deals The score ticked up into early 2025, then fell again in the final deal as the program reached deeper into thin-file borrowers. Source: KBRA pre-sale reports. Non-zero weighted-average FICO excludes the no-score borrowers entirely. | Metric | TAST 2024-2 (issued May 2024) | TAST 2025-1 (issued Mar 2025) | TAST 2025-2 (issued Jun 2025) | |---|---|---|---| | Non-zero WA FICO | 604 | 614 | 600 | | Avg principal balance | $24,861 | $21,381 | $20,943 | | WA APR | 17.10% | 16.64% | 16.90% | | No-score borrowers | n/d | n/d | ~62.0% | FICO declined between the March and June 2025 deals; APR ticked back up; average balance kept shrinking. This does not look like a lender de-risking before a slowdown. It looks like a lender reaching deeper into thin-file borrowers to sustain origination volume while deal size was already contracting. ## What the market was seeing: public bond marks This is the counter-argument, and it matters for intellectual honesty. Public fund holdings from N-PORT filings show Tricolor bonds still marked near or above par through mid-2025. For example, an AB short-duration ETF carrying TAST 2025-1 essentially at par as of May 31, 2025, and an Angel Oak fund carrying even junior TAST 2023-1 paper *above* par as of April 30, 2025. The public market was not broadly pricing distress before September. **What this means**A warehouse lender monitoring only public bond marks would not have seen the collapse coming. The pre-collapse signals were in the collateral-composition data and the structural diligence gaps, not in secondary-market pricing. The signal required reading the prospectus supplements and the ABS-15G filings, not the Bloomberg screens. ## What was not clearly public before September 2025 Intellectual honesty about the limits of the public record matters as much as the signals that were there. The following were **not** visible in the pre-collapse public record: a public warehouse covenant breach or acceleration notice; a public wave of license surrenders; a public ABS payment default or missed remittance; a collapse in senior bond pricing before September 10; or any public disclosure of the alleged double-pledging of collateral. The alleged fraud mechanism, that the same collateral was pledged to multiple warehouse lenders simultaneously, was not detectable from public filings alone. It required access to each warehouse lender's internal records, which were not public. **The pre-collapse public file raised structural questions about funding dependence and diligence adequacy. It did not prove fraud was occurring.** Those are different analytical claims, and conflating them would undermine the analysis. ## The implication for operators still in the market Tricolor's collapse is a fraud story. But the fraud accelerated a structural vulnerability that was visible before the fraud became public. Any lender or investor in subprime auto right now should be asking three questions. **1. What share of my borrower base has no external credit signal?** If it is high, portfolio performance depends entirely on internal underwriting-data quality. That is a concentration of operational risk that belongs in your risk framework. **2. Is my structured-funding program accelerating faster than the collateral quality supports?** A growing ABS program on a weakening pool is a structure under stress even before any default. **3. What does my third-party diligence actually cover?** The ABS-15G scope exclusion in TAST 2025-2 was not unusual. It is standard. The question is whether your warehouse diligence covers the gaps the ABS diligence explicitly does not. Tricolor is the extreme case. The structural questions it raises are not extreme. They are the ordinary risk-management questions a stressed subprime auto market makes urgent. **Sources & notes** Deal sizes, FICO, balances, APR, the ~62% no-score figure, and the rated and unrated program totals are from KBRA's pre-sale and rating reports, for example [TAST 2025-2](https://www.kbra.com/publications/WYLVLYwz/kbra-assigns-ratings-to-tricolor-auto-securitization-trust-2025-2) and [TAST 2025-1](https://www.kbra.com/publications/ScrpWJpX/kbra-assigns-ratings-to-tricolor-auto-securitization-trust-2025-1). Third-party due-diligence scope is from the ABS-15G filings on SEC EDGAR; public bond marks are from fund N-PORT filings on SEC EDGAR. The collateral-fraud allegations ($2.2B pledged vs ~$1.4B real; double-pledging) and the executive charges are from the [U.S. Attorney (SDNY) announcement](https://www.justice.gov/usao-sdny/pr/ceo-cfo-coo-charged-connection-billion-dollar-collapse-tricolor-auto) and are **allegations**; nothing here asserts guilt. This memo is a point-in-time reading of the public record as of September 10, 2025. LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned, and this is not investment, legal, or accounting advice. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). LR LendRisk Analytics Independent market research Continue reading [Vol · 05 · Case study Anatomy of a *toxic book*: the aggregate said fine](https://lendriskanalytics.com/insights/toxic-book.html) [Vol · 04 · Teardown What the *Tricolor* collapse says about reading the slope](https://lendriskanalytics.com/insights/tricolor.html) --- title: "The waiting tax: the most expensive repo is the one you didn't make" url: https://lendriskanalytics.com/insights/the-waiting-tax.html publisher: LendRisk Analytics kind: Servicing description: "In deep subprime, the largest controllable loss isn't the auction price. It's the recovery you forfeit by hesitating, because the odds of getting the car back collapse far faster than the car depreciates." html: https://lendriskanalytics.com/insights/the-waiting-tax.html --- # The waiting tax: the most expensive repo is the one you didn't make [← All insights](https://lendriskanalytics.com/articles.html) Vol · 10 Servicing · 5 min read Servicing # The waiting tax: the most expensive repo is the one you didn't make. Underwriting gets the meetings. Servicing gets the loss. In deep subprime the book is mostly written, the borrowers are mostly going to be late, and the question that actually moves the number is not who you funded. It is what you do in the two weeks after a payment stops. The single most expensive, most controllable line item on a subprime book is the recovery you give up by hesitating to repossess. 16.6× How much likelier a BHPH loan is to be in active repossession 33-46% Subprime loan-balance recovery at sale (S&P), a separate metric from the car's value ~1%/mo Roughly what the car depreciates while you wait (2026 wholesale, Black Book/Manheim) Ask a subprime operator why recovery is low and you will hear about the auction, soft wholesale values, a thin used market, a beat-up unit. All real, all roughly a couple of points a month of depreciation, and all a distraction. The car losing value slowly is not what destroys a recovery. What destroys a recovery is the car you never get back. ## Recovery is a product of two numbers, and you only watch one. Expected recovery on a defaulted loan is the value of the unit *times the probability you actually repossess it.* Operators obsess over the first term and ignore the second. The second is the one that collapses. As a delinquent borrower realizes the repo is coming, they stop answering, let the insurance lapse, park the car somewhere else, pull the plates, disable the GPS, or simply move. Each of those is a step toward a unit you will never see, and they happen on a timeline of days, not months. So while the unit depreciates only about one percent over a month, your odds of getting physical custody can fall by half or more over the same window for an evasive borrower. Multiply the two and expected recovery falls off a cliff that is almost entirely the second term. The auction price you were worried about never even gets a vote, because there is no car at the auction. **You don't lose the car to depreciation. You lose it to waiting.** Depreciation is the slow, visible number everyone watches. The collapse in recovery probability is the fast, invisible number that actually writes the loss. ## This is why the cure notice is a timing tool, not a formality. Most states require a right-to-cure notice, commonly 10 to 21 days, before a lender can lawfully repossess. Operators tend to treat that as a delay to be endured after they have decided to act. That is backwards. The cure clock is the constraint, so the move is to start it *early*, the moment the soft signals appear, a first missed payment, a broken promise, a disconnected phone, not after a hard trigger forces your hand. Serve the notice on the soft signal and the legal window is already open when the hard trigger hits. Serve it late and you are watching the unit bleed value for two to three weeks before you can legally touch it. The hard triggers are the ones that should never wait: a tampered GPS or starter-interrupt device, insurance lapsed on collateral you still own, a payment reversed with no contact since, a second broken promise. Those are not "monitor" events. Those are "the car is being put out of reach" events, and the right response is to move the instant you are lawfully clear. ## The discipline is cheap. The hesitation is not. None of this requires a better auction, a richer borrower, or a tighter underwrite. It requires a written trigger policy, a cure notice that goes out on the soft signal, and the will to dispatch recovery the day a hard trigger fires. It is the cheapest loss-reduction lever in subprime auto, and it is the one most books leave on the table because hesitation never shows up as a line item. It just shows up as a lower recovery rate that everyone blames on the market. Method note Map your triggers and price the cost of every day you wait [Read the method note →](https://lendriskanalytics.com/repo-timing.html) Put a number on it once and it changes how the whole desk behaves. The unit on the lot that "we'll give them another week" is not a courtesy. It is a transfer from your recovery to nobody. The operators who win the institutionalized version of this business are the ones who treat repossession timing as math, run the cure clock ahead of the trigger, and stop paying the waiting tax. **Sources & notes** Repossession and BHPH figures from the Federal Reserve's [May 2026 FEDS note on BHPH lending](https://www.federalreserve.gov/econres/notes/feds-notes/subprime-auto-lending-trends-in-buy-here-pay-here-auto-lending-20260508.html); repossession timing, completion, redemption, deficiency, and fee figures from CFPB auto-repossession data (2022-2025); subprime loan-balance recovery range (33-46%) from S&P / auto-ABS remarketing data via [Auto Remarketing](https://www.autoremarketing.com/subprime/another-view-vehicle-values-recovery-rates/), note this is recovery against the loan balance, a different measure than the car's resale value; 2026 wholesale depreciation (~1%/mo) from Black Book / Manheim indices; right-to-cure periods vary by state, see the [state recovery law map](https://lendriskanalytics.com/repo-map.html). The custody-probability dynamics described here are **modeled scenario priors**, illustrative of the structure, not fitted to a proprietary repossession dataset. Independent analysis, not investment, legal, or accounting advice. LR LendRisk Analytics Independent market research Continue reading [Vol · 09 · Sector note Buy-here-pay-here *grew up*](https://lendriskanalytics.com/insights/bhph-institutionalized.html) [Data narrative The negative equity machine: why long terms on old cars *manufacture loss*](https://lendriskanalytics.com/insights/negative-equity-machine.html) --- title: "Buy-here-pay-here grew up" url: https://lendriskanalytics.com/insights/bhph-institutionalized.html publisher: LendRisk Analytics kind: Sector note description: "BHPH used to be cash-funded and self-insured. The Federal Reserve's 2026 data shows it is now a bank-financed, guarantor-backed asset class, and institutionalization imports finance-company fragility into a segment that used to absorb its own losses." html: https://lendriskanalytics.com/insights/bhph-institutionalized.html --- # Buy-here-pay-here grew up [← All insights](https://lendriskanalytics.com/articles.html) Vol · 09 Sector note · 6 min read Sector note · BHPH # Buy-here-pay-here grew up. The mental image of buy-here-pay-here is a corner lot, a hand-written ledger, and an owner who carries his own paper because no bank will. That business still exists. But the Federal Reserve's 2026 look at the segment describes something else underneath it: a fast-growing, bank-financed, guarantor-backed asset class. The corner lot is being institutionalized, and institutionalization changes what a bad month does. +214% BHPH balance growth since 2018 (vs 34% for traditional finance) $2B+ Bank loan commitments identified to major BHPH dealers 81% Of BHPH bank loans fully backed by guarantors Start with the growth. Per the Fed's note, BHPH loan balances are up roughly **214 percent since 2018**, against about 34 percent for traditional auto finance over the same span. BHPH is still small in absolute terms, around 2 percent of the $1.6 trillion auto market and roughly 5 percent of subprime, but it is not growing like a sleepy retail niche. It is growing like something with outside capital behind it. That capital is the real story. The Fed identifies more than **$2 billion in loan commitments** from banks to major BHPH dealers, with **81 percent** of those bank loans fully backed by guarantors and **65 percent** structured as asset-based lending, both materially higher than for other dealer types. In plain terms: the money behind a growing share of these lots is borrowed, secured against the loan portfolio, and personally guaranteed. The cash-and-carry operator who could ride out a rough quarter on his own balance sheet is being replaced, at the top of the segment, by a leveraged one who cannot. **Leverage is the part that changes the risk, not the borrower.** A self-funded lot that takes a 20 percent charge-off year loses its own money and survives. A bank-financed, guarantor-backed lot that takes the same year is now answering to a borrowing base, an advance rate, and a personal guarantee. The collateral didn't get riskier. The capital structure did. ## The performance was always going to be loud. BHPH sits at the deepest end of the credit curve, and the numbers reflect it. The Fed reports about **10 percent of BHPH balances delinquent** as of Q3 2025, versus 3.8 percent for traditional lenders, with BHPH loans roughly **16.6 times more likely to be in active repossession.** Over half of BHPH balances go to deep-subprime borrowers under a 580 score. Average origination is around **$15,400** at a **25.4 percent** rate, and a meaningful slice is structured on weekly or biweekly pay. None of that is new or alarming on its own, high loss is the business model, and a properly priced BHPH book is built to absorb it. What is new is who eats the loss when the model is funded with someone else's money. A 25 percent coupon is plenty of spread to self-insure against repossession loss. It is a thinner cushion once a chunk of that spread is servicing a bank line and the line has covenants of its own. ## Institutionalization imports finance-company fragility. This is the part worth saying plainly, because it is the whole point. When you wrap a cash retail business in warehouse-style leverage, you import the failure mode of a finance company into a segment that used to fail like a store. The questions that sink leveraged subprime lenders, advance-rate haircuts, a borrowing-base shortfall, a covenant trip that triggers a sweep right when you need liquidity, now apply to the corner lot too. And the guarantor structure means the operator's personal balance sheet is inside the blast radius. The discipline that protects a self-funded lot, repossess fast, keep the recovery, don't let a delinquent unit rot, becomes a balance-sheet necessity once the lot is leveraged. Servicing tempo stops being an operational preference and starts being the thing that keeps you inside your covenants. That is the same lesson the rest of subprime auto keeps relearning: the snapshot looks fine until the slope catches up, and by then the bank has already seen it. Method note See what every day of repo delay costs a leveraged BHPH book [Read the method note →](https://lendriskanalytics.com/repo-timing.html) BHPH growing up is not a bad thing. More capital means more cars financed for people the prime market ignores, which is the point of the segment. But capital arrives with a capital structure, and the operators who thrive in the institutionalized version will be the ones who run the book like the bank already is, watching the slope, naming the channels, and treating recovery timing as a number on the balance sheet rather than a feeling at the lot. **Sources & notes** All BHPH figures from the Federal Reserve's [May 2026 FEDS note, "Subprime Auto Lending: Trends in Buy Here Pay Here Auto Lending"](https://www.federalreserve.gov/econres/notes/feds-notes/subprime-auto-lending-trends-in-buy-here-pay-here-auto-lending-20260508.html) (delinquency, repossession, growth, bank-commitment, guarantor, loan-size, rate, and deep-subprime figures, Q3 2025 data). Market-size context from the same source and industry reporting. Independent analysis, not investment, legal, or accounting advice. LR LendRisk Analytics Independent market research Continue reading [Vol · 02 · Sector note BHPH charge-offs in 2026: what *normal* actually looks like](https://lendriskanalytics.com/insights/bhph-normal.html) [Vol · 10 · Servicing The waiting tax: the most expensive repo is *the one you didn't make*](https://lendriskanalytics.com/insights/the-waiting-tax.html) --- title: "Early payment default isn't a credit event. It's a fraud signal." url: https://lendriskanalytics.com/insights/early-payment-default.html publisher: LendRisk Analytics kind: Underwriting description: "In deep subprime, a loan that defaults in the first three payments rarely went bad. It started bad. EPD is the fingerprint of misrepresentation at origination, and it clusters in a handful of dealers." html: https://lendriskanalytics.com/insights/early-payment-default.html --- # Early payment default isn't a credit event. It's a fraud signal. [← All insights](https://lendriskanalytics.com/articles.html) Vol · 08 Underwriting · 6 min read Underwriting # Early payment default isn't a credit event. It's a fraud signal. A subprime loan that goes 90 days delinquent in year two went bad. A loan that misses its first three payments was bad the day it was written. The industry treats both as credit problems and reaches for the same lever, a higher score floor. That is the wrong lever, because early payment default is rarely about the borrower's credit. It is about what was true on the application, and who sent it. 30-70% Of early payment defaults tied to application fraud ~10% Of dealers tied to nearly all EPD losses $10.4B Record auto-lending fraud exposure, 2026 Early payment default, EPD, means the loan stops performing almost immediately, typically a default inside the first three to six payments. It is rare. Point Predictive's data puts it at under 1 percent of prime auto loans and under 5 percent of subprime. That rarity is exactly what makes it loud. A borrower who could make a down payment, pass a stip check, and then never make a single real payment did not get unlucky. Something in the file was not true. ## It is not the borrower. It is the application. The fraud research is blunt about this. Loans that default in the first six months carry a far higher probability of material misrepresentation than loans that sour later, and analysts put the share of EPD linked to some form of fraud at roughly **30 to 40 percent**, with more recent work tying as much as **70 percent** of early payment defaults to fraudulent applications. The misrepresentation is usually mundane, not cinematic: inflated or fabricated income, fake employment, a straw borrower, a power-booked vehicle value, a synthetic identity stitched together to clear an automated decision. None of that shows up in a credit score. A score measures a real person's real history. It cannot measure whether the person, the income, or the car on the contract is real. That is the entire reason raising your FICO floor does not fix EPD: you punish honest thin-file borrowers, the exact customers a subprime book exists to serve, and you still fund the fraud, because the fraud was engineered to clear whatever floor you set. **EPD is uncorrelated with the thing your underwriting actually measures.** It is a verification failure, not a creditworthiness failure. Treating it as a credit problem means tightening the one dial that cannot move it. ## And it does not come from everywhere. It comes from a few doors. Here is the part that should change how a subprime lender spends its attention. The losses are not diffuse. Point Predictive's analysis finds that close to **100 percent of fraud losses come from fewer than 3 percent of dealers**, and close to **100 percent of early-payment-default losses come from about 10 percent of dealers.** Systematic dealer behavior, repeatedly inflating values, recycling fake employers, can raise the default risk on that dealer's paper by as much as **500 percent.** That is not a borrower-underwriting story. It is a channel-surveillance story. The signal you are looking for is not a number on a single applicant; it is a pattern across a dealer's submissions. A channel that produces three EPDs in its first thirty contracts is not a channel with unlucky customers. It is a channel telling you exactly what it is, in the only language that does not lie. ## The right response, then, is not a tighter borrower. It is a watched door. Two moves, in order. First, move fraud detection *in front of* funding. The same research suggests a finance company can flag half or more of its eventual EPD before the money goes out, using fraud and income-verification models rather than credit score alone. The cheapest EPD is the one you never funded. Second, run EPD as a dealer metric, not a portfolio metric. Track first-payment and early-default rates by source, weight your diligence toward the doors that spike, and put recourse and probation on the channels that earn it before they earn it twice. A new dealer's first thirty contracts tell you more than their references ever will, and the public-record profile of a high-risk dealer, entity cycling, name changes, prior consumer-protection actions, is usually visible before the first loan is even funded. Method note Score a dealer before you board it, and rank the ones you already have The deep-subprime business is built on funding people the rest of the market won't, and most of those people pay. The book does not get killed by the honest thin-file borrower. It gets killed by the small set of loans that were never real, funneled through the small set of dealers that keep sending them. EPD is the alarm on that exact door. Stop reading it as a credit number. Start reading it as an address. **Sources & notes** EPD and fraud figures from Point Predictive's auto-fraud research, including its [2026 Auto Lending Fraud Trends Report](https://pointpredictive.com/press-releases/point-predictive-releases-2026-auto-lending-fraud-trends-report-fraud-exposure-reaches-record-10-4-billion/) (record $10.4B exposure; EPD <1% prime / <5% subprime; dealer-concentration figures) and its [analysis of US auto-fraud losses](https://pointpredictive.com/estimating-auto-fraud-lending-losses-in-the-united-states/); the 70%-of-EPD-tied-to-fraud figure via [Auto Finance News](https://www.autofinancenews.net/allposts/risk-management/fraud-alert-70-of-early-payment-defaults-tied-to-fraudulent-applications/) and [Auto Remarketing](https://www.autoremarketing.com/subprime/point-predictive-spots-record-auto-fraud-exposure-climbing-early-payment-default-risk/). Figures are industry estimates and vary by dataset and definition of EPD. Independent analysis, not investment, legal, or compliance advice. LR LendRisk Analytics Independent market research Continue reading [Vol · 05 · Case study Anatomy of a *toxic book*: the aggregate said fine](https://lendriskanalytics.com/insights/toxic-book.html) [Vol · 10 · Servicing The waiting tax: the most expensive repo is *the one you didn't make*](https://lendriskanalytics.com/insights/the-waiting-tax.html) --- title: "Run the book: a tool-by-tool teardown" url: https://lendriskanalytics.com/insights/run-the-book.html publisher: LendRisk Analytics kind: Case study description: "One blinded $40M subprime auto book, run through all six LendRisk Analytics tools in sequence. Every input, every output, every decision, from 'you look compliant' to a covenant breach six months out, the three dealers causing it, the deal to decline, and the next bad dealer stopped at the door." html: https://lendriskanalytics.com/insights/run-the-book.html --- # Run the book: a tool-by-tool teardown [← All insights](https://lendriskanalytics.com/articles.html) Vol · 06 Case study · Tool-by-tool · 11 min read Blinded walkthrough # Run the book. [Anatomy of a Toxic Book](https://lendriskanalytics.com/insights/toxic-book.html) told the story. This is the operation, the same blinded $40M book, run through all six tools in sequence, every input and every output shown. It starts at *“you look compliant”* and ends six months from a cash sweep, with the three dealers causing it named, one deal declined on the math, and the next bad dealer stopped at the door before a single loan is funded. The book · blinded composite $40M Senior warehouse line 2,000 Active loans · 18 dealers 9.6% NCO · annualised 574 Weighted-avg FICO Every monthly compliance certificate this lender sends its warehouse bank comes back green. Loss is under the cap, delinquency is under the trigger, FICO is above the floor. The CFO believes the book is fine. We are going to run it through six tools, in the order a lender actually faces the problem, and watch “fine” come apart. 1 Tool 05 · Portfolio Risk Calculator Where do we stand against the thresholds our bank is watching? Three numbers off the servicing system, loss rate, severe lates, weighted FICO, scored 0 to 100 against warehouse covenant bands. Portfolio Risk Calculator Live Inputs Loss rate · NCO 9.6% Severe lates · 90+ DPD 4.2% Borrower quality · FICO 574 Composite score 61 / 100 Watch closely Band 0-39 Healthy Band 40-69 Watch · you are here Band 70-100 Action NCO 9.6% ** WARN 90+ DPD 4.2% ** WATCH FICO 574 ** WATCH 560-590 The read **Compliant, but in the warning band, not the healthy one.** A 61 says every covenant has headroom today, yet nothing is comfortable. NCO carries most of the score (it is the 40-point lever), and FICO at 574 sits inside the 560-590 watch zone, not safely above it. This is the number that lets a CFO say “we’re fine.” It is a snapshot. It has no slope. So we ask the next question. [Run it yourself →](https://lendriskanalytics.com/tool.html) 2 Tool 06 · Covenant Breach Projector At our current rate of change, where are we headed, and when? Same metrics, plus how fast each is moving per month, plus the actual covenant caps in the facility agreement. The tool returns the first metric to break and the months of runway left. Covenant Breach Projector · 18-month horizon Live Where you are · rate of change NCO 9.6% Δ +0.65 / mo 90+ DPD 4.2% Δ +0.30 / mo FICO 574 Δ −3 / mo Covenant caps NCO 13.5% · DPD 6.0% · FICO 560 Projection 6 months runway First to break NCO @ 13.5% Second 90+ DPD @ 6.0% · also mo 6 Consequence Cash sweep · advances frozen Why this is the one that lands Your warehouse bank runs this on your tape every month. Most lenders never run it on themselves. The calculator said 61, compliant. The projector says the slope under that 61 puts NCO through its 13.5% cap in **month 6**, and a cash sweep starves new originations exactly when you’d need liquidity to recover. The bank’s surveillance team already sees this trajectory. The gap between the snapshot and the slope is the whole game. [Tricolor](https://lendriskanalytics.com/insights/tricolor.html) was inside covenant the quarter it failed. The read **Six months, not “someday.”** The level is fine; the rate of change is not. Something is steepening the slope. A book-wide retrenchment would strangle the healthy paper to fix a problem we haven’t located yet. So before we touch anything, we find *what* is bending the curve. [Run it yourself →](https://lendriskanalytics.com/covenant.html) 3 Tool 04 · Dealer Scorecard Who is bending the curve, and which relationships do we pull? Break the same tape down by originating dealer. Rank all 18 not by volume but by what predicts loss: net charge-off, early payment default, severe-late rate, and a composite health score. Dealer Scorecard · 18 dealers ranked Live Bottom 3 dealers · flagged Dealer N Score 18 · NCO 16.4% Dealer H Score 22 · NCO 15.1% Dealer C Score 24 · NCO 14.8% Network median Score 71 · NCO 7.7% What the 3 carry Share of book 22% Share of severe lates 51% EPD rate 28% · 3.9× network Cohort skew 2023-H2 · 72-mo paper Three dealers out of eighteen originate **22% of the book but 51% of the severe lates.** Their early-payment-default rate, borrowers missing inside the first three payments, the cleanest signal of a bad deal at inception, runs **3.9× the network.** That is not a macro problem or a servicing problem. It is origination quality at the source: paper impaired the day it was written. Pull those three out and the remaining fifteen dealers are better than benchmark. The action **Pause new originations from Dealers C, H, and N; put existing paper on enhanced watch.** The healthy fifteen keep funding. Re-run the projector with the toxic feed removed and the monthly NCO slope flattens from +0.65 to **+0.22**, runway extends past the 18-month horizon, the breach is gone. The targeted cut protects roughly **140 bps of yield** and restores about **$1.8M of covenant headroom** versus a blunt book-wide pullback. Attribution is what lets you use a scalpel instead of a sledgehammer. [Run it yourself →](https://lendriskanalytics.com/dealers.html) +0.65 → +0.22 Monthly NCO slope, after the cut ~140 bps Yield protected ~$1.8M Headroom restored 4 Tool 01 · Deal Underwriter Dealer H just sent a deal for approval. Do we fund it? Before we finished pausing Dealer H, a fresh application came through the pipe from exactly that channel. Run the borrower, the vehicle, and the terms. The tool returns a defensible *fund / counter / decline* with the math attached. Deal Underwriter Live The deal FICO 545 DTI 47% Vehicle 2014 SUV · 142k mi Wholesale value $9,800 Amount financed $16,500 · 168% LTV Term · APR · down 72 mo · 21% · $500 The math Probability of default 58% Loss given default 70% Expected loss $6,750 Net profit over life −$1,900 Expected ROA −3.6% · target +4% Default curve peaks months 9-15 DECLINE Negative ROA · collateral outlived by the loan Why decline **This is the loan that builds the toxic cohort.** $16,500 financed against a $9,800 vehicle is 168% LTV before tax and fees. A 72-month term on an eleven-year-old SUV means the collateral reaches the end of its reliable life around month 30 while the borrower still owes for another three and a half years. When the car dies, the payments stop, the default curve peaks at months 9-15, and recovery on a dead high-mileage vehicle is thin. Bad car, bad terms, bad borrower, engineered into a default. The honest *counter* exists, 48-month term, $2,500 down, drop the advance, which lifts ROA to about +1.8%, still thin. At the terms as written, you decline. Fund this and you are hand-building next quarter’s scorecard problem. [Run it yourself →](https://lendriskanalytics.com/underwriter.html) 5 Tool 07 · State Recovery Law Map When the bad paper does default, how much do we actually get back? The defaulted loans in the toxic cohort are spread across Texas, Georgia, and Louisiana. Loss given default is not one number, it changes the moment a loan crosses a state line, because four legal levers change with it. State Recovery Law Map · the four levers Live Louisiana Self-help repo No, court order required Right-to-cure notice Yes Deficiency Allowed Wage garnishment Allowed Texas Self-help repo Yes, fast, no court Right-to-cure notice No Deficiency Allowed Wage garnishment Effectively blocked The read **Two identical defaults, two different recoveries.** Louisiana is the only state with no self-help repossession, you go to court first, which is slower and more expensive, lengthening the loss. Texas lets you repossess fast and skip the cure notice, but effectively blocks wage garnishment, so the deficiency is hard to collect. The same charged-off dollar recovers differently depending on geography, which is exactly why the LGD in Step 4 was a modeled number, not a guess, and why concentration by state belongs in every reserve. [Read the method note →](https://lendriskanalytics.com/repo-map.html) ## The whole job, in one pass. The compliance certificate said the book was fine. Run in sequence, the tools said something the aggregate never could: **six months to a cash sweep, caused by three named dealers, writing loans like the one we just declined, and here is the next bad dealer, stopped before they funded a single contract.** That is the difference between a number you report and the book you actually run. Security is finding the toxic paper. Predictability is seeing the breach before it prints. Attribution is naming the dealer, the vintage, and the deal so you can act with a scalpel. None of it requires a Bloomberg terminal, it requires reading your own tape the way your warehouse bank already reads it. **On the numbers.** This is a blinded, rounded composite assembled to demonstrate the workflow end to end, not a single named client. The book mirrors the synthetic 2,000-loan tape you can drive yourself in the [Portfolio Analyzer](https://lendriskanalytics.com/analyze.html); the dealer names are placeholders. Industry context, the ~6.65% subprime delinquency high, Fitch’s 9.81% net-loss index, the Tricolor collapse, and the state recovery rules, is real and cited in [Anatomy of a Toxic Book](https://lendriskanalytics.com/insights/toxic-book.html) and the [recovery map](https://lendriskanalytics.com/repo-map.html). Not investment, legal, or accounting advice. Keep reading The same frame, applied to a real collapse. This walkthrough uses synthetic data to show how attribution, vintage isolation, and covenant runway fit together. For the real-world version, read how the same lens reads the Tricolor failure. [Read: Anatomy of a Toxic Book →](https://lendriskanalytics.com/insights/toxic-book.html) Independent market research, published for readers studying subprime auto credit. Not investment, legal, or accounting advice. LR LendRisk Analytics Independent market research Continue reading [Vol · 05 · Case study Anatomy of a *toxic book*](https://lendriskanalytics.com/insights/toxic-book.html) [Vol · 04 · Teardown What the *Tricolor* collapse actually says](https://lendriskanalytics.com/insights/tricolor.html) --- title: "Anatomy of a toxic book: how an average-looking portfolio hides a covenant breach" url: https://lendriskanalytics.com/insights/toxic-book.html publisher: LendRisk Analytics kind: Case study description: "A blinded case study. The aggregate said compliant. Attribution said six months to a covenant breach. How three dealers and one vintage nearly swept a $40M warehouse line, and how the slope was visible before the damage." html: https://lendriskanalytics.com/insights/toxic-book.html --- # Anatomy of a toxic book: how an average-looking portfolio hides a covenant breach [← All insights](https://lendriskanalytics.com/articles.html) Vol · 05 Case study · 7 min read Case study · Blinded # Anatomy of a toxic book. In September 2025 a subprime auto lender called Tricolor filed for bankruptcy while its warehouse covenants still read compliant. By the time it was over, it had pledged **$2.2B of collateral against $1.4B that actually existed**, JPMorgan had charged off $170M, and its AAA-rated bonds were trading at twelve cents. The aggregate said fine until the week it didn’t. This is how a book hides a covenant breach in plain sight, and how the slope is visible the whole time, if you read your own tape the way your bank does. The book below is a blinded composite, a subprime auto lender on a senior warehouse line, anonymized and rounded. But every benchmark around it is real and cited, and the pattern it walks through is the same one that surfaced publicly at Tricolor, at American Car Center, and across the 2022-2023 ABS vintages. The method is exactly what we run on a live tape; the industry numbers are exactly what the regulators and rating agencies published. ## The setup: a book that looks fine. A subprime auto lender, roughly **2,000 active loans** across **18 originating dealers**, financed on a **$40M senior warehouse facility**. The line carries the standard covenant package: a cumulative net loss cap, a delinquency trigger, and a weighted-average FICO floor. Miss any one and the bank can sweep cash, freeze advances, or reprice the line. At the portfolio level, nothing flashed. Blended 60+ DPD was **6.8 percent**, within a rounding error of the industry benchmark. For context, the subprime 60+ rate hit **6.65 percent in late 2025, its highest reading since the 1990s**, and Fitch’s subprime annualized net loss index reached **9.81 percent in January 2026**, a post-pandemic high. 1 So a 6.8 percent book genuinely looked like the middle of the market. Cumulative net loss was tracking under the covenant cap. On the monthly compliance certificate the lender sent its bank, every box was green. 6.8% Blended 60+ DPD Compliant Every covenant, on paper Month 6 Projected breach, unseen A blended average is a weighted blur. It tells you where the middle of the book sits. It tells you nothing about which loans are pulling the middle in which direction. The entire risk lived in the variance the average erased. ## Security: finding the toxic paper. The first cut is attribution by dealer. Rank all 18 originators not by volume but by the metrics that actually predict loss, early payment default, severe delinquency rate, and net loss per dollar originated. The distribution was not even close to uniform. **Three dealers originated 22 percent of the book but accounted for 51 percent of the severe lates.** Their early-payment-default rate, borrowers who miss inside the first three payments, the cleanest signal of a bad deal at inception, ran roughly **four times the network median**. EPD that high is not a servicing problem or a macro problem. It is an underwriting and origination-quality problem at the source. These three dealers were manufacturing paper that was impaired the day it was written. That concentration is not a modeling convenience. It is how subprime loss actually distributes. LendingTree’s analysis of the default population found **83.7 percent of defaulted auto loans sit with deep-subprime borrowers scoring under 580**, and defaults cluster years two through four of the loan, not at random. 2 Loss is concentrated by origination source and credit tier; an average sprays it evenly across a book and hides exactly where it lives. At Tricolor the concentration was its own kind of tell: after the bankruptcy, roughly **40 percent of 70,000 active loans shared a VIN with at least one other loan**, the same collateral counted twice. 3 **This is the part the aggregate cannot show you.** Pull those three dealers out and the rest of the book is healthy, better than benchmark. Leave them in and blend, and you get a 6.8 percent average that looks like an industry-normal book instead of a clean book carrying a concentrated pocket of toxic paper. Tool 04 · Attribution Rank every dealer in your network by loss, EPD, and severe lates [Open scorecard →](https://lendriskanalytics.com/dealers.html) ## Predictability: seeing the loss before it lands. Attribution tells you *who.* The vintage view tells you *when.* Group the loans those three dealers wrote by origination quarter and plot cumulative net loss against months on book. The 2023 second-half cohort was rolling from 30 to 60 to 90 days past due at roughly **1.4 times the pace** of the lender's seasoned vintages at the same seasoning point. This is not a hypothetical failure mode: Fitch flagged the **2022-2023 vintages as the worst post-pandemic performers**, originated when used-car prices peaked and stimulus-inflated FICO masked the real credit underneath. 4 The cohorts that look fine on a blended line are precisely the ones a vintage cut exposes. Roll rates are leading. Net loss is lagging. The 90-day bucket today is the charge-off in two to four months and the realized loss after that, once you net the depressed auction recovery against the deficiency. Project that cohort forward on its own roll behavior and it lands near **19 percent cumulative net loss**, against an 11 percent book average. That single pocket, left alone, drags the whole book through the covenant. Feed the trajectory into the covenant math, current cumulative loss, monthly slope, the cap in the facility agreement, and it returns a date. At the observed pace, the net loss covenant breaks in **month 6**. Not "someday." Month 6, triggering the cash sweep that starves originations exactly when the lender would need liquidity to reprice and recover. 1.4× Roll pace vs seasoned ~19% Cohort cum. loss (proj.) 11% Book average Tool 06 · Predictability Find the exact month each covenant breaks at your current trajectory [Read the method note →](https://lendriskanalytics.com/covenant.html) How your warehouse bank reads the same tape The bank is not tracking your level. It is tracking your slope. Your monthly certificate shows a compliant average. The bank's surveillance team is running the roll rates underneath it, and they often know your 90-day migration before your own monthly reporting closes. By the time the aggregate moves enough to fail a covenant, the bank has already seen the slope steepen, already started discounting your advance rate, already begun drafting the sweep notice. When Tricolor failed, the warehouse banks took the loss in public: **JPMorgan charged off $170M, Fifth Third disclosed $178-200M, Barclays impaired roughly $148M**, and its $2B of AAA-rated bonds traded down toward twelve cents. 5 The banks did not miss a number. They were watching a slope, and a collateral file, the lender's own reporting was papering over. The lender who reads its own tape the way the bank does is never the one surprised by the call. ## Attribution into action. Because the problem was named precisely, the fix could be precise too, no blunt, book-wide retrenchment that strangles the healthy 78 percent of the portfolio. Three moves, all targeted: **One, stop the bleeding at the source.** Pause new originations from the three flagged dealers pending a network review. The healthy fifteen keep funding. **Two, fix the structural defect.** The toxic cohort skewed toward older, high-mileage vehicles written on 72-month terms, loans engineered to outlive their collateral. Tighten maximum term against vehicle reliability so the loan can no longer mature years after the car is worth less than the payoff. **Three, reprice the risk that remains.** Recut advance-rate assumptions on the impaired cohort so the warehouse exposure reflects the real recovery, not the originated balance. Re-run the covenant projection with those three changes in place and the breach pushes out beyond the projection horizon. The line is preserved. On a representative book of this size, the targeted intervention protects on the order of **140 basis points of portfolio yield** and restores roughly **$1.8M of covenant headroom** versus letting the cohort run, the difference between a managed quarter and an emergency call from your bank. ~140 bps Yield protected ~$1.8M Headroom restored Line held Sweep averted ## The whole job, in one line. The aggregate said the book was fine. Attribution said it had six months. That gap, between the number you report and the slope underneath it, is where lenders get swept and where the work lives. Security is finding the toxic paper before it metastasizes. Predictability is seeing the loss before it prints. Attribution is naming the exact dealer and vintage so you can act with a scalpel instead of a sledgehammer. Deliver those three and the yield protects itself. **On the numbers.** The book itself, the 2,000 loans, 18 dealers, $40M line, the three flagged dealers and the 2023-H2 cohort, is a blinded, rounded composite built to illustrate the method, consistent with the synthetic tape in the Portfolio Surveillance demo. It is not a single named client. Everything stated about the *market*, the 6.65% subprime delinquency high, Fitch’s 9.81% loss index, the 2022-2023 vintage weakness, the default-concentration figures, and every Tricolor fact, is real and sourced below. Not investment, legal, or accounting advice. **Sources.** - Subprime 60+ DPD at 6.65%, highest since the 1990s: CNN Business, Oct 2025 . Fitch subprime annualized net loss index 9.81%, Jan 2026: Auto Remarketing / Fitch . - Default concentration, 83.7% of defaults among sub-580 borrowers; defaults cluster in loan years two to four: LendingTree Auto Loan Defaults Study . - ~40% of 70,000 active Tricolor loans shared a VIN with another loan: DealershipGuy ; Wolf Street . - 2022-2023 ABS vintages the worst post-pandemic performers: Asset Securitization Report . - $1.4B real collateral vs $2.2B pledged; warehouse losses (JPMorgan ~$170M, Fifth Third $178-200M, Barclays ~$148M); $2B AAA ABS trading near 12¢: CNBC ; U.S. DOJ (SDNY) ; Banking Dive . The frame, in three parts How this analysis reads a book. The same lens runs through everything published here, attribution, vintage isolation, and covenant runway. Three questions it keeps coming back to: - A dealer scorecard ranking every originator by loss, EPD, and severe lates - A vintage loss projection isolating the cohorts driving the number - One covenant runway figure, the month each trigger breaks at the current slope Each of these is walked through in its own method note, with the math and a worked example. [Read the method notes →](https://lendriskanalytics.com/tools.html) Independent market research. Not investment, legal, or accounting advice. LR LendRisk Analytics Independent market research Continue reading [Vol · 04 · Teardown What the *Tricolor* collapse actually says about deep subprime auto](https://lendriskanalytics.com/insights/tricolor.html) [Vol · 03 · Methodology Roll rate analysis: how delinquency migration *actually* works](https://lendriskanalytics.com/insights/roll-rates.html) --- title: "What the Tricolor collapse actually says about deep subprime auto" url: https://lendriskanalytics.com/insights/tricolor.html publisher: LendRisk Analytics kind: Teardown description: "Roll rates were inside covenant the quarter Tricolor failed. The warehouse banks did not miss a number, they missed a slope." html: https://lendriskanalytics.com/insights/tricolor.html --- # What the Tricolor collapse actually says about deep subprime auto [← All insights](https://lendriskanalytics.com/articles.html) Vol · 04 Teardown · 6 min read Teardown # What the Tricolor collapse actually says about deep subprime auto. Roll rates were inside covenant the quarter Tricolor failed. The warehouse banks did not miss a number. They missed a slope. The lender did not have the visibility to see what the bank was seeing, and by the time the line was called, there was no runway left to negotiate. Tricolor Auto Acceptance was a Dallas based subprime auto lender focused on Hispanic borrowers across Texas. At its peak it carried over 800 million dollars in warehouse lending capacity, substantial for a mid sized specialty lender. In 2024 that line was called. The company entered liquidation shortly after, leaving thousands of active borrower relationships to be wound down or transferred. The covenant sequence that led to the collapse is the part worth studying. Warehouse credit agreements in subprime auto typically carry three types of performance triggers. A net charge off rate cap, often set at 8 to 12 percent annualised. A 90+ day delinquency concentration limit, often 4 to 6 percent of the pool. A minimum weighted average FICO floor, typically 560 to 590. Breach any one and the warehouse lender has the contractual right to stop advancing against new loans, sweep the borrowing base, or call the line in its entirety. ## The metric was fine. The slope was not. Tricolor reported sub eight percent 90+ DPD in the quarter before liquidation. The covenant cap was around ten. By the conventional snapshot metric the book was inside the line. The CFO could honestly report compliance to the board. Internal monthly reports showed a portfolio operating within covenant tolerance. What the warehouse bank was tracking was the second derivative. Vintage cohorts originated in late 2022 and early 2023 were not just running above expectations on cumulative net loss. They were running above expectations with an accelerating slope. The terminal velocity of the 60 to 90 day roll rate had been climbing for six consecutive months. Plotted as a runway calculation, the book hit covenant in month nine. Plotted as a snapshot, the book looked fine. **The lesson is not that Tricolor missed a number.** The lesson is that the standard reporting cadence at most subprime auto lenders does not produce the slope. Monthly aging reports give you the level. They do not give you the trajectory. The bank already runs the trajectory calculation on your tape every month. ## Why warehouse banks now act faster. Post Tricolor, warehouse lenders across the subprime auto space have materially tightened monitoring cadence and covenant enforcement. Lines that were reviewed quarterly are now reviewed monthly. Breaches that previously triggered a cure period conversation now trigger line restriction. The implicit tolerance that existed in 2021 and 2022, when nearly every subprime portfolio was performing well due to stimulus effects, has been fully withdrawn. The bank does not need to wait for a hard covenant breach to act. A demonstrated trajectory toward breach is increasingly enough for an aggressive review. Lenders who have not updated their covenant monitoring infrastructure to match this enforcement environment are operating with an outdated risk model. ## What to do about it. Build the runway view yourself. Take your current net charge off rate. Add your monthly rate of change. Project both numbers forward 18 months. Compare against your covenant cap. The month the projected number crosses the cap is your runway. Do the same for 90+ DPD and weighted average FICO. If your runway on any metric is under 12 months, your warehouse bank already knows that. They are already modeling it. The strongest position you can be in is the one where you walk into a covenant conversation with your own runway calculation, your own corrective plan, and your own dealer level analysis of where the deterioration is concentrated. The weakest position is the one where you find out about it from them. Method note Run the same runway calculation on your portfolio [Read the method note →](https://lendriskanalytics.com/covenant.html) The math is not complicated. The discipline of running it every month, against every covenant, on every cohort, is what separates the lenders who get blindsided from the lenders who get ahead of the call. LR LendRisk Analytics Independent market research Continue reading [Vol · 03 · Methodology Roll rate analysis: how delinquency migration *actually* works](https://lendriskanalytics.com/insights/roll-rates.html) [Vol · 01 · Market analysis Subprime 60+ DPD hits *6.90%* in Q1 2026](https://lendriskanalytics.com/insights/q1-2026-dpd.html) --- title: "Roll rate analysis: how delinquency migration actually works" url: https://lendriskanalytics.com/insights/roll-rates.html publisher: LendRisk Analytics kind: Methodology description: "A plain English walk through transition matrices, why warehouse banks know your 90 day DPD before you do, and how to build the same view yourself." html: https://lendriskanalytics.com/insights/roll-rates.html --- # Roll rate analysis: how delinquency migration actually works [← All insights](https://lendriskanalytics.com/articles.html) Vol · 03 Methodology · 5 min read Methodology # Roll rate analysis: how delinquency migration actually works. Every warehouse bank that funds a subprime auto book runs the same projection on your tape every month. This is what they are doing, why it tells them where you are headed two to three months before your aging report does, and how to build the same view yourself in a single spreadsheet. Roll rate analysis measures how loans migrate between delinquency states from one month to the next. The core mechanics are simple. Take every loan in the 30 to 59 days past due bucket at month start. At month end, categorise where each one went. Either it cured back to current, stayed in 30 to 59, rolled forward to 60 to 89, or rolled all the way to charge off. The percentage that moved from each state to each other state is the roll rate. Run that across every bucket and you have a transition matrix. Mathematically this is a Markov chain. It represents the probability that a loan in state i at time t will be in state j at time t plus one. Multiply your current state distribution vector by the matrix and you get the expected distribution one period forward. Multiply again and you get two periods forward. This is how warehouse bank credit teams project your 90 day DPD exposure from today's data, and why they often know your portfolio's two month forward trajectory before you do. ## The two rates that matter most right now. In the current environment, the roll rates that drive everything are the 30 to 60 transition and the 60 to 90 transition. The 30 to 60 historically ran 35 to 40 percent in subprime auto. Roughly a third of 30 day delinquent loans cure, a third stay in the bucket, and a third roll worse. In stressed environments like the current one, that figure pushes toward 45 to 50 percent. The 60 to 90 transition is stickier. In stable conditions it runs 50 to 60 percent. In stress it pushes toward 65 to 70. Loans that reach 60 days past due rarely recover. The borrower is structurally distressed by that point, not just timing out a temporary cash flow gap. **The compounding matters.** When both transitions move simultaneously, the effect on 90+ DPD is multiplicative, not additive. A 5 percentage point increase in the 30 to 60 rate combined with a 5 percentage point increase in the 60 to 90 rate does not move 90+ DPD by 10 percent. It moves it by closer to 25 percent over a two month horizon. ## What this means for your reporting cadence. A lender whose 30 day bucket grew by 0.5 percentage points in January should expect 90+ DPD to grow by roughly 0.2 to 0.25 percentage points by March, assuming historical roll rates hold. If roll rates are elevated, which they are right now, that projection is conservative. The actual move could be larger. Running this analysis monthly gives you 60 to 90 days of forward visibility on the metrics warehouse banks are watching. Waiting for the 90+ DPD number to move in your aging report before taking action means responding to a problem that started two to three months earlier. By then your dealer mix, origination quality, and collection priorities should already have been adjusted, not just about to be adjusted. ## How to build it without a credit team. Pull two consecutive monthly snapshots of your active book. Tag each loan with its bucket at month start and its bucket at month end. Pivot one against the other. The resulting nine cell matrix is your transition matrix. Save the matrix every month. After three months you have the trailing average roll rates. After six months you have enough data to spot regime changes. Apply your latest transition matrix to your current state distribution. The result is your projected one month forward distribution. Apply it again and you get two months forward. This is the runway view the bank already has on you. Building it costs you one afternoon of work per month. Method note Project your covenant runway 18 months forward [Read the method note →](https://lendriskanalytics.com/covenant.html) The advantage of doing the analysis yourself is not just the projection. It is the ability to spot when your roll rates change before the change shows up in your headline metrics. Roll rates are the leading indicator. Aging reports are the lagging indicator. The bank works from the leading indicator. The lenders who get caught flat footed work from the lagging one. LR LendRisk Analytics Independent market research Continue reading [Vol · 04 · Teardown What the *Tricolor* collapse actually says about deep subprime auto](https://lendriskanalytics.com/insights/tricolor.html) [Vol · 01 · Market analysis Subprime 60+ DPD hits *6.90%* in Q1 2026](https://lendriskanalytics.com/insights/q1-2026-dpd.html) --- title: "BHPH charge offs in 2026: what normal actually looks like" url: https://lendriskanalytics.com/insights/bhph-normal.html publisher: LendRisk Analytics kind: Sector note description: "A 20 percent annual charge off rate is normal for a properly priced BHPH book. The operators going under are the ones whose recovery model assumed 2021 vehicle prices." html: https://lendriskanalytics.com/insights/bhph-normal.html --- # BHPH charge offs in 2026: what normal actually looks like [← All insights](https://lendriskanalytics.com/articles.html) Vol · 02 Sector note · 4 min read Sector note · BHPH # BHPH charge offs in 2026: what normal actually looks like. A 15 to 25 percent annual charge off rate sounds catastrophic. For a traditional prime or near prime auto lender, it is. For a buy here pay here operation with a properly constructed pricing model, it is normal and expected. The operators going under right now are not the ones with high losses. They are the ones whose recovery model was calibrated to a vehicle market that no longer exists. The buy here pay here model is structurally different from indirect subprime lending. The dealer is the lender. The lender is the dealer. Credit losses are not a surprise. They are priced directly into the interest rate charged, the required down payment, and the collateral haircut applied at origination. A BHPH operator running a 20 percent annualised charge off rate is not in distress. That number is baked into the business model. The business does not fail because it charges off. It fails when one of two things happens. Either the actual charge off rate drifts meaningfully above what the pricing model assumed, or the recovery rate on repossessions falls below what the loss reserve was built to absorb. In 2026, the second of those is the dominant failure mode, and it is not a credit problem. It is a collateral problem. ## Vehicle prices broke the recovery model. BHPH operators work with a specific slice of the used vehicle market. Older, higher mileage units in the 8,000 to 15,000 dollar range. This slice has declined proportionally more than the late model used vehicle market tracked by Manheim. The reason is structural. Late model used inventory rebuilt quickly after the chip shortage cycle ended. Older inventory did not. The depreciation curve that was suppressed in 2021 and 2022 caught up all at once. A repossession that would have cleared 7,500 dollars at auction in 2022 is clearing 5,500 to 6,000 dollars in 2026. After repossession costs, transportation, auction fees, and basic reconditioning, the net recovery on that loan dropped from around 5,500 dollars to closer to 3,500. That difference comes directly out of the recovery rate assumption, which flows directly into net loss. **The number that matters is not your charge off rate. It is your loss given default.** Loss given default equals one minus your recovery rate. If your model assumed 50 percent recovery and you are now running 35 percent, your loss given default went from 50 percent to 65 percent. On a 20 percent charge off rate, that is the difference between a 10 percent net loss and a 13 percent net loss. The book that priced for 10 stops penciling at 13. ## Why this is invisible to most operators. Most BHPH operators track aggregate charge off rate and aggregate cash collections. Those two numbers are insufficient. They tell you about the level of losses. They do not tell you about loss severity, which is the metric driving the actual deterioration. To see what is happening you need to track repossession recovery rate separately, by vintage quarter, and compare it against your underwriting assumption. The operators who have built durable books through this cycle have one thing in common. They track recovery rate as a first class metric, by origination quarter, every month. When recovery rates compressed in early 2025 they saw it in their own data, repriced new originations to a higher down payment requirement, and tightened collateral standards on new loans. The operators who did not track it are now seeing the impact in cash flow without having time to reprice the inflowing book. ## What to do about it. Pull every charge off from the last 24 months. Tag each by origination quarter. Calculate the recovery rate, defined as net auction proceeds divided by remaining principal balance at the time of repossession. Plot it by quarter. Compare against the recovery assumption your pricing model used. If your actual recovery rate is more than five percentage points below your underwriting assumption, your pricing model is broken for current originations. Either raise down payment requirements, shorten loan terms, or tighten the vehicle class you finance. The book you are writing today should reflect the recovery environment of today, not 2022. Method note Calibrate your portfolio against current benchmarks [Read the method note →](https://lendriskanalytics.com/tool.html) The BHPH model is uniquely resilient because the operator controls both sides of the transaction. That resilience only holds if the pricing model stays calibrated to the current environment. The cycle change is the moment that calibration matters most, and the moment most operators miss it. LR LendRisk Analytics Independent market research Continue reading [Vol · 04 · Teardown What the *Tricolor* collapse actually says about deep subprime auto](https://lendriskanalytics.com/insights/tricolor.html) [Vol · 03 · Methodology Roll rate analysis: how delinquency migration *actually* works](https://lendriskanalytics.com/insights/roll-rates.html) --- title: "Subprime 60+ DPD hits 6.90% in Q1 2026" url: https://lendriskanalytics.com/insights/q1-2026-dpd.html publisher: LendRisk Analytics kind: Market analysis description: "Eighteen consecutive quarters of deterioration. The aggregate number is not what should worry you. The shape of the vintage curves underneath it is." html: https://lendriskanalytics.com/insights/q1-2026-dpd.html --- # Subprime 60+ DPD hits 6.90% in Q1 2026 [← All insights](https://lendriskanalytics.com/articles.html) Vol · 01 Market analysis · 4 min read Market analysis # Subprime 60+ DPD hits 6.90% in Q1 2026. Fitch reported the highest reading on record in a series that goes back to 1994. The aggregate number is alarming, but it is not the part you should be reading first. The shape of the vintage curves underneath the headline is the actual signal, and it has been getting steeper for eighteen consecutive quarters without a single quarter of improvement. 6.90% 60+ DPD Q1 2026 3.20% Stimulus low Q3 2021 +116% Rise vs 2021 level The headline is 6.90 percent. That is the highest subprime auto 60+ DPD reading on record in Fitch's series, which goes back to 1994. The number alone does not tell the full story. The trajectory does. This rate did not spike from an external shock. It climbed quarter after quarter from the post stimulus low of 3.2 percent in Q3 2021 without a single quarter of reversal in between. Eighteen consecutive quarters of deterioration is not a cycle. It is a structural shift. ## Why the aggregate hides the actual problem. The 6.90 percent figure masks significant distribution within it. A portfolio average can look manageable while specific dealer channels, geographic concentrations, or FICO tiers are running at 10 to 12 percent 60+ DPD. Aggregate reporting gives you the average. Vintage level reporting tells you which part of your book is driving it. The vintage level data is more concerning than the aggregate. Cohorts originated in 2022 and early 2023 are seasoning into their worst performance at exactly the wrong time. Vehicle values fell 20 to 30 percent from pandemic peaks, which compresses recovery rates on repossessions. When charge off frequency rises and recovery rates fall simultaneously, net loss acceleration follows. The cohorts written in that window are the ones currently driving the headline number higher. **The trajectory matters as much as the level.** At the current rate of deterioration, roughly 0.15 to 0.20 percentage points per quarter, the subprime 60+ DPD rate reaches 7 percent by Q3 2026 absent a meaningful reversal. No macro indicator currently visible suggests that reversal is coming. ## What the underlying story is. Unemployment remains low. Wage growth has decelerated meaningfully since 2022. The borrowers most exposed to subprime auto stress are those for whom cost of living increases have already exhausted the cushion built during the stimulus era. Their disposable income did not keep pace with auto payments structured at peak vehicle values and elevated post pandemic interest rates. The macro question is not whether unemployment rises. It is whether wage growth at the bottom of the income distribution recovers faster than inflation continues to compress household budgets. Until that gap closes, the borrower base under most subprime auto books continues to deteriorate independent of broader labour market data. ## What to do about it. Stop reading the aggregate number first. Build a vintage cohort view of your own book. Group your active loans by origination quarter. Calculate cumulative net loss as a percentage of original balance for each quarter at the same months on book. Plot the curves. The cohorts where the curve at month 18 is meaningfully above the cohorts at month 12 are the ones telling you the story. If your 2022 and 2023 cohorts are tracking 100 to 200 basis points above older vintages at the same seasoning point, you are observing the same deterioration the aggregate Fitch number is reporting, with the advantage of seeing it on your specific book rather than the industry composite. That visibility is the precondition for taking specific action on specific dealer channels rather than blunt portfolio wide retrenchment. Method note See where your portfolio stands against the benchmark [Read the method note →](https://lendriskanalytics.com/tool.html) The 6.90 percent reading is what the market is. It is not what your book is. Knowing the difference, and acting on the difference at the cohort level rather than the portfolio level, is the entire job. LR LendRisk Analytics Independent market research Continue reading [Vol · 03 · Methodology Roll rate analysis: how delinquency migration *actually* works](https://lendriskanalytics.com/insights/roll-rates.html) [Vol · 02 · Sector note BHPH charge offs in 2026: what *normal* actually looks like](https://lendriskanalytics.com/insights/bhph-normal.html) --- title: "The Negative Equity Machine" url: https://lendriskanalytics.com/insights/negative-equity-machine.html publisher: LendRisk Analytics kind: Page description: "The 32-year subprime delinquency record is being read as a borrower-quality problem. The data says it is a loan-structure problem: negative equity underwritten into 77-month loans on collateral that depreciates faster than the note amortizes." html: https://lendriskanalytics.com/insights/negative-equity-machine.html --- # The Negative Equity Machine **Slide series · 6 cards.** Screenshot each card and post in order, or share the whole thread. Live at [lendriskanalytics.com](https://lendriskanalytics.com/insights/negative-equity-machine.html) Lend Risk Analytics 01 / 06 Subprime Auto · 2026 Finding # Subprime didn't get worse borrowers. It got longer loans. Delinquency just hit a **32-year record.** Everyone is calling it a credit problem. The data says it is a loan-structure problem, and the difference is everything. Lend Risk Analytics 02 / 06 The number everyone is quoting ## A 385-month high. Back to January 1994. Subprime auto 60+ day delinquency, and the loss behind it, are both at post-pandemic peaks. But look at the last box. 6.90% Subprime 60+ DPD Jan 2026 · record · was 6.45% yr ago 9.81% Annualized net loss Jan 2026 · post-pandemic high 16.4× Subprime vs prime DPD prime sits at 0.42% 37% Recovery rate was 44% pre-pandemic ▼ Source · Fitch subprime auto ABS index lendriskanalytics.com Lend Risk Analytics 03 / 06 The mechanism ## The term outran the metal. **90% of negative-equity loans now run 72+ months. The average is 77.** A 77-month note amortizes slower than the car depreciates, so the borrower is underwater straight through the window where defaults cluster. What they owe What the car is worth Where defaults hit Source · Edmunds Q1 2026 · illustrative structure lendriskanalytics.com Lend Risk Analytics 04 / 06 Why it compounds ## It doesn't reset. It digs deeper. 1 31% of trade-ins are underwater The average is 4.3 years old and carries **$7,183** in negative equity, a record, up 42% in five years. ↓ 2 The $7,183 rolls into the next loan To keep the payment livable, the term stretches to 77 months. The new loan starts thousands above a car that is already depreciating. ↓ 3 They surface even later, or never 26% now roll more than **$10,000** of old debt forward. Each cycle the hole gets deeper, not shallower. Source · Edmunds Q1 2026 lendriskanalytics.com Lend Risk Analytics 05 / 06 The proof it is structure, not credit ## Severity gave it away. If this were only weaker borrowers, defaults would rise but recoveries would hold. Instead **recoveries are collapsing.** That only happens when the collateral was never worth the loan. Recovery rate · pre-pandemic 43.7% Recovery rate · today 37.0% So loss-given-default rose to 63¢ / $1 When a 77-month loan on an old car fails at month 30, the lender eats **63 cents on every dollar.** The term wrote the loss in on day one. Source · Fitch subprime auto ABS index lendriskanalytics.com Lend Risk Analytics 06 / 06 If you hold subprime paper ## Three things to do before your next vintage. 1 **Stop underwriting the FICO. Underwrite the structure.** Term length on collateral age predicts loss better than the score on a deep-subprime book. 2 **Cap term by vehicle age.** A 77-month note on an 8-year-old car is a guaranteed negative-equity window. Match the term to the metal. 3 **Track loss-given-default, not just delinquency.** Falling recoveries are the early warning your DPD rate hides. LendRisk Analytics · Independent market research lendriskanalytics.com Sources & method **Fitch Ratings** subprime auto ABS index (via Auto Remarketing, May 2026): 60+ DPD 6.90% Jan 2026 (record, vs 6.45% yr ago); annualized net loss 9.81%; recoveries 37.0% TTM vs 43.73% pre-pandemic; prime 60+ DPD 0.42%. **Edmunds** Q1 2026 insights: 30.9% of trade-ins underwater; average negative equity $7,183 (+42% in 5 yrs); 90.2% of negative-equity loans 72+ months, 43% at 84 months, average term 77.4 months; average underwater trade-in age 4.3 years; 26% roll more than $10,000. The balance-versus-value chart is illustrative of the structure, not a per-loan plot. Subprime-vs-prime ratio (16.4x) computed from the two cited DPD figures. Loss-given-default (63%) is the complement of the 37% recovery rate. Independent analysis, not affiliated with or representing any employer. --- title: "The credit strength Washington is trying to outlaw" url: https://lendriskanalytics.com/insights/repossession-liability-turn.html publisher: LendRisk Analytics published: 2026-07-10 kind: Original analysis description: "The Fed's May 2026 BHPH note treats a 16.63x-higher repossession rate as a credit strength, faster recovery, lower loss-given-default, and $2B+ of bank commitments rated lower risk. Warren's February probe treats the same act as consumer harm. Both readings cannot hold. What happens to the banks' LGD assumption when the regulatory cost of repossession rises." html: https://lendriskanalytics.com/insights/repossession-liability-turn.html --- # The credit strength Washington is trying to outlaw [← All insights](https://lendriskanalytics.com/articles.html) Sector note · The LGD contradiction Original analysis · 11 min read Where two public readings collide # The credit strength Washington is trying to outlaw. Public record through July 10, 2026 · LendRisk Analytics In May, the Federal Reserve published a note explaining why banks are comfortable lending against buy-here-pay-here paper. Its answer, in part: BHPH operators repossess cars *16.63 times* more often than traditional lenders, and that fast, aggressive recovery lowers loss-given-default, which is why banks rated more than $2 billion of these commitments as *lower* risk. Three months earlier, a Senate probe called that same repossession behavior "inexcusable" and demanded to know how often it is done in error. Two arms of the public record are looking at the identical act and reaching opposite conclusions. They cannot both be right, and the gap between them is a credit assumption nobody has priced. 16.63× BHPH loans more likely to be in active repossession vs traditional lenders (Fed, 2025:Q3) $2B+ Bank commitments the Fed says were rated lower risk than loans to traditional auto dealers 12 Industry recipients of Warren's Feb 5, 2026 repossession probe 1.73M Vehicles repossessed in 2024, most since 2009 (Cox / Experian) This is not a story about a single operator. It is about a single assumption that sits underneath the whole bank-to-BHPH funding stack, and about the fact that the assumption is now the subject of a federal investigation. The best way to see it is to read what each side actually wrote. ## What the Fed actually said The Fed's May 8, 2026 FEDS Note, *Subprime Auto Lending: Trends in Buy Here Pay Here Auto Lending*, was written into the wreckage of the Tricolor bankruptcy, the note describes loans to BHPH borrowers seeing their reported probability of default rise "by nearly 150% from the second to third quarter of 2025." That is the credit-quality half of the story, and it is deteriorating. But the note goes out of its way to explain a countervailing strength, and the mechanism it names is repossession. "One such loss mitigation strategy is auto repossession, in which BHPH dealer behavior appears to noticeably differ from that of traditional auto lenders." Federal Reserve FEDS Note · May 8, 2026 How different? The note quantifies it precisely: "BHPH loans are 16.63 times more likely to be in active repossession status." In the third quarter of 2025, "approximately 5% of BHPH balances were in active repossession, compared to less than half a percent for traditional auto lender balances." Repossession is not an edge case in this model. It is the loss-mitigation engine. And here is the sentence that matters most, the one that turns an operational habit into a bank credit input: "The repossession and subsequent vehicle sale could reduce loss-given-default for the consumer auto loans, which should translate to reduced credit risk for bank lending to BHPH dealers when these consumer auto loan receivables serve as collateral." Federal Reserve FEDS Note · May 8, 2026 Read that carefully, because it is the entire argument. The consumer loans are the collateral behind the bank's line to the dealer. If those loans default, what protects the bank is how fast and how cheaply the dealer can turn the car back into cash. A high, frictionless repossession rate *is* the recovery assumption. The Fed then reports the consequence at the bank level: "over $2 billion in loan commitments that we identify were rated by these large banks as being of lower risk compared to loans to traditional auto dealers." Roughly 78% of BHPH volume goes to subprime borrowers, against 27% for traditional lenders, and the banks still rated the exposure as safer, because they were pricing the recovery, not the borrower. The banks did not misjudge the borrower. They priced a repossession regime, and that regime is what is now under investigation. ## What Washington is doing to the same act On February 5, 2026, the ranking member of the Senate Banking Committee opened a formal probe into auto repossession practices, sending letters to a dozen recipients. The list is not incidental. It is a map of exactly the operators the Fed's collateral assumption depends on: **America's Car-Mart, DriveTime, Byrider, and CarHop** among the BHPH names, alongside Chase Auto, Capital One, Toyota Financial, GM Financial, and Ally, plus the industry bodies AFSA, the American Recovery Association, and NIADA. The framing is the opposite of the Fed's. Where the Fed sees a loss-mitigation strength, the probe sees a consumer-harm problem to be measured and curbed: "Car repossession is a devastating disruption to someone's life, and it is inexcusable when that repossession is in error." Sen. Warren, Senate Banking Committee (minority) · Feb 5, 2026 The probe's stated targets are error rates, illegal and mistaken repossessions, cars seized while the borrower is current or has an agreement in place, and the practices around them, sent at a moment when the letter argues the CFPB's capacity to police those errors has been deliberately weakened. It lands against a backdrop the same reporting supplies: 1.73 million vehicles repossessed in 2024, the most since 2009, and a subprime auto delinquency rate that reached 6.74% in December, the highest in records going back to the early 1990s (Fitch). Repossession is rising, and so is the political cost of doing it aggressively. These two documents are describing one behavior. The Fed calls the 16.63× a reason to rate the credit lower-risk. The Senate calls it a reason to open an investigation. That is not a nuance. It is a direct contradiction in how the same public record values the same act. ## Why both readings cannot hold The Fed's LGD benefit is not free-standing. It is entirely a function of the *cost* of repossession, how many days from default to recovered vehicle, how much friction and legal expense per repossession, and how confident the operator is that the seizure sticks. Cheap, fast, unchallenged repossession is what produces the low LGD. That cost structure is precisely what every consumer-protection lever raises: **Right-to-cure and notice periods** add days between default and lawful seizure, and in deep subprime, days are the whole game. **Wrongful-repossession liability** converts a recovery into a loss plus a penalty plus a reserve against the next one. **Restrictions on the enforcement tooling**, GPS trackers and starter-interrupt devices, which are how many BHPH operators keep repossession cheap, raise the marginal cost of every recovery. **Redemption and reinstatement rights**, already embedded in UCC Article 9 and expanded by many state statutes, give the borrower more paths to pull the car back out of the pipeline. Each one is individually modest. Together they move exactly the variable the Fed's low LGD is built on. **Inference, labeled** This is the analytical claim of the piece, not something either document states: if the regulatory cost of aggressive repossession rises, the LGD the banks priced rises with it, and the "lower risk" rating on that $2B+ of commitments is the thing that re-rates. The Fed already shows the probability-of-default half moving, up nearly 150% post-Tricolor. The loss-given-default half is the part still being carried at the favorable number, and it is the part now under federal probe. Correlation and mechanism, not a prediction of any specific rating action. LendRisk's own [waiting-tax](https://lendriskanalytics.com/insights/the-waiting-tax.html) work makes the mechanism concrete: recovery is collateral value multiplied by the probability you actually get the car back, and in deep subprime the second number collapses far faster than the car depreciates. Every friction the probe would add, cure periods, wrongful-repo liability, restrictions on the tooling that keeps repossession cheap, pushes on that second number, which pushes on LGD. Neither the Fed nor the Senate publishes a recovery-cost curve. The chart below is not that curve. It is a hypothetical, built with assumed inputs, showing only the shape of what happens to recovery and LGD if repossession friction rises, an illustration of the mechanism, not a measurement of it. ⚠ Illustrative example · not real data Hypothetical, recovery and LGD if repossession friction rises Assumed inputs only, invented for this illustration and not drawn from any filing, study, or model. Shows the direction and shape of the effect described above, not a forecast or an actual figure. Recovery = collateral value × probability of recovery; LGD = 1 − recovery, before carry cost. ** Recovery rate (% of balance), assumed ** Implied loss-given-default, assumed * All figures on this chart are invented for illustration and do not appear in the Fed note, the Senate probe, or any other source cited on this page. Nobody has published a real recovery-cost curve for BHPH repossession friction, this is a stand-in showing what the mechanism described above would look like if someone did, not what the numbers actually are. Method note · runs in your browser Skip the hypothetical, model what added repossession friction costs your own recovery rate [Read the method note →](https://lendriskanalytics.com/repo-timing.html) ## The number that has to move Put the two documents side by side and the tension is exact. | The same act | The Fed's reading (May 2026) | The probe's reading (Feb 2026) | |---|---|---| | 16.63× repossession rate | Loss-mitigation strength | Evidence to investigate | | Fast, low-friction recovery | Lowers LGD → lower credit risk | Where errors and illegal seizures hide | | GPS / starter-interrupt tooling | Implicit in the cheap-recovery assumption | Practices under scrutiny | | Net effect on the bank line | $2B+ rated lower risk | Rising regulatory cost, unpriced | Left column is the identical behavior. The two right columns are both public, both official, and point in opposite directions. Nobody has published the reconciliation, because there isn't one, one of the two valuations gives. None of this asserts that any bank will be downgraded, that any operator is failing, or that the probe will produce a rule. It asserts something narrower and, for a lender, more useful: the favorable LGD assumption underneath the bank-to-BHPH stack rests on a repossession regime that is now a live political target, and the cost of that regime moves in one direction under every plausible intervention. The Fed priced the strength. It did not price the fragility of the thing that produces the strength. ## Run this on your own book The Fed did this read at the sector level and stopped at "recovery is a strength." The same move runs on a single book, a method note on this site walks through it. **Re-price recovery under friction.** The favorable LGD is a function of days-to-recover and cost-per-repossession. Push both up, +30 days, +60 days, higher legal and resale cost, and recovery falls before a single loan goes bad, the same shape as the hypothetical above, worked through in the [method note](https://lendriskanalytics.com/repo-timing.html). None of this asserts any bank is downgraded or any operator is failing. It asserts something narrower and more useful to a desk: the favorable LGD under the whole bank-to-BHPH stack rests on a repossession regime that is now a live political target, and the cost of that regime moves one direction under every plausible intervention. The Fed priced the strength and stopped. Reading the other half, and running it on your book before someone else runs it for you, is the entire job. See it on a real book The same read, run end-to-end on a synthetic $40M book. [Open the worked example →](https://lendriskanalytics.com/sample-report.html) **Sources & notes** The 16.63× repossession multiple, the ~5% vs <0.5% active-repossession share (2025:Q3), the "loss mitigation strategy" and "reduce loss-given-default … reduced credit risk for bank lending to BHPH dealers" language, the "over $2 billion in loan commitments … rated … as being of lower risk," the 78% vs 27% subprime shares, and the "nearly 150%" post-Tricolor probability-of-default increase are all quoted or drawn from the Federal Reserve FEDS Note, *Subprime Auto Lending: Trends in Buy Here Pay Here Auto Lending*, published May 8, 2026, at [federalreserve.gov](https://www.federalreserve.gov/econres/notes/feds-notes/subprime-auto-lending-trends-in-buy-here-pay-here-auto-lending-20260508.html). The February 5, 2026 auto-repossession probe, the twelve recipients (CarHop, DriveTime, Byrider, America's Car-Mart, Chase Auto, Capital One, Toyota Financial Services, GM Financial, Ally Financial, AFSA, the American Recovery Association, and NIADA), the "inexcusable when that repossession is in error" quote, and the focus on error rates and illegal/mistaken repossessions are from the Senate Banking Committee (minority) release at [banking.senate.gov](https://www.banking.senate.gov/newsroom/minority/with-trump-sidelining-cfpb-warren-launches-probe-into-the-auto-lending-industry-as-car-repossessions-skyrocket), corroborated by [CNN, Feb 5, 2026](https://www.cnn.com/2026/02/05/business/car-prices-repossession-elizabeth-warren). The 1.73 million 2024 repossessions (most since 2009; Cox Automotive / Experian) and the 6.74% December subprime delinquency rate (Fitch) are as reported by CNN in the same piece. Redemption/reinstatement and default rights referenced generally reflect UCC Article 9 and state right-to-cure statutes; no specific state enforcement action is asserted here. **The recovery/LGD chart is a hypothetical illustration with invented inputs, it does not appear in, and is not derived from, the Fed note, the Senate release, or any other source on this page, and is included only to show the shape of the mechanism, not an actual or forecast figure.** **Where this note reasons beyond what a document literally states, chiefly the claim that rising repossession cost raises the banks' priced LGD, it is labeled as an inference.** Point-in-time reading of the public record through July 10, 2026. LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned, and this is not investment, legal, or accounting advice. Read this differently, or catch a number I got wrong? I want to know. [Send it through →](https://lendriskanalytics.com/contact.html). LR LendRisk Analytics Independent market research Continue reading [Servicing · Recovery The waiting tax: the most expensive repo is *the one you didn't make*](https://lendriskanalytics.com/insights/the-waiting-tax.html) [What the Tape Said · Issue 3 What the tape said: *America's Car-Mart*](https://lendriskanalytics.com/insights/carmart-stress-signals.html) --- title: "LendRisk Analytics \u00b7 Independent research on subprime credit" url: https://lendriskanalytics.com/about.html publisher: LendRisk Analytics kind: Page description: "Independent market research on subprime auto and buy-here-pay-here credit. The What the Tape Said series, post-mortems on the lenders that failed, method notes and benchmarks, every figure read from the public record." html: https://lendriskanalytics.com/about.html --- # LendRisk Analytics · Independent research on subprime credit # LendRisk Analytics Independent market research on subprime credit: buy-here-pay-here and subprime auto lenders, the credit unions and banks that fund them, and the securitizations that carry the paper. Every figure is cited to a named public source or labeled synthetic, and the method is printed beside the number. This homepage is a JavaScript application. Every article and page is static HTML with a markdown twin. Start at [/llms.txt](https://lendriskanalytics.com/llms.txt); everything in one file: [/llms-full.txt](https://lendriskanalytics.com/llms-full.txt); text edition index: [/text/index.md](https://lendriskanalytics.com/text/index.md); sitemap: [/sitemap.xml](https://lendriskanalytics.com/sitemap.xml). ## Articles - Modified, and not paying. What the Tape Said, Issue 14 (What the Tape Said, Issue 14). Every federally insured credit union files one line for loans it has modified for borrowers in trouble, and since 2024 a second line for the modified loans that are already late again. The first has more than doubled since the definition changed. One dollar in four on it is not paying. What the line says about next year's charge-offs, where it says nothing, and who leaves it blank. Every figure rebuilt from the raw NCUA archives; inferences labeled. Text: /text/insights-modified-and-not-paying.md - Current, on tape. What the Tape Said, Issue 13 (What the Tape Said, Issue 13). Fourteen auto lenders, forty-four securitizations, 1.15 million loans on their latest monthly tapes. Every one of them grants payment extensions, and every extension turns a past-due account current. Reported 60+ delinquency against the same figure with recently extended loans added back, lender by lender, with what happened to those loans six months later. Every figure recomputed from the filed loan-level records; inferences labeled. Text: /text/insights-current-on-tape.md - What the tape said: the lot behind the branch (What the Tape Said, Issue 12). One line on the NCUA call report counts cars already repossessed and not yet sold. Roughly seven in ten credit unions leave it blank. Across eleven consecutive annual cohorts, the institutions filling it in charged off more the following year, every time. Every figure recomputed from the raw filings; inferences labeled. Text: /text/insights-the-lot-behind-the-branch.md - Past due, unchanged (What the Tape Said, Issue 11). Eleven buy-here-pay-here stores, thirty months of real monthly numbers. The share of customers behind held steady. What each dollar of loss cost in interest moved, bottomed in late 2025, and is climbing back. What these operators did about it. Text: /text/insights-past-due-unchanged.md - What the tape said: the one-way door (What the Tape Said, Issue 10). Credit unions began securitizing their own loans in November 2019. Twenty-five deals and roughly $8.4 billion later, the dollars are still small. The structure is not. Why a funding tool adopted under liquidity stress has not been retired now that the stress has eased. Every figure sourced; inferences labeled. Text: /text/insights-the-one-way-door.md - The cash problem (What the Tape Said, Issue 9). In buy-here-pay-here, cash leaves the day you sell the car and comes back over four to five years. Seven failures from 2023 to 2026, re-read as cash events: American Car Center, U.S. Auto Sales, Tricolor, PrimaLend, Automotive Credit, FinBe, and America's Car-Mart. Every figure sourced; inferences labeled. Text: /text/insights-the-cash-problem.md - What the tape said: current, on paper (What the Tape Said, Issue 8). An account gets an extension. The past-due clock resets. The tape shows current. Nothing about the borrower has changed. Where that gap lives in the SEC record, what the Philadelphia Fed found, and the disclosure failure that cost America's Car-Mart a non-reliance finding. Every figure sourced; inferences labeled. Text: /text/insights-current-on-paper.md - What the tape said: the future of subprime (What the Tape Said, Issue 7). A forward read on subprime auto. Severity has moved into origination and become forecastable, the industry benchmark is dissolving under composition drift, and the verification layer will be built by a rating agency, a consortium or a vendor. What each outcome costs the operators being measured. Every figure sourced; inferences labeled. Text: /text/insights-the-future-of-subprime.md - What the tape said: three stress cycles, one missing layer (What the Tape Said, Issue 6). A deep study of the U.S. macroeconomy and subprime auto across 1997-98, 2008-09, and 2022-26. Three macro regimes, the same three causes of death, and the measurement layer the sector was told to build in 1998 and still has not. Every figure sourced; inferences labeled. Text: /text/insights-three-cycles-one-missing-layer.md - What the tape said: Tricolor, PrimaLend, Car-Mart (What the Tape Said, Issue 5). Three subprime auto lenders failed or nearly failed in nine months, funded by JPMorgan, Fifth Third, Barclays, CIBC and Silver Point. A comparative postmortem on what each institution actually missed, and why the answer is different in all three cases. Every figure sourced; inferences labeled. Text: /text/insights-three-failures-one-blind-spot.md - What the tape said: Credit Acceptance (CACC) (What the Tape Said, Issue 4). A stress-signal snapshot on Credit Acceptance: an 8.2-point forecast miss on the 2022 vintage, the worst in a decade, against ABS funding costs that fell from 8.6% to 5.1% over the same two years. Deterioration and stabilization, read from the same tape. Every figure sourced; inferences labeled. Text: /text/insights-cacc-stress-signals.md - What the tape said: America's Car-Mart (What the Tape Said, Issue 3). A read of what is publicly visible about America's Car-Mart right now: a $300M distressed-fund term loan, a June 2026 forbearance covering five simultaneous covenant defaults, an $18M waiver fee, and 66 days on the clock. The credit book was improving; the funding architecture is what broke. Straight from the filings. Text: /text/insights-carmart-stress-signals.md - What the tape said: CarMax Auto Finance (What the Tape Said, Issue 2). A read of what is publicly visible about CarMax Auto Finance right now: a $71.3M lifetime-loss revision on 2022 and 2023 vintages, an allowance that climbed after management called the peak, and a nonprime shelf whose structure is quietly tightening. Straight from the filings, with every figure sourced. Text: /text/insights-carmax-stress-signals.md - What the tape said: Tricolor Holdings (What the Tape Said, Issue 1). A post-mortem on what was publicly visible about Tricolor Holdings before its September 2025 collapse: funding dependence, a thin-file collateral pool (62% no credit score in the final deal), and diligence blind spots, read straight from the public filings. Text: /text/insights-tricolor-tape.md - The waiting tax: the most expensive repo is the one you didn't make . In deep subprime, the largest controllable loss isn't the auction price. It's the recovery you forfeit by hesitating, because the odds of getting the car back collapse far faster than the car depreciates. Text: /text/insights-the-waiting-tax.md - Buy-here-pay-here grew up . BHPH used to be cash-funded and self-insured. The Federal Reserve's 2026 data shows it is now a bank-financed, guarantor-backed asset class, and institutionalization imports finance-company fragility into a segment that used to absorb its own losses. Text: /text/insights-bhph-institutionalized.md - Early payment default isn't a credit event. It's a fraud signal. . In deep subprime, a loan that defaults in the first three payments rarely went bad. It started bad. EPD is the fingerprint of misrepresentation at origination, and it clusters in a handful of dealers. Text: /text/insights-early-payment-default.md - Run the book: a tool-by-tool teardown . One blinded $40M subprime auto book, run through all six LendRisk Analytics tools in sequence. Every input, every output, every decision, from 'you look compliant' to a covenant breach six months out, the three dealers causing it, the deal to decline, and the next bad dealer stopped at the door. Text: /text/insights-run-the-book.md - Anatomy of a toxic book: how an average-looking portfolio hides a covenant breach . A blinded case study. The aggregate said compliant. Attribution said six months to a covenant breach. How three dealers and one vintage nearly swept a $40M warehouse line, and how the slope was visible before the damage. Text: /text/insights-toxic-book.md - What the Tricolor collapse actually says about deep subprime auto . Roll rates were inside covenant the quarter Tricolor failed. The warehouse banks did not miss a number, they missed a slope. Text: /text/insights-tricolor.md - Roll rate analysis: how delinquency migration actually works . A plain English walk through transition matrices, why warehouse banks know your 90 day DPD before you do, and how to build the same view yourself. Text: /text/insights-roll-rates.md - BHPH charge offs in 2026: what normal actually looks like . A 20 percent annual charge off rate is normal for a properly priced BHPH book. The operators going under are the ones whose recovery model assumed 2021 vehicle prices. Text: /text/insights-bhph-normal.md - Subprime 60+ DPD hits 6.90% in Q1 2026 . Eighteen consecutive quarters of deterioration. The aggregate number is not what should worry you. The shape of the vintage curves underneath it is. Text: /text/insights-q1-2026-dpd.md - The Negative Equity Machine . The 32-year subprime delinquency record is being read as a borrower-quality problem. The data says it is a loan-structure problem: negative equity underwritten into 77-month loans on collateral that depreciates faster than the note amortizes. Text: /text/insights-negative-equity-machine.md - The credit strength Washington is trying to outlaw . The Fed's May 2026 BHPH note treats a 16.63x-higher repossession rate as a credit strength, faster recovery, lower loss-given-default, and $2B+ of bank commitments rated lower risk. Warren's February probe treats the same act as consumer harm. Both readings cannot hold. What happens to the banks' LGD assumption when the regulatory cost of repossession rises. Text: /text/insights-repossession-liability-turn.md - What the tape said: the bottom moved more . Used cars are worth a third more than before the pandemic and subprime lenders recover eleven points less on them. The averages hide why. Fifty thousand repossessions from the public loan tapes, split by the year each contract was written: the cohorts that make up nearly all of what is being repossessed are recovering less than a year ago, and the index only looks flat because newer paper is replacing older. Every figure computed from the raw filings. Text: /text/insights-set-at-signing.md ## Methods, tools and data - Reading a loan tape: vintage curves, aging, and attribution . How a loan-level CSV becomes a portfolio read: static-pool vintage loss curves, delinquency aging, dealer attribution, and concentration screens, with the formulas, thresholds, and column vocabulary written out in the open. Text: /text/analyze.md - Articles . Plain English analysis on auto credit performance, vintage loss curves, dealer attribution, covenant monitoring, and market stress. Text: /text/articles.md - The Borrowing-Base Certificate · Lender-Side Market Brief . Tricolor's warehouse lenders had every standard structural protection and still lost roughly $370 million between two banks, because every protection sits downstream of a self-reported borrowing-base certificate. A market brief on the monitoring gap between field exams and the emerging practice of independent monthly recomputation of certificates from loan tapes. Independent research from the public record, not an audit, not advice. Text: /text/borrowing-base-lender.md - The Borrowing-Base Certificate · Market Brief . The monthly borrowing-base certificate is self-reported, and after Tricolor the banking system re-rated the entire BHPH sector because it cannot tell a clean certificate from a fabricated one. This market brief explains what changed, why it runs through one monthly document, and what independent verification of that document would have to look like. Independent market research from the public record, not an audit, not advice. Text: /text/borrowing-base.md - Covenant runway: the month the trigger breaks . The full arithmetic of covenant runway: project each metric linearly from its current level and monthly slope, and the first threshold crossed sets the headline. A method note on why the slope, not the snapshot, is the signal. Text: /text/covenant.md - Data & sources . Every dataset, filing, and publication LendRisk Analytics draws on, government statistics, Federal Reserve research, ratings-agency data, industry reports, and public company disclosures. All public, none proprietary or loan-level. Text: /text/data.md - Reading the dealer channel . A method note on scoring dealer channels in an indirect auto book: loss rate, severe lates, early payment default, and a volume-weighted composite health score, with the full weights, stress caps, and banding published and a worked synthetic example. Text: /text/dealers.md - The segments carrying the loss . A method note on segment-level loss attribution for direct auto books: the bin edges, flag thresholds, and cross-cut logic that show which credit bands, collateral profiles, terms, and geographies are carrying the net loss. Worked examples are synthetic. Text: /text/loss-drivers.md - Method & Data . How LendRisk Analytics builds its method notes and write-ups: where the data comes from, how the synthetic worked examples are constructed, the math behind every metric, and what is illustrative versus market-sourced. Text: /text/method.md - The four levers of recovery, state by state . A method note on the four state-law levers that decide what a defaulted auto loan returns: self-help repossession, right-to-cure notice, deficiency judgments, and wage garnishment. Includes the full scoring weights and how all fifty states and D.C. land. Text: /text/repo-map.md - Recovery decay: what a day of delay costs . A method note on repossession timing in BHPH and deep subprime: expected recovery equals vehicle value times the probability of custody, and the second term collapses far faster than the car depreciates. The full decay model, trigger ledger, and a worked synthetic table. Text: /text/repo-timing.md - Sample Walkthrough · Direct Lender . A blinded direct-lender portfolio walkthrough. No dealers to blame, so loss attribution comes from the credit box itself: which FICO band, term, LTV, vehicle, affordability, and geography are eating the book. We find the slope, rank the segments, project the covenant breach, and tighten the box that stops it. Text: /text/sample-report-direct.md - Sample Walkthrough . A worked example on a synthetic $40M subprime auto book. It shows how to find the slope the aggregate hides, attribute it to three dealers, date it by vintage, and project the covenant breach. Illustrative analysis on made-up data, not a real portfolio. Text: /text/sample-report.md - When the term outlives the car . A method note on matching loan term to vehicle useful life: straight-line depreciation versus amortization, the negative-equity window, and why long terms on aged cars carry a structural path to default. Text: /text/term-matcher.md - The composite read: three numbers against warehouse bands . The full arithmetic of a three-input composite read for auto lending books: annualised loss rate, 90-plus-day delinquency, and weighted-average FICO scored 0-100 against standard warehouse covenant bands, with every threshold and weight published as a reference table. Text: /text/tool.md - Methods · LendRisk Analytics . Ten plain-English method notes on how a subprime auto book gets read, vintage curves, covenant runway, recovery, dealer channel, plus worked examples on synthetic books. Independent market research. Text: /text/tools.md - The economics of a single deal . A method note on single-deal auto-loan economics: how probability of default, loss given default, expected loss, and lifetime ROA combine into a fund, counter, or decline read, with benchmark assumptions and a worked synthetic example. Text: /text/underwriter.md --- title: "Reading a loan tape: vintage curves, aging, and attribution" url: https://lendriskanalytics.com/analyze.html publisher: LendRisk Analytics kind: Page description: "How a loan-level CSV becomes a portfolio read: static-pool vintage loss curves, delinquency aging, dealer attribution, and concentration screens, with the formulas, thresholds, and column vocabulary written out in the open." html: https://lendriskanalytics.com/analyze.html --- # Reading a loan tape: vintage curves, aging, and attribution [← All insights](https://lendriskanalytics.com/articles.html) Methods · Note 01 Method note · 7 min read Methods · Reading the tape # Reading a loan tape: vintage curves, aging, and attribution. LendRisk Analytics · Method note · 2026 A loan-level file is a set of claims about a portfolio. This note publishes the method for testing those claims: how a raw CSV becomes static-pool vintage loss curves, delinquency aging, dealer attribution, and concentration screens. The formulas, thresholds, and column vocabulary below are the complete method, in the open. None of it is exotic, it is the same arithmetic behind every rating-agency pre-sale table, applied one loan at a time. ## What the method needs from the file A tape arrives as one row per loan, and no two servicing systems label their columns the same way. So the first step is mechanical: map whatever headers exist onto a standard field set, matching each header exactly first, then by substring, case-insensitive. Seven fields carry the read, origination date, dealer or source, original balance, delinquency status, charge-off amount, recoveries, and a credit score. Everything else (term, LTV, state, mileage, vehicle age) sharpens the segmentation but is not required. The full matching vocabulary: | Field | Header names accepted | |---|---| | Origination date | orig_date, origination, orig, funded, book_date, contract_date, open_date | | As-of date | as_of, asof, report_date, snapshot, statement_date, data_date | | Dealer / source | dealer, source, originator, seller | | Original balance | orig_balance, original_balance, amount_financed, funded_amount, loan_amount, financed, original_amount | | Current balance | current_balance, balance, principal_balance, outstanding, curr_balance | | Delinquency status | dpd, days_past_due, delinquency, delq, bucket, status, loan_status | | Charge-off | charge_off, chargeoff, co_amount, gross_loss, loss_amount, charged_off | | Recoveries | recovery, recoveries, recovered, recovery_amount | | Credit score | fico, score, credit_score, bureau | | First-default timing | epd, first_default, months_to_first_default, months_to_first_delinquency, mob_first_delq | | LTV | ltv, loan_to_value | | Term | term, term_months, orig_term | | Model year / vehicle age | vehicle_year, model_year, vyear, car_year, year · vehicle_age, veh_age, car_age, age | | Mileage | mileage, miles, odometer | | Collateral type | vehicle_type, collateral, make, body, segment | | State | state, st, geo, region | | Loan identifier | loan_id, loanid, id, account, contract, acct | Bold rows are the core fields the read depends on. This is the actual detection vocabulary, published as method. One quiet assumption sits under everything: the as-of date. Every age calculation runs from origination to the snapshot date, and if the file does not carry one, the reader has to assume today's date, after which every months-on-book figure inherits that guess. ## The definitions, written out Delinquency status is the messiest column on any tape, so it gets parsed permissively. Text containing "charge" (or exactly "co") is charged off; "repo" is repossession; "paid" or "payoff" is closed clean. Anything numeric is days past due: 90 or more is the 90+ bucket, 60 or more is 60-89, 30 or more is 30-59, anything else is current. From there, four definitions do almost all of the work. **Cumulative net loss** is the sum of max(0, charge-off − recoveries), divided by the sum of *originated* balance. Both choices matter: net of recoveries, and divided by originated, never current, balance. Dividing by current balance flatters a shrinking book. Where a loan is marked charged-off or repossessed but no loss amount exists, the current balance stands in as the gross loss (originated balance if that is missing too). **60+ DPD** is the share of active loans, paid-off and charged-off excluded from the denominator, sitting at 60+, 90+, or in repossession. **Months on book** is whole calendar months from origination to the as-of date. **Early payment default** is a first default within three months of origination; when the tape has no first-default column, the proxy is severe delinquency at six or fewer months on book. Averages are balance-weighted throughout. | Measure | Rule as applied | Bands / presets | |---|---|---| | Cumulative net loss | Σ max(0, charge-off − recoveries) ÷ Σ originated balance | quiet ≤8% · watch 8-13% · elevated >13% Fitch reference 9.81% | | 60+ DPD | 60+/90+/repossession ÷ active loans | quiet ≤5% · watch 5-7% · elevated >7% industry reference 6.65% | | Early payment default | first default ≤3 months; proxy: severe status at ≤6 months on book | — | | Dealer flag | net loss >1.5× the median dealer and >10% absolute | "clean" 4% absolute | top 6 by loss share | | Concentration | HHI = 10,000 × Σ (dealer share of balance)²; top-3 share alongside | — | | Credit bins | score at origination | The actual thresholds, bins, and flag rules, published in full. The Fitch and industry reference points are the benchmark figures displayed alongside the same metrics on the original dashboard. ## Vintage curves: hold age constant A single loss number for a whole book is a blend of cohorts at different ages, and blends hide slopes. The static-pool discipline is simple: group loans by origination quarter, compute each cohort's cumulative net loss against its *own* originated balance, and index by months on book rather than calendar date. Losses then become comparable at matched age. Here is the shape the method is built to catch, in a worked example: | Vintage | Loans | Originated | Age (MOB) | Cum net loss | 60+ DPD | EPD rate | |---|---|---|---|---|---|---| | 2023 Q1 | 380 | $6.2M | 14 mo | 4.1% | 3.2% | 1.9% | | 2023 Q3 | 372 | $6.0M | 14 mo | 9.7% | 6.8% | 5.6% | Synthetic and illustrative, these numbers describe the shape of the method, not any real portfolio. Both rows are measured at fourteen months on book, which means the observations come from different calendar dates. The blended book behind this example prints roughly 6.4% cumulative net loss: under the 8% watch line, under the 9.81% Fitch reference, apparently unremarkable. The blend is calm because the mature vintages have flattened and the youngest are too green to show loss yet, while the one vintage that matters is running at more than twice its sibling's pace at identical age. The aggregate is not lying. It is averaging. **Inference**A book whose younger vintages sit above its older vintages at matched age is deteriorating at origination, not merely aging. That is a different problem from a seasoned book working through old loss, and it predicts where the blended number goes as those vintages season. The aggregate cannot make this distinction; the curve exists to make it. ## Attribution: who carried the loss The same static-pool numbers, cut by dealer: loan count, share of originated balance, cumulative net loss, and EPD rate, ranked by loss. Two mechanical flags do the sorting. A dealer running above 1.5× the *median* dealer's net loss and above 10% absolute is flagged; one below 0.8× the median reads clean. The median, not the mean, is the yardstick, so a single blown-up channel cannot drag the benchmark upward and hide its peers behind it. EPD is the sharper of the dealer numbers. A loan that ages twenty months before defaulting failed slowly, for reasons that can include the borrower's life. A loan that never really performs, first default inside three months, usually failed at the point of sale. **Inference**A dealer channel with an elevated early-payment-default rate is, more often than not, a sourcing and underwriting problem rather than a servicing problem. The loan was weak when it was written. Dealer × vintage is therefore the attribution cut that turns a loss number into a cause. Share of total net dollar loss, synthetic demo book In the illustrative book used throughout this note, three flagged dealer channels wrote about a fifth of the originations and carried nearly half the loss. Synthetic, illustrative only. Flag rule as published above: dealer net loss above 1.5× the median dealer and above 10% absolute. ## Concentration: how lumpy is the book The third cut asks how much of the book depends on any one thing. Across dealer balances, the Herfindahl-Hirschman index, 10,000 × the sum of squared shares, plus the top-three share give the channel answer. Across risk dimensions, a segment screen runs the published bins over credit score, LTV, term, vehicle age, geography, and dealer, and flags a segment only when all four conditions hold: at least 15 loans, loss at 1.35× the book average or worse, at least 7% of total dollar loss, and above 4% absolute. The four conditions together keep the screen honest, a loss multiple alone flags trivia, and a loss share alone flags whatever happens to be big. The read A blended loss number is a claim; a vintage curve at matched age is a test. **The read never comes from the aggregate.** It comes from holding age constant, then asking which vintages, which dealers, and which segments actually carried the dollars, and whether the newest paper is running above the oldest at the same age. ## Limits Honesty about what this method cannot see matters as much as the method. A single snapshot yields one point per vintage, its current age and cumulative loss to date. That sketches a curve *across* vintages, but it is not a true static-pool triangle: without monthly archives you cannot watch one cohort's path over time, only where each cohort stands now. Recoveries lag charge-offs, so young cohorts read worse than they will settle. Small cohorts are noise dressed as signal, which is why the segment screen refuses to flag anything under 15 loans. The deeper limit is that the arithmetic trusts the tape. Status fields are self-reported by the servicer; a deferral or extension resets days past due without curing risk, and nothing in the file distinguishes a cured loan from a re-aged one. And no amount of column-mapping can verify the file itself, that the collateral exists, that the balances are real, or that the same loans are not pledged in two places at once. Arithmetic on a tape tests internal consistency. It cannot test whether the tape is true. **Sources & notes** The benchmark reference points cited in this note, a 9.81% cumulative net loss reference attributed to Fitch and a 6.65% industry 60+ DPD reference, are the figures that were displayed alongside the same metrics on the original dashboard version of this page. All worked examples, including the two-vintage table and the dealer loss-share bar, are **synthetic and illustrative**; they describe the shape of the method, not any real portfolio. All formulas, thresholds, bins, and the column-detection vocabulary are published in full above. LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned. Not investment, legal, or accounting advice. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). LR LendRisk Analytics Independent market research Continue reading [Methods · Note 02 The segments *carrying the loss*](https://lendriskanalytics.com/loss-drivers.html) [Methods · Note 03 Reading the *dealer channel*](https://lendriskanalytics.com/dealers.html) --- title: "Articles" url: https://lendriskanalytics.com/articles.html publisher: LendRisk Analytics kind: Page description: "Plain English analysis on auto credit performance, vintage loss curves, dealer attribution, covenant monitoring, and market stress." html: https://lendriskanalytics.com/articles.html --- # Articles Insights · Editorial # From the data, not the headlines. Plain English analysis on auto credit performance, written for subprime and near-prime auto lenders, BHPH operators, and anyone who wants the numbers to mean more than the headline. [Article of the week What the Tape Said · Issue 16 Tape read Entry, *not exit.* A borrower scored 661 to 780 is prime to every bureau and every rating agency. At Ford that borrower reaches 60 days past due within two years 1.1% of the time. At Santander, 29.3%. Twenty-eight times the rate, same band, same window. Then the convergence: among loans that did reach 60 days down, the share charged off within a year runs 68.1% at Exeter, 68.9% at Santander and 67.8% at World Omni, a prime captive. Entry into distress is where subprime risk lives. Exit is close to a constant. Counted from 907,334 loans in the public tapes. Read the brief → Tape read 27.9x Best to worst lender, same 661-780 band, reaching 60 days down 1.5x The same gap once the borrower is already 60 days down 24.00% Pool loss on Exeter's 2022 trust, every class A note paid in full 1,844 Basis points between reported and extension-adjusted 60+ 14 min read · filings through July 2026](https://lendriskanalytics.com/insights/entry-not-exit.html) [What the Tape Said · Issue 15 Tape read The bottom *moved more.* Used cars are worth a third more than before the pandemic and subprime lenders recover eleven points less on them. The averages hide why. Fifty thousand repossessions from the public loan tapes, split by the year each contract was written: a 2022 Exeter contract came back with 25 cents on the dollar in the past twelve months and a 2024 contract with 47, at the same auctions. The three vintages that make up 98% of what Santander repossessed in 2026 each recovered less than in 2025, and its pooled rate did not move, because newer paper replaced older. Every figure computed from the raw filings. Read the brief → Tape read 50,865 Repossessions with a sale price, two subprime shelves, 2021 to 2026 24.7% Exeter recovery on 2022 contracts, last twelve months of sales 47.3% Same lender, same auctions, 2024 contracts 50.0% Santander pooled, flat on 2025, the paper inside it down 15 min read · filings through July 2026](https://lendriskanalytics.com/insights/set-at-signing.html) [Issue · 14 Research brief Modified, *and not paying.* Every federally insured credit union files one line for loans it has modified for a borrower in trouble, and since 2024 a second line for the modified loans that are already late again. The first has more than doubled since the definition changed. One dollar in four on it is not paying. Every figure rebuilt from the raw NCUA archives. Sep 2026 Read the brief →](https://lendriskanalytics.com/insights/modified-and-not-paying.html) [Issue · 13 Research brief Current, *on tape.* Forty-four registered auto securitizations, fourteen lenders, 1.15 million loans. Every lender grants extensions, prime captives included. On the subprime shelves one dollar in five outstanding was extended in the last six months, and three in ten of those loans are 60+ again within six. Every filing linked. Sep 2026 Read the brief →](https://lendriskanalytics.com/insights/current-on-tape.html) [Issue · 12 Research brief The lot behind *the branch.* One line on the NCUA call report counts cars already repossessed and not yet sold. Seven in ten credit unions leave it blank. Among the ones that fill it in, the size of that number sorts next year's charge-offs, and it has done so in eleven consecutive annual cohorts, including inside the group delinquency calls clean. Aug 2026 Read the brief →](https://lendriskanalytics.com/insights/the-lot-behind-the-branch.html) [Issue · 11 Operator study Past due, *unchanged.* Eleven buy-here-pay-here stores turned in their numbers every month for two and a half years. The share of customers behind did not move: 31.7 percent at the start of 2024, 32.4 by mid-2026. What a dollar of loss cost them in interest moved from $1.14 down to $0.83 and back. The warning light everybody watches sat still through all of it. Aug 2026 Read the study →](https://lendriskanalytics.com/insights/past-due-unchanged.html) [Issue · 10 Research brief The *one-way door.* Credit unions have sold conforming mortgages for decades. Consumer paper was the part that stayed home, until a Tampa credit union sold bonds backed by its own auto loans in November 2019. For four years issuance tracked the liquidity cycle almost exactly. Then the pressure eased and it kept going. Twenty-five deals, roughly $8.4 billion, and every figure traced to source at the end. Aug 2026 Read the brief →](https://lendriskanalytics.com/insights/the-one-way-door.html) [Issue · 09 Research brief The cash *problem.* Seven subprime auto lenders failed or nearly failed between 2023 and 2026, and not one was killed by demand. Every trigger was cash: a pulled bond deal, an over-advance, a maturity wall, a revolver traded away. Credit supplied the shock; funding architecture picked who drowned. Aug 2026 Read the brief →](https://lendriskanalytics.com/insights/the-cash-problem.html) [Issue · 08 Method brief Current, *on paper.* An account gets an extension. The past-due clock resets. The tape shows current, and nothing about the borrower has changed. The Philadelphia Fed now says the headline rate "likely overstates" borrower distress, and no regulator aggregates the fields that would measure it. Aug 2026 Read the brief →](https://lendriskanalytics.com/insights/current-on-paper.html) [Issue · 07 Deep study The future of *subprime.* Six issues of reading the tape after the fact, and here is the forward view. Severity has moved somewhere it can be forecast. The benchmark everyone quotes is dissolving under its own composition. The verification layer nobody built in 1998 gets built by a rating agency, a consortium or a vendor. Four calls, one of them falsifiable. Aug 2026 Read the study →](https://lendriskanalytics.com/insights/the-future-of-subprime.html) [Issue · 06 Deep study Three stress cycles, *one missing layer.* 1997 buried twelve lenders in a boom. 2008 broke the economy and the senior bonds held. 2025 set an all-time delinquency high with unemployment in the low fours. Three macro regimes, the same three causes of death, and a measurement layer the sector was told to build in 1998 and still has not. Aug 2026 Read the study →](https://lendriskanalytics.com/insights/three-cycles-one-missing-layer.html) [Issue · 05 Comparative postmortem Three failures, *one blind spot.* Tricolor, PrimaLend and Car-Mart failed or nearly failed in nine months, funded by JPMorgan, Fifth Third, Barclays, CIBC and Silver Point. The three cases look identical from a distance and share almost nothing up close. What each institution actually missed, and who caught what everything expensive did not. Aug 2026 Read the postmortem →](https://lendriskanalytics.com/insights/three-failures-one-blind-spot.html) [Issue · 04 Stress signal What the tape said: *Credit Acceptance.* The 2022 vintage missed its own forecast by 8.2 points, the worst miss in a decade, and 2023 and 2024 are still seasoning. At the same time, CACC's ABS funding cost fell from 8.6% to 5.1%, and the quarterly bleed just hit a three-year low. Deterioration and stabilization, read from the same tape. Aug 2026 Read the brief →](https://lendriskanalytics.com/insights/cacc-stress-signals.html) [Issue · 03 Stress signal What the tape said: *America's Car-Mart.* The credit book was getting better. The funding architecture is what broke. A $300M distressed-fund term loan, forbearance on five simultaneous covenant defaults, an $18M waiver fee, and 66 days on the clock. In subprime auto, the funding line kills faster than the credit line. Jul 2026 Read the brief →](https://lendriskanalytics.com/insights/carmart-stress-signals.html) [Sector · New Original analysis The credit strength *Washington is trying to outlaw.* The Fed's May note treats a 16.63× repossession rate as a credit strength, lower loss-given-default, and $2B+ of bank commitments rated lower risk. Warren's February probe calls the same act "inexcusable." Both readings can't hold. What re-rates when the regulatory cost of repossession rises. Every figure sourced to the Fed note and the Senate release. Jul 2026 Read the note →](https://lendriskanalytics.com/insights/repossession-liability-turn.html) [Issue · 02 Stress signal What the tape said: *CarMax.* Not a collapse story, which makes it the more useful read. Management called the provisioning peak in June 2025, then added $71.3M in lifetime losses on the same vintages a quarter later, while the nonprime shelf quietly added cushion and lost margin. Every figure sourced to the filings. Jun 2026 Read the brief →](https://lendriskanalytics.com/insights/carmax-stress-signals.html) [Issue · 01 Post-mortem What the tape said: *Tricolor Holdings.* What was publicly visible before the September 2025 collapse, read straight from the filings. Funding dependence, a thin-file pool with 62% no-score borrowers, and diligence blind spots. The signals were in the prospectus supplements and the ABS-15G, not the Bloomberg screens. Sep 2025 Read the post-mortem →](https://lendriskanalytics.com/insights/tricolor-tape.html) [Vol · 06 Case study Run the book: a *tool-by-tool* walkthrough of a $40M line. The same blinded book run through all six tools in sequence, every input, every output, every decision. From "you look compliant" to a covenant breach six months out, three dealers named, one deal declined on the math. May 2026 Read article →](https://lendriskanalytics.com/insights/run-the-book.html) [Vol · 10 Servicing The waiting tax: the most expensive repo is *the one you didn't make.* In deep subprime the loss is in servicing, not the auction. Recovery is value times the odds you actually get the car back, and the second number collapses far faster than the car depreciates. Pairs with the recovery-decay method note. Jun 2026 Read article →](https://lendriskanalytics.com/insights/the-waiting-tax.html) [Vol · 09 Sector note Buy-here-pay-here *grew up.* The Fed's 2026 data shows BHPH is now bank-financed and guarantor-backed, balances up 214% since 2018. Institutionalization imports finance-company fragility into a segment that used to absorb its own losses. Jun 2026 Read article →](https://lendriskanalytics.com/insights/bhph-institutionalized.html) [Vol · 08 Underwriting Early payment default isn't a credit event. *It's a fraud signal.* Up to 70% of early payment defaults trace to application fraud, and nearly all of it comes from ~10% of dealers. The fix isn't a higher score floor, it's a watched door. Jun 2026 Read article →](https://lendriskanalytics.com/insights/early-payment-default.html) [Vol · 05 Case study Anatomy of a *toxic book*: the aggregate said fine. Tricolor read compliant the quarter it pledged $2.2B against $1.4B of real collateral. A blinded book anchored to the public record, Fitch, the NY Fed, the DOJ filing, showing how three dealers and one vintage steer a $40M line toward a sweep, and how the slope was visible the whole time. May 2026 Read article →](https://lendriskanalytics.com/insights/toxic-book.html) [Vol · 04 Teardown What the *Tricolor* collapse actually says about deep subprime auto. Roll rates were inside covenant the quarter it failed. The warehouse banks did not miss a number, they missed a slope. Here is what the bank was tracking that the lender was not. Mar 2026 Read article →](https://lendriskanalytics.com/insights/tricolor.html) [Vol · 03 Methodology Roll rate analysis: how delinquency migration *actually* works. A plain English walk through transition matrices, why warehouse banks know your 90-day DPD before you do, and how to build the same view yourself with a single spreadsheet. Feb 2026 Read article →](https://lendriskanalytics.com/insights/roll-rates.html) [Vol · 02 Sector note BHPH charge-offs in 2026: what *normal* actually looks like. A 20% annual charge-off rate is normal for a properly priced BHPH book. The operators going under are not the ones with high losses. They are the ones whose recovery model assumed 2021 vehicle prices. Jan 2026 Read article →](https://lendriskanalytics.com/insights/bhph-normal.html) [Vol · 01 Market analysis Subprime 60+ DPD hits *6.90%* in Q1 2026. Eighteen consecutive quarters of deterioration. The aggregate number is not what should worry you. The shape of the vintage curves underneath it is the actual signal. Jan 2026 Read article →](https://lendriskanalytics.com/insights/q1-2026-dpd.html) --- title: "The Borrowing-Base Certificate \u00b7 Lender-Side Market Brief" url: https://lendriskanalytics.com/borrowing-base-lender.html publisher: LendRisk Analytics kind: Page description: "Tricolor's warehouse lenders had every standard structural protection and still lost roughly $370 million between two banks, because every protection sits downstream of a self-reported borrowing-base certificate. A market brief on the monitoring gap between field exams and the emerging practice of independent monthly recomputation of certificates from loan tapes. Independent research from the public record, not an audit, not advice." html: https://lendriskanalytics.com/borrowing-base-lender.html --- # The Borrowing-Base Certificate · Lender-Side Market Brief Market brief · Independent research Analysis from the public record. Not an audit and not advice. Where you sit: [**Operator** You file the certificate](https://lendriskanalytics.com/borrowing-base.html) [**Warehouse lender** You rely on it](https://lendriskanalytics.com/borrowing-base-lender.html) LendRisk Analytics · The Borrowing-Base Certificate · Lender Side # A clean certificate and a fabricated one look identical on paper. Why the sector-wide re-rate is a measurement problem, and what independent verification of the borrowing-base certificate would have to look like **Series**  Market brief **Data**  Fed FEDS Note · May 2026 **Record**  Tricolor Ch.7 · Sept 2025 **Status**  Public record Research · Not advice $2.2B Collateral Tricolor pledged $1.4B Collateral that existed ~$370M Losses · Fifth Third + JPMorgan 12 : 1 Certificates filed per field exam +150% Sector PD re-rate · one quarter What's going on Tricolor's warehouse lenders had every standard structural protection: advance rates of 60-80%, special-purpose entities, guarantees. Fifth Third took a roughly **$200 million impairment** and JPMorgan a **$170 million charge-off** anyway, because the protections were all computed against a borrowing-base certificate the borrower invented, loans pledged to multiple banks at once, loans already sold into securitizations, loans that never existed, delinquent paper dressed up as eligible. The market's answer so far has been to reprice everything: reported probability-of-default on BHPH facilities rose **nearly 150% in one quarter** after the collapse. That protects no one and costs margin on every clean credit in the sector. This brief examines the narrower answer the public record points toward, independent monthly recomputation of each borrower's certificate from its loan tape, and asks what that verification would actually have to look like for the certificate to stop being taken on trust. 01 · What Tricolor proved The protections were set right. The number under them was not. Record · bankruptcy filings, bank disclosures, federal charges The uncomfortable part Nothing in the Tricolor structure was lazy. The advance rates were conservative. The SPEs were in place. The guarantees were signed. Post-mortems have not found a covenant that should have been tighter, what they found is that **every protection in an ABL facility is arithmetic performed on a self-reported number**, and the number was fiction. Roughly $2.2 billion pledged against $1.4 billion that existed. A 70% advance rate against a fabricated collateral figure is still fabricated. The scheme was also not exotic. Double-pledging across warehouse lines, continuing to pledge loans after selling them into securitizations, and re-aging delinquent paper into eligibility are all things that **a recomputation from the loan tape is positioned to catch**, not because any single tape proves the loans exist, but because fabrication leaves arithmetic seams: balances that don't roll, aging that doesn't match payment history, populations that shift in ways originations can't explain. *Structural protection is downstream of measurement.* If the measured number is invented, the advance rate, the SPE, and the guarantee are all invented with it. 02 · The monitoring gap Twelve certificates a year. One field exam. Practice · standard ABL surveillance cadence The cadence mismatch Standard practice on an ABL facility is a field exam roughly annually, more often for new or troubled credits, with perhaps a quarterly desk review between. The exam samples a pool, tests it, and moves on. Meanwhile the borrower files a certificate **every month**, and every month's advance is computed from it. Between exams, the certificate is taken on trust; the exam itself sees one month out of twelve, months after the fact. The post-Tricolor response, re-rating the whole sector, is what a portfolio does when it cannot measure borrower by borrower. It is expensive in both directions: **spread is given back on clean credits** that deserve better terms, and **the next fabricated tape is not caught**, because sector-level repricing does not read anyone's collateral. The gap is not a pricing problem. It is a measurement problem wearing a pricing costume. | Layer | Cadence | What it actually verifies | |---|---|---| | Borrowing-base certificate | Monthly | Nothing, it is the borrower's own statement | | Desk review | Quarterly, if that | Internal consistency of the documents as filed | | Field exam | ~Annual | A sampled pool, one point in time, in arrears | | Independent recomputation · emerging | Monthly | Certificate vs. loan tape, every filing, variance traced to cause | Pricing the sector because you cannot measure the borrower *gives back margin on the clean books and still misses the dirty one.* Illustrative arithmetic, not an observed quote: if borrower-level measurement supported pricing a clean credit 50-100bps inside a sector-repriced facility, that would be $500K,$1.0M a year of spread on a $100M line. Synthetic figures, shown only to size what measurement is worth to both sides of a facility. 03 · What verification would have to look like Every certificate, recomputed from the tape it claims to summarize. Inference · where post-Tricolor surveillance points The structural answer, described The verification the market is converging toward after Tricolor is not a new covenant. It is independent recomputation of the certificate itself. Two inputs per borrower per month: the certificate as filed, and the loan-level tape underneath it, a standard CSV export that every mainstream BHPH DMS (Verifacto, DealerCenter, Frazer, Wayne Reaves, DealerClick) already produces. Eligible collateral is recomputed under **the facility's own eligibility rules**, delinquency thresholds, charge-off exclusions, ineligible collateral types, concentration limits, and reconciled against the reported number, with every variance quantified and traced to a cause. Most months, on most borrowers, a recomputation confirms the certificate, **which is itself the information**: it is what lets a clean credit be measured as a clean credit. The month it stops confirming, the facility has a named cause and a current tape, not a year-old sample. That is the entire difference between a variance and a headline. Where certificates and tapes typically diverge, six recurring cause classes, from the mechanical to the substantive: | Cause | What it looks like on the tape | What it usually means | |---|---|---| | Reconciliation imbalance Mechanical | Beginning balance plus originations, minus collections and charge-offs, does not roll to the ending balance. | Process error at best; at worst, balances that exist only on the certificate. | | Dealer attribution Structural | Loans attributed to the wrong lot or related entity, moving collateral between books. | Often benign in a multi-entity group, and the exact seam double-pledging hides in. | | Layout mismatch Mechanical | The DMS export's rows or columns shift between cycles, silently changing what gets counted. | Pure data hygiene, but it moves reported eligibility without anyone deciding anything. | | Undocumented metric Data | A figure on the certificate with no derivable source anywhere in the tape. | The strongest flag in the Tricolor record: reported numbers no data supported. | | Aging misclassification Substantive | Delinquency buckets inconsistent with the payment history underneath them. | Re-aging, the standard route by which ineligible paper stays eligible. | | Eligibility disagreement Substantive | Certificate and recomputation apply the facility's rules differently to the same loan. | A genuine interpretive difference, the one class that is an argument, not an error. | What recomputation catches · stated precisely A tape reconciliation cannot prove a loan exists in the world, that is what a field exam's verification procedures are for. What it catches is the arithmetic residue of both error and invention: balances that do not roll month to month, aging inconsistent with payment history, eligible populations that move in ways originations cannot explain, and reported figures with no derivable source in the data. **It closes the gap between the spot-checks. It does not replace them.** 12 / yr Filings a monthly cadence reads 6 Recurring variance cause classes 100% Of the tape, not a sample **Sources & framing.** Sector figures, 65% asset-based share of BHPH bank facilities, 81% guarantor coverage, 60-80% advance rates, and the ~150% quarter-over-quarter rise in reported probability of default from 2025:Q2 to 2025:Q3, are from the Federal Reserve FEDS Note, *"Subprime Auto Lending: Trends in Buy Here Pay Here Auto Lending"* (May 8, 2026). Tricolor figures, the September 2025 Chapter 7 filing, approximately $2.2B pledged against roughly $1.4B of actual collateral, Fifth Third's ~$200M impairment and JPMorgan's ~$170M charge-off, are from the bankruptcy record, bank disclosures, and federal fraud charges as publicly reported; allegations are allegations until adjudicated. Field-exam cadence reflects standard ABL practice as described in examiner handbooks and industry guidance; individual facilities vary. Basis-point and dollar spread figures are illustrative, not observed quotes. Interpretive and forward-looking statements are labeled as inference. This is independent market research from the public record. It is not an audit, attestation, investment, legal, or accounting advice. LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned. The point Underwrite the borrower, not the headline. The sector re-rate is what pricing looks like when measurement fails. Borrower-level measurement, every certificate, every month, recomputed from the tape, is what lets a clean credit be read as a clean credit, and what surfaces the other kind while it is still a variance and not a headline. However the market gets there, facility terms, lender requirements, or borrowers volunteering the proof, the direction after Tricolor points one way: the certificate stops being taken on trust. Independent market research · Not an audit, attestation, or advice Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). --- title: "The Borrowing-Base Certificate \u00b7 Market Brief" url: https://lendriskanalytics.com/borrowing-base.html publisher: LendRisk Analytics kind: Page description: "The monthly borrowing-base certificate is self-reported, and after Tricolor the banking system re-rated the entire BHPH sector because it cannot tell a clean certificate from a fabricated one. This market brief explains what changed, why it runs through one monthly document, and what independent verification of that document would have to look like. Independent market research from the public record, not an audit, not advice." html: https://lendriskanalytics.com/borrowing-base.html --- # The Borrowing-Base Certificate · Market Brief Market brief · Independent research Analysis from the public record. Not an audit and not advice. Where you sit: [**Operator** You file the certificate](https://lendriskanalytics.com/borrowing-base.html) [**Warehouse lender** You rely on it](https://lendriskanalytics.com/borrowing-base-lender.html) LendRisk Analytics · Market Brief # The certificate is now the whole conversation. Why the monthly borrowing-base certificate became the sector's pressure point, and what independent verification of it would have to look like **Series**  Market brief **Data**  Fed FEDS Note · May 2026 **Record**  Tricolor Ch.7 · Sept 2025 **Status**  Public record Research · Not advice 65% BHPH bank facilities that are ABL 81% Guarantor-backed 60-80% Typical advance rate +150% Sector PD re-rate · one quarter ~$370M Tricolor losses · two banks What's going on In September 2025, Tricolor, one of the largest BHPH operators in the country, filed Chapter 7. Its executives are federally charged with pledging roughly **$2.2 billion of collateral against $1.4 billion that actually existed**: loans pledged to multiple banks at once, loans that had already been sold into securitizations, and loans that were never real. In May 2026 the Federal Reserve published the plumbing of the whole sector in a FEDS Note, and the numbers above are from it. The banking system's response was not surgical. Banks' reported probability-of-default on BHPH borrowers rose **nearly 150% in a single quarter**, the whole sector, not the borrowers who did something wrong. If you run a clean book, you are now paying for Tricolor. This brief explains why that happened, why it runs through one monthly document, and what independent verification of that document would have to look like. It describes the market's problem, not anyone's product. 01 · What just happened The banks re-rated the sector without reading its tapes. Source · Federal Reserve FEDS Note · May 2026 The re-rate Before Tricolor, banks actually rated BHPH facilities as slightly **lower** risk than loans to traditional dealers, the structures looked strong: 65% asset-based, 81% guarantor-backed, advances at 60 to 80 cents on the dollar of receivable value. From the second to the third quarter of 2025, the reported probability of default on the sector's bank facilities rose nearly 150%. Read that carefully. The banks did not discover that every operator's book got worse in ninety days. They discovered that **they could not tell which books were real**, so they repriced all of them. That is what pricing the sector instead of the borrower looks like, and it means the spread a clean book pays now carries someone else's fraud premium. *Banks are pricing the sector, not the borrower.* A clean book pays the same fraud premium as a dirty one, because from the bank's chair they look identical. 02 · The document at the center One self-reported PDF governs the whole facility. Structure · how an ABL warehouse actually works The mechanics An asset-based facility works like this: every month the operator files a **borrowing-base certificate**, its own statement of how much eligible collateral sits in the book, and the bank advances against it. The bank's verification is a field exam, typically annual, sometimes with a quarterly desk review between. That leaves roughly **twelve self-reported certificates for every one independent look**. Every filing in between is taken on trust. Tricolor is what it looks like when that trust is abused with intent. Its lenders had every standard protection, 60-80% advance rates, special-purpose entities, guarantees, and Fifth Third and JPMorgan still lost roughly $200 million and $170 million respectively. The protections did not fail because they were set wrong. They failed because they all sit downstream of the reported number, and the reported number was invented. **A 70% advance rate against a fabricated collateral figure is still fabricated.** | Structural protection | Sector prevalence | What it silently assumes | |---|---|---| | Advance rate of 60-80% | Standard | The reported collateral exists and is eligible | | Personal / corporate guarantee | 81% of facilities | There is an estate worth chasing after the loss | | Special-purpose entity | ~13% of dealers | The assets inside the SPE are real and unencumbered | | Field exam | Typically annual | Nothing drifts, or breaks, in the eleven months between | Every structural protection in the stack sits downstream *of a self-reported number.* The certificate is the single point of failure, and the banking system just learned it, publicly. 03 · What verification would have to look like The reported number, traced back to the tape. Method · monthly recomputation from the DMS tape Inference · where the market is converging The structural gap is the twelve-to-one ratio: twelve self-reported certificates for every field exam. The answer the market has been converging on since Tricolor is not more guarantees or lower advance rates, those all sit downstream of the reported number. It is **independent monthly recomputation of the certificate itself**, at the same cadence it is filed. Mechanically, that means two inputs each month: the certificate filed with the lender, and the loan tape underneath it, a CSV export every mainstream DMS produces (Verifacto, DealerCenter, Frazer, Wayne Reaves, DealerClick). From the tape, eligible collateral is recomputed under **the facility's own eligibility rules**, delinquency thresholds, charge-off exclusions, ineligible collateral types, concentration limits, and the recomputed number is reconciled against the number on the certificate. The essential property is independence of computation: someone other than whoever filed the certificate rebuilds it from the raw data, every month it is filed, whether that is a lender's own surveillance desk, a third party, or a walled-off function inside the operator. When the two numbers diverge, the divergence has a cause. Across the sector's reporting stack, the causes cluster into a handful of repeat offenders, this is where certificates and tapes typically diverge: | Variance cause | What it looks like in the data | |---|---| | Reconciliation imbalance Mechanical | Principal activity that doesn't balance across the month, beginning balance plus originations minus payments and charge-offs doesn't equal ending balance for a dealer or a pool. | | Dealer-attribution error Mechanical | Receivables booked under the wrong dealer or entity, shifting concentration math and sometimes eligibility. | | Row-layout mismatch Reporting | Two parts of the operation reporting the same metric on different layouts, so the roll-up silently double-counts or drops rows. | | Undocumented metric Reporting | A number on the certificate that nobody can define from the tape, no formula, no source column, no owner. | | Aging misclassification Eligibility | Accounts sitting in a delinquency bucket that doesn't match their payment history, moving them across the eligible / ineligible line. | | Eligibility disagreement Judgment | A genuine difference in how a facility rule reads against an account, flagged, argued both ways, and labeled as interpretation, not error. | Why the taxonomy matters Notice what dominates that table: not fraud. Reconciliation drift, attribution errors, layout mismatches, undefined metrics, ordinary operational entropy. These are exactly the variances an annual field exam eventually finds, eleven months late and framed less charitably. Monthly recomputation finds them in the month they occur, assigns each one a cause rather than just a flag, and separates confirmed findings from inference, because a reconciliation that overstates its own certainty is worth less than no reconciliation at all. That is also why the practice matters commercially and not just operationally. A lender pricing the sector has no instrument that distinguishes a clean book from the sector; twelve months of independent certificate-to-tape reconciliation is that instrument. Whether any given lender re-prices on it is that lender's call. But the arithmetic of what borrower-level pricing is worth, labeled as arithmetic, not a promise, is below, and it is not subtle. The arithmetic · illustrative, not a forecast If independent monthly verification moved a lender from sector pricing to borrower pricing by even **50 basis points** on a **$100M facility**, that is **$500K a year**. At 100 bps it is $1M. That is the size of the wedge the sector's fraud premium has driven between clean books and their own cost of funds. Whether any lender moves at all depends on the book and the lender, this is arithmetic, labeled as arithmetic. 50-100 bps The spread question on the table $500K,$1M Per year, per $100M facility · illustrative 12 : 1 Certificates filed per field exam **Sources & framing.** Sector figures, 65% asset-based share, 81% guarantor coverage, 60-80% advance rates, ~13% SPE usage, and the ~150% quarter-over-quarter rise in reported probability of default, are from the Federal Reserve FEDS Note, *"Subprime Auto Lending: Trends in Buy Here Pay Here Auto Lending"* (May 8, 2026). Tricolor figures, the September 2025 Chapter 7 filing, approximately $2.2B pledged against $1.4B of actual collateral, Fifth Third's ~$200M impairment and JPMorgan's ~$170M charge-off, are from the bankruptcy record, bank disclosures, and the federal fraud charges as publicly reported; allegations are allegations until adjudicated. The basis-point arithmetic is illustrative, not a forecast. This is independent market research from the public record. It is not an audit, attestation, investment, legal, or accounting advice. LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned. The point The certificate stopped being paperwork. The sector got re-rated because one operator's tape was fiction and nobody could tell. The books that get priced as borrowers rather than as a sector will be the ones where the reported number and the recomputed number can be shown, monthly, independently, to be the same number. The market has stopped extending that trust for free; what the tape shows is what will replace it. Market brief · Public record · Research, not advice Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). --- title: "Covenant runway: the month the trigger breaks" url: https://lendriskanalytics.com/covenant.html publisher: LendRisk Analytics kind: Page description: "The full arithmetic of covenant runway: project each metric linearly from its current level and monthly slope, and the first threshold crossed sets the headline. A method note on why the slope, not the snapshot, is the signal." html: https://lendriskanalytics.com/covenant.html --- # Covenant runway: the month the trigger breaks [← All insights](https://lendriskanalytics.com/articles.html) Methods · Note 04 Method note · 6 min read Method note · Covenant runway # Covenant runway: the month the trigger breaks. LendRisk Analytics · Method note · 2026 Every warehouse facility carries triggers: a net charge-off cap, a delinquency ceiling, a weighted-FICO floor. The question that matters in practice is not whether a portfolio sits inside them today. It is which trigger breaks first, and in what month. This note publishes the full arithmetic of that calculation, the standard bands the market sets those triggers within, and why the monthly slope carries more signal than any snapshot. ## The runway arithmetic The method needs three numbers per covenant metric: where the metric stands today, how far it moves each month, and where the trigger sits. Project each metric forward on a straight line, **value at month n = current + n × monthly Δ**, and read off the first month at or past the line. For a cap-type covenant (charge-offs, delinquency), the breach month is **ceil((cap − current) ÷ slope)**. For a floor-type covenant (weighted FICO), it is **ceil((current − floor) ÷ |slope|)**. The ceiling function is not decoration: covenants test at reporting dates, so the first monthly test at or beyond the threshold is the breach month, even if the line is crossed mid-month. Run that for every trigger in the facility, and the **earliest breach month is the runway headline**. One number: how many monthly reporting cycles remain before the first covenant conversation stops being hypothetical. | Quantity | Rule | Convention | |---|---|---| | Projected value | value(n) = current + n × Δ/mo | Linear, 18-month horizon | | Breach month, cap-type | ceil((cap − current) ÷ slope) | Slope ≤ 0 → no breach projected | | Breach month, floor-type | ceil((current − floor) ÷ /slope/) | Slope ≥ 0 → no breach projected | | Already at threshold | breach month = 0 | At or past the trigger today | | Runway | min of all breach months | Earliest breach sets the headline | The complete logic. Breaches projected beyond month 18 are reported as none projected, a convention about forecast humility, not a claim of safety. ## The standard bands Warehouse facilities in subprime auto tend to set triggers inside recognizable ranges: **net charge-off caps at 8 to 10 percent, 90-plus-day delinquency caps at 4 to 6 percent, and weighted-FICO floors at 560 to 590**. Below the covenant line, the working bands most monitoring frameworks use are tighter still, charge-offs read as comfortable under 4 percent and as a watch item from 4 to 8; delinquency reads as comfortable under 3 and as a watch item from 3 to 5; a weighted FICO under 560 reads as elevated risk, with 560 to 590 the watch zone. The same runway logic applies to any threshold, not just facility covenants: a CECL or allowance-coverage trigger, a concentration limit, or an internal board risk tolerance. Anything with a current level, a monthly drift, and a line it must not cross has a breach month. **Inference**A facility whose caps sit well outside the typical bands, a 13.5 percent charge-off cap against a market norm of 8 to 10, is telling you something about how it was negotiated. Wide caps rarely come free; they tend to arrive paired with lower advance rates, faster sweep triggers, or tighter eligibility carve-outs elsewhere in the document. The headline cap is one term in a system of terms. ## Why the slope, not the snapshot Two portfolios each report 9.6 percent charge-offs against a 13.5 percent cap. Identical snapshots, identical headroom: 3.9 points. One is drifting up at a tenth of a point per month, 39 months to the line, beyond any forecast worth taking literally. The other is deteriorating at 0.65 points per month, six months to the line. The snapshot cannot tell these two books apart. **The slope is the entire signal.** That is the same lesson the public record taught in the [Tricolor collapse](https://lendriskanalytics.com/insights/tricolor.html): the level looked survivable right up until the trajectory said otherwise. The slope framing also reorders which trigger matters. Runway is a race between ratios, headroom divided by slope, not a ranking of headroom. Consider a synthetic baseline book, illustrative only: | Metric | Today | Slope / mo | Trigger | Breach month | |---|---|---|---|---| | NCO rate | 9.30% | +0.18pp | cap 13.50% | M24 · beyond window | | 90+ DPD | 6.90% | +0.09pp | cap 10.00% | M35 · beyond window | | Weighted FICO | 578 | −1 pt | floor 570 | M8 · first to break | Synthetic baseline, illustrative only. Runway headline: 8 months, set by the least dramatic-looking gap on the page. **Inference**In this baseline the charge-off line, the number everyone watches, has more than four points of headroom and two years of runway. The FICO floor, eight points away and drifting one point a month, breaks first. A monitoring routine that ranks covenants by headroom in native units will consistently watch the wrong trigger; ranking by headroom-over-slope is the whole discipline. ## A worked example: month six Take the fast-deteriorating book from above, synthetic and illustrative throughout. Charge-offs stand at **9.6 percent**, worsening at **0.65 points per month**, against a **13.5 percent cap**. Headroom is 3.9 points; 3.9 ÷ 0.65 = 6.0, and the ceiling of 6.0 is 6. The sixth monthly test lands exactly on the cap. | Month | Projected NCO | Headroom to 13.50% cap | |---|---|---| | M0 · today | 9.60% | 3.90pp | | M1 | 10.25% | 3.25pp | | M2 | 10.90% | 2.60pp | | M3 | 11.55% | 1.95pp | | M4 | 12.20% | 1.30pp | | M5 | 12.85% | 0.65pp | | M6 · breach | 13.50% | 0.00pp | Synthetic worked example. Each row is one reporting cycle; the trigger breaks at the sixth test. Headroom at today's reading Of the 13.5% cap, 9.6 points are already consumed. At +0.65pp per month, the green segment lasts six reporting cycles. Synthetic example. Bar is proportional: 9.6 of 13.5 points consumed (71%), 3.9 points (29%) remaining. The read A covenant does not break when the portfolio gets bad. It breaks when **headroom divided by slope runs out.** Both numbers sit on the monthly servicing tape, and the warehouse bank computes them every cycle whether or not the operator does. Runway is not a private calculation. It is a shared clock, and the only question is who reads it first. ## What a tripped trigger actually does The mechanics matter because they start before the breach month. Most facilities carry intermediate trigger levels short of an event of default, and the first consequence is usually the **cash sweep**: collections that would ordinarily flow back to the operator as excess spread are instead trapped in the structure and swept to pay down the facility. The operator's own cash flow becomes the cure mechanism. Alongside the sweep come advance-rate step-downs, eligibility exclusions that shrink the borrowing base, and, at the covenant line itself, cure periods measured in days, not quarters. Sequencing differs by trigger. Delinquency breaches tend to draw the fastest response, because 90-plus-day delinquency is a leading indicator of the charge-offs to come; a charge-off breach confirms what the delinquency tape already said months earlier. And in every case the formal review starts well before month zero of the projection, the lender sees the same slope the operator does, and initiates the conversation on its own schedule. ## Limits This method is a straight line, and portfolios do not move in straight lines. Roll rates compound, so deterioration tends to accelerate late, a linear breach month is usually the optimistic case for a worsening book. Denominators move: rapid origination growth mechanically flattens a charge-off ratio while a shrinking book steepens it, with no change in underlying credit. Seasonality, tax-refund cycles in particular, puts real curvature in monthly delinquency. Covenant definitions vary by document, trailing three-month annualized versus cumulative static-pool measures can put the same portfolio months apart on the same chart. And the arithmetic cannot see the relationship: waivers, amendments, and lender discretion decide what a breach actually costs, and none of that is in the tape. Treat the breach month as a planning number, not a prediction. **Sources & notes** The covenant bands cited here, net charge-off caps of 8 to 10 percent, 90-plus-day delinquency caps of 4 to 6 percent, weighted-FICO floors of 560 to 590, and the tighter monitoring bands beneath them, describe typical warehouse facility structures in subprime auto; individual facility documents vary and always control. All portfolio figures in this note, including the baseline scenario and the month-six worked example, are **synthetic and illustrative**; they describe no actual company. This note publishes, in full, the logic previously implemented as an interactive projector at this address. LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned. Not investment, legal, or accounting advice. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). LR LendRisk Analytics Independent market research Continue reading [Methods · Note 05 The composite read: *three numbers* against warehouse bands](https://lendriskanalytics.com/tool.html) [Methods · Note 06 Recovery decay: *what a day of delay* costs](https://lendriskanalytics.com/repo-timing.html) --- title: "Data & sources" url: https://lendriskanalytics.com/data.html publisher: LendRisk Analytics published: 2026-05-08 kind: Page description: "Every dataset, filing, and publication LendRisk Analytics draws on, government statistics, Federal Reserve research, ratings-agency data, industry reports, and public company disclosures. All public, none proprietary or loan-level." html: https://lendriskanalytics.com/data.html --- # Data & sources [← Methods](https://lendriskanalytics.com/tools.html) Data & sources # Where the numbers come from. Everything on this site is built on **public data**: government statistics, central-bank research, ratings-agency series, industry reports, and the disclosures companies file themselves. This page lists all of it, what each source covers, and where it shows up in the work, so any figure can be traced back to a named, checkable origin. **What this publication does not use.** No loan-level or borrower-level tapes from any lender. No proprietary or subscription datasets presented as our own. No vehicle-history or title-history data. Worked examples in the method notes are **synthetic**, generated from the stated formulas to illustrate mechanics, and are calibrated to, never copied from, the public sources below. 01 · Government & regulators ## Official statistics and filings. Primary public record, the backbone of the sector work. | Source | What it covers | Used in | |---|---|---| | Federal Reserve | FEDS Note: "Subprime Auto Lending: Trends in Buy Here Pay Here Auto Lending" (May 8, 2026), bank-facility structure, guarantor coverage, advance rates, and the post-Tricolor probability-of-default re-rate. | BHPH & certificate briefs | | Federal Reserve Bank of New York | Household Debt and Credit Report (Consumer Credit Panel / Equifax), the 620 subprime line, auto-loan delinquency by tier, origination shares by credit score. | Loss attribution, DPD notes | | Consumer Financial Protection Bureau | Consumer Credit Trends and auto-finance Data Point reports, market monitoring, enforcement actions, negative-equity and repossession research. | Underwriting & recovery notes | | Federal Highway Administration | Highway Statistics Series, Table VM-1 , annual vehicle-miles traveled and the ~13,500-mile average annual mileage used in term-to-life matching. | Term / useful-life note | | Bureau of Transportation Statistics | National Transportation Statistics, Average Age of Automobiles and Trucks in Operation , fleet-age and vehicle-miles series. | Term / useful-life note | | U.S. Senate Banking Committee | February 2026 auto-repossession inquiry (minority release), the letter, its twelve recipients, and the error-rate framing. | Repossession-liability note | | SEC EDGAR | 10-K / 10-Q / 8-K / ABS-15G filings , issuer disclosures, financing terms, and securitization due-diligence scope. | Stress-signal briefs | | NHTSA | vPIC & recall databases , vehicle identification, safety, and recall data. | Collateral context | 02 · Ratings & securitization ## The ABS record. Cumulative-loss assumptions, credit enhancement, and deal-level disclosure. | Source | What it covers | Used in | |---|---|---| | Fitch Ratings | U.S. Auto Loan ABS Index (subprime net-loss and 60+ day delinquency series), the sector loss benchmark. | Benchmarks, loss attribution | | S&P Global Ratings | Auto ABS presale reports , expected-cumulative-net-loss (ECNL) assumptions and structure detail, deal by deal. | Stress-signal briefs | | KBRA | Auto ABS rating & surveillance reports , ratings, credit-enhancement levels, and collateral pool summaries. | Tricolor & CACC briefs | | Moody's Ratings | Auto ABS indices & presales , cumulative-net-loss expectations on a comparable (un-haircut) basis. | Stress-signal briefs | 03 · Industry & market data ## Published market series. Financing trends, valuations, and vehicle longevity, all public reports. | Source | What it covers | Used in | |---|---|---| | Experian | State of the Automotive Finance Market (quarterly report), loan-term distribution (the 73-84-month share, now ≈⅓ of loans) and subprime origination and delinquency shares. | Loss attribution, term note | | TransUnion | Quarterly Credit Industry Insights Report (CIIR) , auto-loan 60+ day delinquency by risk tier (1.45% in Q3 2025) and origination trends. | Benchmarks | | Cox Automotive | Manheim Used Vehicle Value Index (MUVVI) , monthly wholesale used-vehicle valuation and depreciation trend. | Recovery & term notes | | Edmunds | Quarterly Insights Report, negative-equity data , share of trade-ins underwater (29.3% in Q4 2025) and average amount owed ($7,214), behind the LTV axis. | Loss attribution, negative equity | | iSeeCars | Longest-Lasting Cars Study (2025) , odds of a model reaching 250,000 miles, from ~400 million vehicles analysed; the basis for the reliability-tier ordering. | Term / useful-life note | | RepairPal | RepairPal Reliability Ratings , average annual repair cost by brand (e.g. Toyota ≈$634 vs Porsche ≈$2,345); the ordering behind the tier repair-cost column. | Term / useful-life note | | Consumer Reports | Annual Auto Reliability Survey and Cost of Car Ownership study, reliability and ten-year ownership cost by brand. | Term / useful-life note | | NIADA / NABD | NIADA Used Car Industry Report and National Alliance of Buy Here Pay Here Dealers operating benchmarks, BHPH and independent-dealer metrics. | BHPH sector notes | 04 · Company disclosures ## What issuers report themselves. The stress-signal briefs read only what a company has publicly filed or announced. | Source | What it covers | Used in | |---|---|---| | Credit Acceptance (CACC) | FY2025 Form 10-K, quarterly earnings releases, ABS financing 8-Ks , and the forecasted-collection-rate-by-vintage table. | CACC brief | | America's Car-Mart | Form 10-K/10-Q, credit-facility and covenant-waiver 8-Ks , and term-loan disclosures. | Car-Mart brief | | CarMax | 10-K and CarMax Auto Finance (CAF) segment disclosures , provisioning and loss-reserve detail. | CarMax brief | | Tricolor Holdings | Rated/unrated ABS reports, ABS-15G due-diligence filings, and the bankruptcy docket , plus the public fraud allegations. | Tricolor post-mortem | **On how these are used** Market figures are cited to a named source at the point they appear. Where a note reasons beyond what a document literally states, that step is labelled as inference. Worked examples are synthetic and generated from published formulas; they describe no actual lender, book, borrower, or vehicle. Company names appear only as subjects of analysis drawn from their own public disclosures. LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned. Nothing here is investment, legal, or accounting advice. Think a source is missing, or read one differently? [Send it through →](https://lendriskanalytics.com/contact.html). --- title: "Reading the dealer channel" url: https://lendriskanalytics.com/dealers.html publisher: LendRisk Analytics kind: Page description: "A method note on scoring dealer channels in an indirect auto book: loss rate, severe lates, early payment default, and a volume-weighted composite health score, with the full weights, stress caps, and banding published and a worked synthetic example." html: https://lendriskanalytics.com/dealers.html --- # Reading the dealer channel [← All insights](https://lendriskanalytics.com/articles.html) Methods · Note 03 Method note · 6 min read Method note · Dealer channel # Reading the dealer channel. LendRisk Analytics · Method note · 2026 Every indirect auto book eventually asks the same question: which dealers are sending the bad paper? The portfolio-level loss number cannot answer it, because an aggregate is an average, and averages launder concentration. This note publishes the full method behind a dealer scorecard, three signals, exact weights and stress caps, a volume adjustment, and a three-band reading, then walks it through a synthetic twelve-channel book to show why a handful of channels usually carry most of the deterioration. 40/30/30 Signal weights: loss rate / severe lates / early payment default ≥ 70 High-risk line on the 0-100 composite score 48% Of losses from the 3 worst channels in the worked book (synthetic) 1.8x Worst channel's loss share vs. its volume share (synthetic) ## Three signals, read together A dealer channel is scored on three numbers, each doing a different job. **Net charge-off rate (NCO)** is the realised loss number, settled fact, entirely backward-looking. **90+ days past due** is the severe-lates share, the leading indicator: most accounts that reach ninety days roll forward into charge-off, so this is next quarter's loss number arriving early. **Early payment default (EPD)** is the share of loans that went 60+ days delinquent within their first six months on book. It is the cleanest dealer-level underwriting signal in the set, because a loan that fails almost immediately was usually never good to begin with, and origination is where the dealer sits. The same arithmetic runs on any grouping key. An indirect lender cuts the book by dealer channel. A buy-here-pay-here operator can run the identical score by sales rep or by vehicle class. A multi-lot group can run it by location. The method does not change; only the grouping does. ## The score: weights, caps, and the volume adjustment The composite health score runs 0 to 100, higher meaning worse. Each signal is first normalised against a stress cap, divide the channel's rate by the cap, truncate at 1.0, so no single blown-out metric can dominate beyond its weight. The caps are 15% for NCO, 12% for 90+ DPD, and 10% for EPD, levels set against published subprime auto benchmark families (Fitch, S&P Global, TransUnion). The three normalised signals are then weighted 40 / 30 / 30 and multiplied by a volume weight. The volume weight uses a log scale on the channel's trailing-90-day loan count: **min(1.2, 0.7 + 0.3 · log₁₀(loans ÷ 60))**. A channel with 60 loans or fewer is damped to 0.70; roughly 190 loans earns about 0.85; 600 loans reaches 1.00; the weight caps at 1.20 near 2,800 loans. The full specification: | Component | Definition | Rule | |---|---|---| | NCO rate | Realised net charge-offs, trailing | 40% weight · stress cap 15% | | 90+ DPD | Severe lates; leading loss indicator | 30% weight · stress cap 12% | | EPD | 60+ DPD within first 6 months on book | 30% weight · stress cap 10% | | Volume weight | Log scale on 90-day loan count | min(1.2, 0.7 + 0.3·log₁₀(loans÷60)) | | Composite | 100 × (0.4·nNCO + 0.3·nDPD + 0.3·nEPD) × vol. weight | capped at 100 | | Bands | High risk / watch / healthy | ≥70 · 50-69 · The scoring specification in full. Standard screens on top of the ranking: high-risk only (score ≥ 70) and top-five channels by volume. **Inference**The volume weight is doing two quiet jobs. Damping small channels to 0.70 keeps a 40-loan lot with one bad quarter from topping the ranking on noise. And because the weight only reaches 1.0 at around 600 loans, the high-risk band is deliberately hard to enter: a mid-sized channel has to be running near the stress caps on all three signals at once before it crosses 70. The band is built to flag conviction, not variance. ## Three bands, three different reads **High risk (score ≥ 70).** The channel is harming the book more than it helps. The standard read: cap or restrict new originations from the channel, look at dealer recourse on the existing paper, and pull a set of representative loans for underwriting review. **Watch (50-69).** The deterioration is visible but not yet conclusive. The standard read: pull a recent vintage sample and look for underwriting drift, and tighten stip verification or down-payment requirements before the channel migrates into the top band. **Healthy (below 50).** A channel to grow into. The useful move is diagnostic in the other direction: study what its stip stack and deal structure look like, and check whether the weaker channels differ in ways that explain the gap. ## A worked book: twelve synthetic channels Here is the method applied to a synthetic twelve-channel portfolio built to mirror a typical mid-sized subprime book: 1,388 loans, $17.9M of unpaid balance. Weighted by balance, the book runs 7.5% NCO, 5.1% 90+ DPD, 4.3% EPD. Every name and number below is synthetic and illustrative. | Channel | Loans (90d) | UPB | NCO | 90+ DPD | EPD | Score | Band | |---|---|---|---|---|---|---|---| | Auto Barn Motors · TX | 142 | $1.85M | 13.2% | 9.4% | 8.5% | 68 | Watch | | Pilot Point Motors · OK | 96 | $1.24M | 11.7% | 8.1% | 7.2% | 56 | Watch | | Coastal Drive · FL | 188 | $2.46M | 10.4% | 7.2% | 6.4% | 55 | Watch | | Lonestar Cars · TX | 74 | $0.92M | 9.9% | 6.8% | 5.9% | 44 | Healthy | | Sunrise Auto Group · CA | 121 | $1.58M | 8.1% | 5.4% | 4.6% | 39 | Healthy | | Phoenix Auto Plaza · AZ | 88 | $1.12M | 7.4% | 5.0% | 4.2% | 34 | Healthy | | Heritage Motors · NC | 156 | $1.98M | 5.8% | 3.9% | 3.1% | 28 | Healthy | | Northstar Auto · MN | 64 | $0.85M | 5.2% | 3.5% | 2.8% | 22 | Healthy | | Cascade Auto Sales · OR | 92 | $1.18M | 4.6% | 3.1% | 2.4% | 21 | Healthy | | Sterling Motorcars · WA | 108 | $1.42M | 4.1% | 2.8% | 2.0% | 19 | Healthy | | Riverside Auto · CA | 135 | $1.72M | 3.8% | 2.5% | 1.8% | 18 | Healthy | | Atlantic Auto Group · NJ | 124 | $1.58M | 3.4% | 2.2% | 1.5% | 15 | Healthy | Synthetic and illustrative. Scores computed exactly per the specification above. Note the strictness: even a 13.2% NCO channel lands at 68, inside the watch band, not high risk, because its 142-loan count damps the composite to 0.81 of its raw value. Loss attribution, worked book The three worst-scoring channels produce nearly half the book's net losses on less than a third of its volume. Synthetic twelve-channel book; shares computed from the worked table above. The read A book running **7.5% weighted NCO** sounds like one number. It is not. Inside the worked book, channels run from 3.4% to 13.2%, and the three worst channels carry **47.8% of the losses on 30.7% of the volume**. The portfolio average is not where the risk lives. The spread is. ## Why a handful of dealers carry the deterioration Deterioration in an indirect book rarely arrives evenly. Dealers differ in how disciplined their finance office is, in the customer base their lot attracts, and in how hard marginal deals get pushed through at month-end. Those differences compound: the channel with the loosest structure attracts the applications the tighter channels decline. The honest concentration measure is loss share against volume share. In the worked book, the worst channel produces 18.1% of the losses on 10.2% of the volume, a 1.8x ratio, and runs 1.8x the book's weighted-average NCO. Its six-month trend line is also the steepest in the book, up 3.4 points over the last three months of the series. **Inference**The forward-looking claim embedded in this method: because loss production is concentrated, volume decisions at two or three channels move the whole book. On the worked numbers, halving originations from the worst channel would remove roughly 9% of forward loss production while giving up about 5% of volume, an asymmetric trade. That projection assumes the channel's rates persist, which is exactly what the trend line exists to check. ## Limits What this method cannot see. **The stress caps are fixed choices.** 15 / 12 / 10 are defensible against published subprime benchmarks, but a deep-subprime book and a near-prime book should not share caps; a real deployment recalibrates them to the book's own vintage history. **EPD depends on honest data.** Deferrals, re-ages, and due-date changes silently repair early payment default; a channel whose paper gets serviced generously will score better than its underwriting deserves. **Small channels stay jumpy.** The log damping softens small-sample noise but cannot fix it, a 20-loan channel's score is an anecdote, not a statistic. **The score is descriptive, not causal.** It cannot distinguish a dealer problem from a geography or collateral problem that happens to share a lot. And **channel labels can hide common ownership**, two clean-scoring lots run by the same finance office are one exposure wearing two names. The score ranks where to look. It does not explain what you will find. **Sources & notes** Stress-cap levels are set with reference to published subprime auto benchmark families from Fitch Ratings, S&P Global, and TransUnion. All twelve channels in the worked example, names, states, and every number, are **synthetic and illustrative**; the dealer names are fictional and any resemblance to real businesses is coincidental. Composite scores, loss-attribution shares, and weighted averages are computed exactly per the published specification. LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned. Not investment, legal, or accounting advice. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). LR LendRisk Analytics Independent market research Continue reading [Methods · Note 04 Covenant runway: the month the *trigger breaks*](https://lendriskanalytics.com/covenant.html) [Methods · Note 05 The composite read: three numbers against *warehouse bands*](https://lendriskanalytics.com/tool.html) --- title: "The segments carrying the loss" url: https://lendriskanalytics.com/loss-drivers.html publisher: LendRisk Analytics kind: Page description: "A method note on segment-level loss attribution for direct auto books: the bin edges, flag thresholds, and cross-cut logic that show which credit bands, collateral profiles, terms, and geographies are carrying the net loss. Worked examples are synthetic." html: https://lendriskanalytics.com/loss-drivers.html --- # The segments carrying the loss [← All insights](https://lendriskanalytics.com/articles.html) Methods · Note 02 Method note · 7 min read Method note · Loss attribution # The segments carrying the loss. LendRisk Analytics · Method note · 2026 Every direct auto book reports one cumulative net-loss number, and that number is a weighted average of segments that behave nothing like each other. This note publishes, in full, a method for answering the question the aggregate cannot: *which* credit bands, collateral profiles, terms, and geographies are actually carrying the loss, and which combinations the single-axis view hides. All bin edges, formulas, and thresholds are stated exactly. All worked numbers are synthetic. ## Why the aggregate is the wrong unit In an indirect book, loss attribution naturally starts with the channel: which sources sent the paper that went bad. That is its own method (Note 03). A direct-to-borrower book has no channel axis at all. The underwriting decision belongs to the lender, so attribution has to come from the credit box itself, which FICO band, vehicle type, vehicle age, mileage bucket, advance rate, term, payment burden, and geography are running hot relative to the rest of the book. A book-level cumulative net loss of, say, 9% is not actionable on its own. It could be a uniformly mediocre book, or a mostly sound book subsidizing three segments running at twice the average. The credit-box response to those two situations is completely different, and the aggregate cannot tell them apart. Segment attribution exists to tell them apart. ## Four numbers per segment The method computes the same four quantities for every segment, on originated balance rather than loan count wherever dollars matter. **Cumulative net loss rate** is gross charge-offs minus recoveries, divided by the segment's originated balance. **Share of total loss** is the segment's net-loss dollars over the whole book's net-loss dollars, the materiality axis. **Default frequency**, the PD proxy, is charged-off plus repossessed units over all units in the segment. **Loss severity**, the LGD proxy, is net loss on the defaulted loans divided by those same loans' originated balance. A fifth number, 60+ days past due among still-active accounts, serves as the forward indicator. The frequency-severity split is the point of the exercise. A segment can carry outsized loss because too many of its loans default, deep credit tiers, stretched payment burdens, or because each default recovers badly, aged luxury collateral, very high mileage. The first is an approval and structure problem; the second is a collateral and advance-rate problem. Same loss dollars, different fix. ## The segment grid Bins are fixed, not data-driven. A segment definition that drifts with the book cannot be tracked across snapshots, so the cut points below stay constant and the book moves through them. These are the exact edges the method uses. | Dimension | Bin edges, as used | |---|---| | Credit (FICO) | Vehicle age falls back to model year against the as-of date when age is not stated directly. These edges are method choices, tuned to subprime auto; a near-prime book would cut the credit and LTV axes differently. ## The flag line: elevated and material A segment is flagged only when two conditions hold at once: its loss rate runs at **1.2× the book average or worse**, and it carries **at least 3.5% of the book's total net loss**, with a minimum of 20 loans in the bin. Both conditions matter. Without the materiality bar, a tiny 40-loan bucket at 3× book dominates the list while moving nothing; without the multiple, big segments get flagged just for being big. Flagged segments are then ranked by an impact score: **share of loss × (multiple − 1)**. That is, roughly, the loss the book would not have taken had the segment performed at book average, expressed in share points. A segment at 1.6× book carrying 21% of loss outranks one at 2.0× carrying 12%, which is the correct ordering for anyone deciding what to tighten first. Above 1.4× book, the flag is treated as severe. At book level, the context bands are: cumulative net loss above 8% reads elevated and above 13% severe; severity above 70% severe; 60+ DPD above 7% severe. In the per-dimension view, a bin above 1.35× book reads hot and below 0.7× reads cool. | Rule | Threshold, as used | |---|---| | Segment flag, elevated | loss rate ≥ 1.2× book | | Segment flag, material | ≥ 3.5% of total net loss | | Minimum sample | 20 loans (25 for cross-cuts) | | Ranking score | share of loss × (multiple − 1) | | Severe segment | ≥ 1.4× book | | Cross-cut flag | ≥ 1.5× book and ≥ 3% of loss | | Bin heat, dimension view | hot > 1.35× · cool The full flag logic. When nothing clears both segment bars, the honest output is that risk is diffuse, the loss is everywhere and nowhere, and the credit box is not the lever. **Inference**A flag is a statement about where the lever is, not a verdict on the loans. A flagged credit band says the approval line is the instrument; a flagged collateral bin says advance rates and valuation are; a flagged geography usually says the box is fine but its application is not uniform. The method locates the decision. It does not make it. The read The aggregate is calm because the majority is calm. In the synthetic example below, a book at 9.4% cumulative net loss is really **a handful of segments running 12-19% stacked on a majority running near 6%**. The decision-relevant number is the multiple against book, not the mean, the mean is what the multiple hides in. ## The cross-cuts the aggregate hides Single-axis attribution has a blind spot: risk factors arrive on the same loan. The method therefore crosses six fixed pairs, **credit × term, vehicle age × mileage, vehicle type × vehicle age, geography × credit, LTV × term, and affordability × credit**, and flags any intersection at 1.5× book or worse that carries at least 3% of total loss on 25+ loans. This is where the classic concentrations surface: deep-subprime credit written at 73 months and beyond, and old luxury collateral, where frequency and severity go bad together. Here is a worked read on a fully synthetic direct book, roughly 2,400 loans, cumulative net loss 9.4%, default frequency 14%, severity 58%. The top flagged segments, ranked by the impact score above: | # | Segment | Cum net loss | × book | Share of loss | Default freq | LGD | |---|---|---|---|---|---|---| | 1 | Term · 73 mo+ | 15.0% | 1.6× | 21% | 24% | 64% | | 2 | Credit · Synthetic and illustrative, no actual portfolio. Segments overlap: the same loan sits in several rows, so shares do not sum. Note the ranking: #1 beats #2 on impact (21 × 0.6 > 12 × 1.0) despite the lower multiple. One bin, two shares, the 73 mo+ term bucket In the synthetic book, loans written past 72 months are about 13% of originated balance but carry roughly one net-loss dollar in five. Synthetic, illustrative book. Not data from any actual lender or portfolio. The cross-cuts sharpen the same picture. In the worked book, **credit **Inference**Stacked risk factors on one loan behave closer to multiplicatively than additively in loss dollars. A borrower who is deep-subprime *and* at 84 months *and* above 145 LTV is not three moderate risks. It is one loan carrying all three, defaulting early with maximum balance at risk and minimum equity. The single-axis view disassembles that loan into bins; the cross-cut is what reassembles it. ## Where the bin edges come from The worked numbers on this page are synthetic, but the *structure*, where the lines are drawn and which direction the flags point, is not guesswork. It is set to match published, public data, none of it loan-level, borrower-level, or vehicle-history data: - The credit axis uses the 620 subprime line the Federal Reserve Bank of New York Consumer Credit Panel/Equifax uses, and the direction of the credit flags follows its finding that auto delinquency concentrates in below-620 borrowers, auto 90+ day delinquency reached 5.02% in 2025:Q3, the highest since 2020. - The term axis singles out 73 months and beyond because Experian reports roughly a third of loans now sit in the 73-84-month band, and its data shows subprime delinquency rising fastest there. - The sector loss level the multiples are measured against tracks the Fitch subprime auto ABS net-loss index and the Federal Reserve's May 2026 FEDS Note on subprime and BHPH lending; the negative-equity pressure behind the LTV axis follows published Edmunds negative-equity data. In other words: a reader who wants to know whether the 73-month flag or the sub-620 flag is a real phenomenon can check it against the public record, not take this note's word for it. The full list is under [Data & sources](https://lendriskanalytics.com/data.html). ## Limits What this method cannot see is as important as what it can. **It does not separate segment from seasoning.** Cumulative loss favors older vintages; a segment concentrated in early originations will read hot partly because it is old. A vintage-controlled cut is the check. **It does not establish causation.** Segments correlate, deep credit tiers also carry longer terms and higher LTVs, so a flagged bin may be carrying a neighbor's risk. **The materiality bars hide small, emerging problems** by design; a new 30-loan segment at 3× book will not surface until it grows. **Recovery lag flatters recent defaults**, understating current-period severity. **Input quality binds everything:** LTV depends on which book value was used at origination, and PTI on stated income. And the method says nothing about what the segment earns, a 1.5× loss segment written at a sufficiently higher yield may still be sound economics. That is a different analysis. **Sources & notes** The attribution method is published in full: every bin edge, formula, and threshold above is stated exactly as used. The cut points are calibrated to **public, non-proprietary data**, the [Federal Reserve Bank of New York](https://www.newyorkfed.org/microeconomics/hhdc) Household Debt & Credit report (Consumer Credit Panel/Equifax; 620 subprime line, auto delinquency by tier), [Experian](https://www.experianplc.com/newsroom/press-releases/2026/new-experian-automotive-report-shows-nearly-one-third-of-automot) State of the Automotive Finance Market (term-length and subprime-delinquency shares), the [Federal Reserve FEDS Note](https://www.federalreserve.gov/econres/notes/feds-notes/subprime-auto-lending-trends-in-buy-here-pay-here-auto-lending-20260508.html) on subprime/BHPH lending (May 2026), the Fitch subprime auto ABS net-loss index, and Edmunds negative-equity data. None of these are loan-level or borrower-level. The exact edges remain this publication's own method choices, reasonable practitioners draw them elsewhere, but the direction and materiality of every flag can be checked against those public series. All portfolio figures in the worked table, the cross-cut examples, and the chart are **synthetic and illustrative**; they describe no actual lender, book, or borrower. Full source list under [Data & sources](https://lendriskanalytics.com/data.html). LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned. Not investment, legal, or accounting advice. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). LR LendRisk Analytics Independent market research Continue reading [Methods · Note 03 Reading the *dealer channel*](https://lendriskanalytics.com/dealers.html) [Methods · Note 04 Covenant runway: *the month the trigger breaks*](https://lendriskanalytics.com/covenant.html) --- title: "Method & Data" url: https://lendriskanalytics.com/method.html publisher: LendRisk Analytics kind: Page description: "How LendRisk Analytics builds its method notes and write-ups: where the data comes from, how the synthetic worked examples are constructed, the math behind every metric, and what is illustrative versus market-sourced." html: https://lendriskanalytics.com/method.html --- # Method & Data [← Home](https://lendriskanalytics.com/) Method & data # What the numbers are, and what they are not. Every method note and write-up on this site is built to be defensible. This page documents where the data comes from, how the synthetic worked examples are constructed, the math behind each metric, and the clear line between what is illustrative and what is sourced from published industry data. 01 · The example data ## The sample books are synthetic. On purpose. The 2,000-loan example book, the direct-lender segment book, and the dealer scorecard portfolio are all **synthetic**. No real borrower, dealer, or lender data appears anywhere on this public site. The example loans are generated to mirror the structure of a real mid-market subprime book: realistic distributions of FICO, term, advance rate, vehicle age, mileage, affordability, and geography, with default and recovery behavior calibrated to the **level and shape of the published Fitch subprime auto ABS index** and ABS-EE-style loan-level distributions. Synthetic is the right choice here, not a limitation. It lets the write-ups demonstrate exactly how attribution and covenant math behave, without exposing anyone's portfolio. It also means the worked-example numbers are illustrative of **method**, never a claim about any specific lender's results. If a figure on this site is labeled **demo** or **blinded composite**, it came from a synthetic book and should never be read as a market statistic. Market statistics are cited separately, to a named source. 02 · The sample reports ## Sample reports use synthetic composite books. The two sample data reports walk one synthetic composite book through the full analysis. They are illustrative, built to show the format and the reasoning on made-up data, not drawn from any real portfolio. Every number in them is synthetic. The published articles and method notes apply the same reasoning to industry data and synthetic worked examples. 03 · The math ## Every metric, defined. - Net loss (charge-off − recovery) ÷ originated balance, per loan, aggregated by vintage, dealer, or segment. - EPD · early payment default First default within 3 months on book. A tell for underwriting quality, not seasoning. - LGD · loss given default Net loss ÷ defaulted balance. The severity side of loss, separate from frequency. - Default frequency Charge-off plus repossession over the cohort. - Vintage / static pool Loans grouped by origination quarter and aged at equal months-on-book, so newer paper is not flattered by being younger. - Roll / transition rate Share of a delinquency bucket that migrates to the next bucket month over month. - Covenant slope & runway Linear fit over the trailing 6 monthly readings, projected forward to the month each cap or trigger is crossed. - Segment flag A segment is surfaced only when it runs materially above the book average and carries a real share of total loss, so a tiny high-loss bucket cannot cry wolf. 04 · Benchmarks ## Where the market numbers come from. When a number is described as market data, it is cited to a named, public source. The write-ups carry their citations inline. The recurring benchmarks are: Fitch subprime auto ABS index S&P Global TransUnion Edmunds NY Fed NABD Manheim 05 · What the write-ups are, and are not ## Independent research, not a decision. The write-ups are **independent market research**: analysis built from public, named sources, walked through on synthetic worked examples. A worked underwriting example returns an **indicative** probability of default and expected loss to show how the framing works; it is not a credit approval, a score, or a guarantee. The state recovery-law breakdown is a **general reference** for understanding exposure, not legal advice; statutes change and edge cases exist, so verify with counsel before acting. Nothing here is an audit of any portfolio, an offer of credit, investment or legal advice, or a regulatory determination. It is research, published to be read and checked. 06 · Independence ## Published independently. LendRisk Analytics is an **independent research publication** with no position in, and no affiliation with, any company mentioned. Nothing here uses any proprietary data or systems. The work stands on the method and the sources, which are documented here so they can be checked. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). --- title: "The four levers of recovery, state by state" url: https://lendriskanalytics.com/repo-map.html publisher: LendRisk Analytics kind: Page description: "A method note on the four state-law levers that decide what a defaulted auto loan returns: self-help repossession, right-to-cure notice, deficiency judgments, and wage garnishment. Includes the full scoring weights and how all fifty states and D.C. land." html: https://lendriskanalytics.com/repo-map.html --- # The four levers of recovery, state by state [← All insights](https://lendriskanalytics.com/articles.html) Methods · Note 07 Method note · 7 min read Method note · Recovery law # The four levers of recovery, state by state. LendRisk Analytics · Method note · 2026 Two loans with the same borrower, the same vehicle, and the same default date can return very different amounts, because one was written in Georgia and the other in Louisiana. This note publishes the method behind our state recovery-law map: the four statutory levers that decide what a defaulted auto loan actually gives back, the exact weights assigned to each, and how all fifty states plus D.C. land when you score them. 1 State requires judicial process for every repossession (Louisiana) 20 Jurisdictions require a pre-repossession right-to-cure notice 4 States block wage garnishment for most consumer debt 28 Jurisdictions score a clean 100 across all four levers ## Recovery is a legal variable, not a portfolio constant When an auto loan defaults, the number you eventually record as recovery is the output of a legal pipeline, and the pipeline is built by the state, not by the contract. Four questions decide almost everything. Can you take the vehicle without a court order? Must you first send the borrower a notice and give them a window to cure? Once the vehicle is sold, can you pursue the shortfall as a deficiency judgment? And if you win that judgment, can you actually collect it from wages? A book spread across state lines is carrying a blended answer to those four questions, whether or not anyone has ever written it down. ## The four levers, and the weights on them The method scores each jurisdiction out of 100. A state earns points for each lever that runs in the recovering party's favor. The weights are the exact ones used in the map. | Lever | What earns the points | Weight | |---|---|---| | Self-help repossession | Permitted under UCC §9-609, the vehicle can be recovered without a court order, provided there is no breach of the peace | 35 pts | | No mandatory right to cure | No pre-repossession notice-and-cure window is required before taking the collateral | 20 pts | | Deficiency judgment | The shortfall after sale can be pursued as a deficiency judgment | 20 pts | | Wage garnishment | Garnishment is available to collect that deficiency from wages | 25 pts | | Total | A clean, fast, low-cost, collectable recovery environment | 100 pts | Scores map to four tiers: Strong 90-100 · Standard 70-89 · Friction 50-69 · Restrictive below 50. Three of the classifications carry named statutory anchors. Louisiana is the only state with no self-help repossession; recovery there runs through judicial executory process, so every repossession starts in court. The right-to-cure requirement covers 20 of the 51 jurisdictions, largely the states that enacted the Uniform Consumer Credit Code, plus California under the Rees-Levering Act. And Wisconsin bars a post-repossession deficiency only when the balance at default was $1,000 or less (Wis. Stat. 425.209), so on a typical auto contract a deficiency is generally still available there. **Inference**The 35/20/20/25 split is an analytical judgment, not a statute. It encodes a view: access to the collateral is worth more than any single collection right, and the practical ability to garnish is worth more than the paper right to a judgment. Re-weight the levers and the middle of the ranking shuffles, but the outliers stay outliers under any defensible weighting, because they are missing levers entirely, not merely points. ## How the country splits Score all 51 jurisdictions and only five distinct values occur: 100, 80, 75, 55, and 45. That is because the deficiency lever never differentiates in this dataset, every jurisdiction permits a deficiency on a typical auto balance, so its 20 points act as a floor. What actually moves the map is the cure requirement, the garnishment block, and Louisiana's judicial-only rule. The counts: **28 jurisdictions score a clean 100.** Eighteen score 80, meaning a cure notice is required but everything else is intact. Three score 75, Texas, Pennsylvania, and North Carolina, where no cure notice is required but wage garnishment is blocked. South Carolina scores 55, the only state that stacks a cure requirement on top of a garnishment block. Louisiana scores 45, alone in the Restrictive tier. Where 51 jurisdictions land on the recovery score Most of the map is benign. The differentiation lives in a handful of states, and it concentrates in the collection levers, not the repossession lever. **Strong · 90-100 **Standard · 70-89 **Friction · 50-69 **Restrictive · below 50 Segment widths proportional to jurisdiction counts. Scores computed from the lever weights above. **Inference**Because self-help repossession is near-universal, two books can post identical repossession experience and still diverge widely in what they ultimately collect. The levers that vary most across states are the ones that operate *after* the auction. A severity assumption that treats a Texas shortfall like a Georgia shortfall is assuming a collection right that Texas does not grant. ## Ten states that show the spread The full map carries all 51 jurisdictions; these ten cover every distinct score and every lever combination that occurs in the data. | State | Self-help | Right to cure | Deficiency | Garnishment | Score | Tier | |---|---|---|---|---|---|---| | Georgia | Allowed | Not required | Available | Available | 100 | Strong | | New York | Allowed | Not required | Available | Available | 100 | Strong | | California | Allowed | Required (Rees-Levering) | Available | Available | 80 | Standard | | Colorado | Allowed | Required (UCCC) | Available | Available | 80 | Standard | | Wisconsin | Allowed | Required | Available* | Available | 80 | Standard | | Texas | Allowed | Not required | Available | Blocked | 75 | Standard | | Pennsylvania | Allowed | Not required | Available | Blocked | 75 | Standard | | North Carolina | Allowed | Not required | Available | Blocked | 75 | Standard | | South Carolina | Allowed | Required | Available | Blocked | 55 | Friction | | Louisiana | Judicial only | Required | Available | Available | 45 | Restrictive | *Wisconsin bars a deficiency only when the balance at default was $1,000 or less (Wis. Stat. 425.209); on a typical auto contract it is generally available. The garnishment block in TX, PA, NC, and SC applies to most consumer debt, so a deficiency judgment there is far harder to convert to cash. ## One loan, three states To see the levers work, take one synthetic contract: a $11,400 balance at default, the vehicle recovered and sold at auction for $6,200, leaving a $5,200 shortfall. **Every number in this example is invented for illustration**, no real loan, borrower, or portfolio is depicted. The contract is identical in all three columns. The state is the only thing that changes. | Step | Georgia (100) | South Carolina (55) | Louisiana (45) | |---|---|---|---| | Path to the vehicle | Self-help, no court order | Self-help, after the cure window | Judicial executory process, court first | | Notice before acting | None required | Right-to-cure notice, borrower may reinstate | Filing precedes recovery | | Shortfall after sale | $5,200 claim available | $5,200 claim available | $5,200 claim available | | Collecting it | Wage garnishment available | Garnishment blocked, recovery effectively stops at the auction proceeds | Garnishment available, at the end of the judicial channel | Synthetic and illustrative. The paper deficiency is identical in all three states; the collectable deficiency is not. The read The recovery lever that matters most is the one used last. Self-help repossession is near-universal, so the front of the pipeline looks the same almost everywhere. What divides the map is whether the shortfall after auction is collectable at all. In four states the deficiency judgment is largely ornamental, **recovery effectively stops at the auction proceeds**, and in one state the process does not even start without a court. ## Limits This method sees four flags per state. Real statutes are not flags. Cure windows differ in length and in reinstatement terms; garnishment states differ in exemptions and caps, and "available" is never the same thing as "easy". The score also has no time axis: it does not measure how long a judicial process takes, and county-level docket speed can matter as much as the statute itself. It says nothing about breach-of-peace litigation exposure in self-help states, nothing about the voluntary-surrender share of a book, which changes how often these levers are even pulled, and nothing about contract- or program-specific terms, which control in practice. Finally, statutes drift. A static classification decays, and any specific repossession or collection question belongs with licensed counsel in the relevant state, not with a scoring model. **Sources & notes** Lever classifications summarize state statutory frameworks: self-help repossession under UCC §9-609 as adopted state by state; pre-repossession notice-and-cure requirements largely under the Uniform Consumer Credit Code in adopting states and, in California, the Rees-Levering Automobile Sales Finance Act; Wisconsin's small-balance deficiency bar under Wis. Stat. 425.209; and Louisiana's judicial executory process. Classifications are directional, one-flag-per-lever summaries, not statutory citations for any particular action. The worked example is **synthetic and illustrative**; no real loan, borrower, or portfolio is depicted. LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned. Not investment, legal, or accounting advice. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). LR LendRisk Analytics Independent market research Continue reading [Methods · Note 09 The economics of a *single deal*](https://lendriskanalytics.com/underwriter.html) --- title: "Recovery decay: what a day of delay costs" url: https://lendriskanalytics.com/repo-timing.html publisher: LendRisk Analytics kind: Page description: "A method note on repossession timing in BHPH and deep subprime: expected recovery equals vehicle value times the probability of custody, and the second term collapses far faster than the car depreciates. The full decay model, trigger ledger, and a worked synthetic table." html: https://lendriskanalytics.com/repo-timing.html --- # Recovery decay: what a day of delay costs [← Methods](https://lendriskanalytics.com/tools.html) Methods · Note 06 Method note · 6 min read Method note · Recovery & servicing # Recovery decay: what a day of delay costs. LendRisk Analytics · Method note · 2026 This note answers one question. On a delinquent BHPH or deep-subprime auto loan, what does each day of hesitation between *the file turning* and *the truck rolling* cost in expected recovery? Below is the full working model, the decay-curve assumptions, the trigger ledger, and a worked synthetic table, published openly so anyone with a book can check it against their own history. The companion essay is [the waiting tax](https://lendriskanalytics.com/insights/the-waiting-tax.html). 10% Of BHPH balances delinquent, Q3 2025, vs 3.8% traditional · Fed 16.6× More likely to be in active repossession than a traditional loan · Fed 4-5 d Median assignment-to-completion once repo is ordered · CFPB 27-38% Share of repo assignments that ever complete · CFPB ## Two curves, not one The core identity is simple. Expected recovery on day *d* past due = **net resale value(d) × probability of physical custody(d) − recovery costs(d)**. In buy-here-pay-here the asset is the loan and the collateral is the only real protection, so everything about repossession timing reduces to how those two curves move against each other. They move at completely different speeds. Used vehicles are depreciating slowly right now, roughly **1% a month** in 2026 wholesale data (Black Book and Manheim indices). The odds you ever get the car back are the fast term: they collapse the moment a delinquent borrower stops answering, lets insurance lapse, disables the GPS, and starts hiding the unit. The car is not the bleed. The custody probability is, and it is the term most operators never put a number on. One denominator discipline matters here. Net resale value is what the unit itself would fetch at sale. It is deliberately *not* multiplied by an ABS-style loan-balance recovery rate, the 30s,40s percent figures lenders report measure recovery against the loan balance, a different denominator, and folding them in would double-count loss severity. ## The slow term: what the car does At the 2026 wholesale norm of ~1% a month, ninety days of delay costs a unit about 3% of its value from depreciation alone. The model adds a condition drag for wear, miles, and neglect while the car sits with a non-paying borrower, about 0.095% of value per day past due, capped at 14%. Even at a stress case of 1.5% or more a month, the vehicle side of the identity moves single-digit percentages over a full quarter. If that were the whole story, waiting would be cheap. ## The fast term: the custody curve Custody probability is modeled as a logistic slide from a high starting point down to a posture-dependent floor. The parameters below are the actual presets the model runs on, three borrower postures, each with a starting probability, a floor, a midpoint (the day past due at which half the slide has happened), and a steepness scale. | Borrower posture | Odds at day 0 | Floor | Midpoint of slide | Steepness scale | |---|---|---|---|---| | Cooperative | 97% | 66% | day 78 | 15 d | | Going quiet | 95% | 30% | day 46 | 10 d | | Hiding it | 90% | 7% | day 28 | 7 d | Modeled scenario priors, not measured rates. On top of the posture curve, each hard trigger present cuts custody odds a further 16% (multiplicative, capped at a 55% total penalty, with a 2% floor). GPS dormancy or device tampering escalates the file one posture worse automatically. Read the "hiding it" row again. Half the slide is done by day 28, and the floor is 7%. On a concealed unit, a month of delay converts a near-certain recovery into a long-shot skip file, while the car itself has lost barely 1% of its value. **Inference**The shape of these curves is the interpretive step, but it is anchored to a public pattern: CFPB repossession data shows capture is heavily front-loaded, a median of ~4-5 days from assignment to completion when it completes, yet only ~27-38% of assignments ever complete, and ~96% of completions land within 90 days. Repossession outcomes look bimodal. Either you secure the unit early, or the file drifts into a long tail where most assignments never convert. A steep early decay with a low floor is the curve that pattern implies. ## The trigger ledger: stage the file vs repo now The model does not treat days past due as the decision variable. Behavior is. Two tiers of signals govern the file's footing. | Tier | Signal | |---|---| | Hard · repo footing | GPS shows the unit dormant 72h+ or out of the area | | Hard · repo footing | Starter-interrupt or GPS tampered with or disabled | | Hard · repo footing | Insurance lapsed, collateral now uninsured | | Hard · repo footing | Payment reversed / NSF and no contact since | | Hard · repo footing | Broke a second promise-to-pay | | Hard · repo footing | No contact for 10+ days (skip behavior) | | Soft · stage the file | Missed within the first 3 payments (early-payment default) | | Soft · stage the file | First scheduled payment missed | | Soft · stage the file | Partial payment only | | Soft · stage the file | Broke first promise-to-pay | | Soft · stage the file | Phone disconnected / mail returned | | Soft · stage the file | Paying later each month (deteriorating pattern) | The staging logic: any hard trigger puts the file on repo footing, dispatch if lawful, or serve the right-to-cure notice today and pre-stage the recovery agent if not. Two or more soft signals, or any early-payment default, means stage and serve: open the legal window now so a hard trigger becomes a same-day repo instead of a three-week scramble. One soft signal is a watch-close; log it and tighten contact. The clock that gates all of this is legal, not behavioral. In the model, the cure notice goes out on day 10 past due, and the lawful repo day is that notice day plus the state and contract cure window. Some states allow repossession at default with no notice; others require a cure or notice period, on the order of ~10 days in Rhode Island and Maryland, ~15 in Pennsylvania, ~21 in Illinois. The lawful day is the best day the decay curve will ever offer, which is why the staging logic is built around reaching it pre-positioned rather than starting the process there. The read Auction math tells you what the car is worth. Timing math tells you whether you will ever hold the car. Because **custody probability collapses far faster than the vehicle depreciates**, the recovery decision is won or lost in the days around the lawful gate, not at the sale. The operators who treat the cure notice as the start of the process, rather than the end of a staged one, pay the difference. ## A worked example: the decay table The table below runs the model end to end on a **synthetic** default file: an $8,500-net unit, 1.0%/month depreciation, "going quiet" posture with no hard triggers, cure notice served day 10, a 15-day cure window (lawful day 25), $875 in fixed recovery and disposal costs, and $25/day of holding cost from the notice day. Every figure is illustrative. | Day past due | Unit net value | Custody odds | Costs accrued | Expected net recovery | |---|---|---|---|---| | Day 10 · notice served | $8,391 | 93% | $875 | $6,952 | | Day 25 · lawful day | $8,229 | 88% | $1,250 | $5,984 | | Day 40 | $8,069 | 72% | $1,625 | $4,182 | | Day 60 | $7,857 | 43% | $2,125 | $1,242 | | Day 90 | $7,543 | 31% | $2,875 | −$552 | | Day 120 | $7,236 | 30% | $3,625 | −$1,451 | Synthetic and illustrative throughout. Note the columns: over 110 days the unit loses about 14% of its value, while the custody odds lose two-thirds of theirs. Past day 90 the expected recovery on this file is negative, the holding and recovery costs exceed what the collapsing odds are worth. The waiting tax, days 25 to 60 On the synthetic file above, $5,984 of expected recovery is on the table on the lawful day. Holding to day 60 leaves $1,242 of it. Synthetic example, model output. Custody odds are modeled scenario priors, not measured rates. The decomposition is the point. At 30 days past due on this file, the next 30 days of waiting cost the unit about 4% of its value, and cost the custody odds about 49% of theirs. Almost the entire waiting tax is the second term. Early custody also keeps the borrower's redemption and reinstatement options alive: in CFPB data, redemptions run ~22-34% of completed repossessions, and most happen within 30 days of the repo. **Inference**The same CFPB data shows ~94% of disposals still end in a deficiency, meaning the sale of the unit rarely clears the balance, and the deficiency claim against a deep-subprime borrower is worth little in practice. If the recovered vehicle is, realistically, most of what a BHPH operator will ever collect on a defaulted loan, then the custody-odds curve is not one input among many. It is the loss model. ## Limits Honesty about what this method cannot see. **The custody curves are priors, not measurements.** No public source reports a daily custody probability by borrower posture, days past due, GPS status, or broken promises; the posture parameters and trigger penalties are calibrated to the directional CFPB and Fed findings, not fitted to an event-level repossession dataset. The structural claim, custody odds collapse faster than vehicle value, so timing beats auction math, is well supported. The exact coordinates are estimates that should be recalibrated against an operator's own repo history before any dollar figure is trusted. The ~1%/month depreciation figure is a market-wide wholesale index, not any particular lane or segment. The model ignores agent capacity, lot constraints, bankruptcy stays, and the economics of reinstatement. And the legal gate is state- and contract-specific: cure, notice, and self-help rules vary widely, and nothing here substitutes for confirming them in the relevant jurisdiction. **Sources & notes** BHPH delinquency (10% vs 3.8%), the 16.6× active-repossession intensity, average origination ($15.4k) and APR (25.4%) are from the Federal Reserve's [FEDS note on buy-here-pay-here lending (May 2026)](https://www.federalreserve.gov/econres/notes/feds-notes/subprime-auto-lending-trends-in-buy-here-pay-here-auto-lending-20260508.html). Assignment-to-completion timing (median ~4-5 days), completion share (~27-38%), the 90-day completion concentration (~96%), redemption share (~22-34%, most within 30 days), deficiency incidence (~94% of disposals), and typical recovery costs (agent ~$500, disposal ~$300-350, forwarders adding ~$30-100+) are from CFPB auto-repossession data, 2022-2025. Wholesale depreciation near ~1%/month is from Black Book and Manheim index data. State cure and notice windows (e.g. RI ~10, MD ~10, PA ~15, IL ~21 days) are set by state statute and contract; see the companion note on [state recovery law](https://lendriskanalytics.com/repo-map.html). The worked decay table, the split bar, and all custody-probability figures are **synthetic and illustrative** model output, not measured rates. LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned. Not investment, legal, or accounting advice. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). LR LendRisk Analytics Independent market research Continue reading [Methods · Note 07 The four levers of *recovery*, state by state](https://lendriskanalytics.com/repo-map.html) --- title: "Sample Walkthrough \u00b7 Direct Lender" url: https://lendriskanalytics.com/sample-report-direct.html publisher: LendRisk Analytics published: 2026-04-30 kind: Page description: "A blinded direct-lender portfolio walkthrough. No dealers to blame, so loss attribution comes from the credit box itself: which FICO band, term, LTV, vehicle, affordability, and geography are eating the book. We find the slope, rank the segments, project the covenant breach, and tighten the box that stops it." html: https://lendriskanalytics.com/sample-report-direct.html --- # Sample Walkthrough · Direct Lender Worked example · direct lender A model data report for a lender that funds borrowers directly. Numbers drawn from a synthetic direct-lender dataset. Names and identifying details are invented. How they lend: [**Indirect lender** Dealer attribution](https://lendriskanalytics.com/sample-report.html) [**Direct lender** Borrower segments](https://lendriskanalytics.com/sample-report-direct.html) LendRisk Analytics · Worked Example · Synthetic Data # No dealer to blame, so we read the credit box. A worked example on a synthetic direct-to-borrower book · Every figure is illustrative **Prepared**  LendRisk Analytics **Date**  May 2026 **As-of**  April 30, 2026 **Loans**  2,400 **Line size**  $40M senior warehouse Worked example · Synthetic 2,400 Active loans $38.7M Originated balance 9.6% Cum net loss 62% Loss severity · LGD 561 WA FICO How to read this This is the direct-lender version of the dealer report, for a book that funds borrowers directly. There is no dealer to attribute the loss to, so the attribution comes from the credit box itself: which FICO band, which term, which advance rate, which vehicle, which affordability tier, which geography is running hot. It is here to show what the analysis looks like. The book gets read for the slope the aggregate hides, then every segment is ranked by how much of total loss it actually carries, not how big it is. Same tools, borrower axis instead of dealer axis: the dashboard, the segment loss-driver intelligence, the vintage static pool, the vehicle-to-term matcher, and the covenant projector. 01 · We open the tape On paper, you look completely fine. Tool · Portfolio Dashboard Walking the book Start with the headline numbers. Cumulative net loss running **9.6%**, weighted FICO **561**, and a loss severity (LGD) of **62%**, meaning when a loan goes down the book recovers about 38 cents on the dollar of balance. The NCO covenant cap is 13.5%, so today it sits comfortably under it. On a direct book the loss splits into two pieces worth separating right away: how often loans default (frequency) and how badly they hurt when they do (severity). The aggregate hides which one is moving. That distinction is the whole game on a book with no dealers to point at. *A book is a movie, not a photograph.* The number that matters isn't where you are. It's how fast you're moving, and which segments are pushing. 02 · The slope, not the snapshot Watch the column nobody puts on the dashboard. Tool · Covenant Projector · trend read What I'm pointing at Look at the "Monthly Δ" column. Net loss is not parked at 9.6%, it is climbing **0.65 points every month.** 90+ DPD is climbing 0.30. The level is comfortable. The trajectory is not. That slope is a straight line fit through your last six monthly readings. | Metric | Current | Monthly Δ | Covenant cap | Headroom | Status | |---|---|---|---|---|---| | NCO (annualised) | 9.60% | +0.65 pp | 13.50% | 3.90 pp | ⚠ 6 months | | 90+ DPD | 4.20% | +0.30 pp | 6.00% | 1.80 pp | ⚠ 6 months | | Loss severity (LGD) | 62% | +0.5 pp | n/a | n/a | ⚠ Elevated | | WA FICO | 561 | −1.4 / mo | 545 floor | 16 pts | ✓ Watch | | Advance rate (blended) | 118% LTV | +0.4 pp | n/a | n/a | ⚠ Drifting | Benchmark: Fitch subprime auto ABS, Jan 2026, NCO 9.81% · 60+ DPD 6.65%. At market on the level, above market on the rate of change, and your advance rate is drifting up. At this slope, your book *breaches its NCO cap in six months.* That puts it in October. 03 · Which segments are eating the book No dealers. So we rank the credit box. Tool · Segment Loss-Driver Intelligence The direct-lender version of attribution On an indirect book the attribution would be a dealer scorecard. This book has no dealers, so every loan gets binned across the dimensions the lender actually controls, credit band, term, advance rate, down payment, geography, vehicle, affordability, and rank each segment by its **share of your total loss times how far above the book average it runs.** A small bucket that loses a lot does not get to cry wolf; this is loss you can feel. | Segment | Loans | % of loss | Net loss | Default freq | LGD | Flag | |---|---|---|---|---|---|---| | Credit FICO Book average net loss 9.6% · segment flagged when it both runs ≥1.2× the book and carries ≥3.5% of total loss · LGD = loss on defaulted balance. Reading it back to you Two things jump out. First, your **deep-subprime credit band under 520** is doing the heavy lifting, a quarter of your total loss at 3.1× the book, and it fails on both axes: it defaults more often *and* recovers worse. Second, look at the LGD column on **LTV 145%+ and 13-year-old vehicles**, 71% and 73%. Those segments do not default the most, but when they do you barely recover. That is an advance-rate and collateral problem, not a borrower problem. So your loss has two separate engines: a frequency engine in deep-subprime credit, and a severity engine in thin-equity, old-collateral loans. You fix them with two different levers. Three segments, a fifth of the book. *Fifty-nine percent of the damage.* 04 · The combinations the average hides It's never one box. It's two stacked. Tool · Segment Intelligence · cross-cut Why single dimensions undersell it A single segment understates the risk, because the real damage lives where two bad attributes stack. A 73-month term is elevated. Deep-subprime credit is elevated. Put a 73-month term *on* a sub-520 borrower and the loss is not additive, it compounds. Here are the intersections carrying outsized loss. FICO < 520 × Term 73mo+ 2.6× book 11% of loss 34% default 148 loans ** LTV 145%+ × Vehicle 13yr+ 2.3× book 8% of loss 77% LGD 96 loans ** ZIP Tier 4 × FICO < 540 1.9× book 7% of loss 27% default 171 loans ** Zero down × PTI 20%+ 1.8× book 6% of loss 25% default 134 loans ** The credit box looks fine one rule at a time. *The losses live where two rules overlap.* 05 · When it started The damage has a date stamp. Tool · Vintage Static Pool Is it still happening Knowing which segments is half of it. Knowing when tells you whether your underwriting has already drifted back or is still loose. We line up every origination quarter as its own static pool and age them at the same months on book. | Vintage | Loans | Orig $ | Avg MOB | 60+ DPD | Charged off | Cum net loss | |---|---|---|---|---|---|---| | 2023 Q1 | 438 | $7.0M | 39 | 6.3% | 9.1% | 9.6% | | 2023 Q2 | 441 | $7.1M | 36 | 6.6% | 9.0% | 9.4% | | 2023 Q3 ▲ | 452 | $7.3M | 33 | 9.0% | 11.5% | 12.7% | | 2023 Q4 ▲ | 447 | $7.2M | 30 | 9.4% | 11.0% | 12.1% | | 2024 Q1 | 451 | $7.3M | 27 | 6.1% | 5.4% | 6.3% | | 2024 Q2 | 171 | $2.8M | 24 | 4.9% | 3.2% | 3.5% | ▲ Flagged cohorts: 2023 Q3 and Q4 exceed 12% cumulative loss at month 30-33, about 1.4× the 2023 Q1-Q2 rate at the same seasoning. Common factor: a loosening of the term and advance-rate caps in the back half of 2023. Reading it back to you The break is the **back half of 2023**, when the share of 73-month-plus loans and 145%+ advances spiked in your originations. 2024 is already coming in cleaner, so the box partly self-corrected, but the loose 2023 vintage is still on your book bleeding, and it lines up exactly with the hot segments from the last two sections. 06 · Why those loans default You didn't underwrite a borrower. You underwrote a car. Tool · Vehicle-to-Term Matcher The mechanism behind the severity This explains the LGD column. Your highest-severity segments are **long terms and high advances on old vehicles.** Put the depreciation curve next to the amortization curve and the loss draws itself. Loan balance vs. vehicle value · 73mo+ term, 11-year-old vehicle, 145% advance (illustrative) Loan balance (73mo+ amortization) Vehicle value (depreciation) Where defaults cluster Reading it back to you At a 145% advance you are underwater the day the loan funds. On an 11-year-old vehicle the value falls off a cliff while a 73-month note barely moves early. The negative-equity gap is widest exactly where these loans default, **months 18 to 36.** A borrower who hits a bump there cannot sell or refinance out, the car is worth thousands less than they owe, so you get the keys back and recover 30 cents on the dollar. That is your 71-73% LGD, drawn as a picture. The term outran the metal. *The loan runs to 73; the car was gone by 36.* 07 · Where this ends Now we put it on the calendar. Tool · Covenant Breach Projector The part the bank cares about most Take the slopes measured above and run them to the line. Same math a warehouse bank will run, just earlier, while the inputs can still change. NCO Covenant · Breach projected 6 months Current 9.6% · cap 13.5% · slope +0.65 pp/mo Projected breach: **October 2026** 2 of 6 mo consumed 90+ DPD Trigger · Breach projected 6 months Current 4.2% · trigger 6.0% · slope +0.30 pp/mo Projected breach: **October 2026** 2 of 6 mo consumed FICO Floor · Compliant 16 pt headroom Current 561 · floor 545 · drift −1.4 pts/mo Projected floor: **11 months out** ~3 of 16 pts consumed Post-fix · NCO slope 18+ months After tightening the three credit-box rules: slope +0.23 pp/mo Breach pushed beyond 18-month horizon Line preserved Reading it back to you NCO and 90+ DPD both break in **month six. The same month. October.** A breach trips your cash sweep, the sweep halts new originations, and you lose the line right when you would need it. The fourth card is this same book after we tighten the box. Hold that thought. 08 · The fix Three rules, not a book-wide pullback. Tool · Segment Intelligence + Underwriting policy Why not just stop lending The panic move is to slam the whole credit box shut and watch your volume collapse. You do not need to. Because we attributed the loss to named segments and combinations, you can re-cut the box surgically: tighten the three rules carrying the loss, and keep funding everything else exactly as you do today. Credit-box fix · modelled Three changes, applied to **new originations only**: floor FICO at 540 when term exceeds 72 months, cap LTV at 130% on vehicles aged 10 years or older, and require a minimum down payment when PTI exceeds 20%. Existing paper stays on the book under enhanced monitoring. Everything outside these rules funds untouched, that is roughly 80% of your current volume. +0.65 → +0.23 Monthly NCO slope ~135 bps Loss avoided · new vintages ~$1.7M Covenant headroom restored Reading it back to you Those three rules sit right on top of the hot segments and the worst combinations. They flatten your slope from 0.65 to 0.23 a month, push the breach past the 18-month horizon, and cut roughly 135 basis points of expected loss out of your new vintages, all while leaving about 80% of your volume untouched. The green covenant card from the last section is this scenario. Three rules tightened. *The line preserved. Eighty percent of volume untouched.* 09 · What you do Monday Three actions. One is immediate. Output · the page you keep What the analysis points to Here is where the numbers lead. Short enough to push to an underwriting team in a week, documented enough to stand up on a bank's next surveillance call. 1 Immediate · This week Floor FICO at 540 on terms over 72 months The sub-520 × 73mo+ combination is your single worst pocket, 2.6× the book and 34% default. Stop writing that intersection on new applications today. This one rule does most of the work to flatten your slope and remove the October breach risk. Document the change and the data behind it for the bank. 2 30 days · Advance-rate policy Cap LTV at 130% on vehicles 10 years and older This is your severity lever, not your frequency lever. The 145%+ LTV and 13yr+ vehicle segments do not default the most, they hurt the most when they do, at 71-73% LGD. Capping the advance on old collateral keeps you from being underwater on day one, which is the whole reason recoveries are so thin. Pair it with the 60-month term cap on the same vehicle age. 3 Ongoing · Monthly Re-run the segment scan monthly; watch ZIP Tier 4 and zero-down ZIP Tier 4 and zero-down sit above the book but below the urgent threshold. Both bear watching the next two cycles. If either crosses 1.8× the book on rising volume, add a down-payment or geography overlay. A monthly segment refresh catches a drifting box before it reaches the aggregate, which is the whole point: you find it here, not on the bank's call. **Method & lineage.** Net loss = (charge-off − recovery) ÷ originated balance, computed per loan and binned across credit, term, LTV, vehicle type, vehicle age, mileage, affordability (PTI), down payment, income type, and geography (ZIP tier). A segment is flagged when it loses materially faster than the book average *and* carries a real share of total loss, so a tiny high-loss bucket does not cry wolf. Combinations cross two dimensions to surface concentrations the single-axis view misses. Default frequency = charge-off plus repo over the segment; LGD = net loss over defaulted balance. Covenant slope is a linear fit over the prior 6 months of monthly NCO readings. The balance-versus-value chart in section 06 is illustrative of the structure, not a per-loan plot. This was prepared on a synthetic direct-lender book for illustration; the method is the same one a risk desk would apply to a live book. LendRisk Analytics Independent market research the research desk LendRisk Analytics · lendriskanalytics.com Worked Example · May 2026 Synthetic data · Illustrative only The point The average hides the answer. On a direct book the loss is never spread evenly. It lives in a few corners of the credit box, where two bad attributes stack, and the book-level number smooths all of it away. Ranking the segments by what they actually cost is what turns a calm aggregate into a decision the underwriting team can act on. --- title: "Sample Walkthrough" url: https://lendriskanalytics.com/sample-report.html publisher: LendRisk Analytics published: 2026-04-30 kind: Page description: "A worked example on a synthetic $40M subprime auto book. It shows how to find the slope the aggregate hides, attribute it to three dealers, date it by vintage, and project the covenant breach. Illustrative analysis on made-up data, not a real portfolio." html: https://lendriskanalytics.com/sample-report.html --- # Sample Walkthrough Data report · synthetic composite This is a model data report, written the way the analysis usually unfolds. Numbers drawn from a synthetic 2,000-loan demo book. Names and identifying details are invented. How they lend: [**Indirect lender** Dealer attribution](https://lendriskanalytics.com/sample-report.html) [**Direct lender** Borrower segments](https://lendriskanalytics.com/sample-report-direct.html) LendRisk Analytics · Worked Example · Synthetic Data # How to read a subprime book. A worked example on a synthetic composite book · Every figure is illustrative **Prepared**  LendRisk Analytics **Date**  May 2026 **As-of**  April 30, 2026 **Loans**  2,000 **Line size**  $40M senior warehouse Worked example · Synthetic 2,000 Active loans $32.4M Originated balance 9.6% NCO · annualised 4.2% 90+ DPD 574 WA FICO How to read this This is a worked example: one synthetic book read the same way a warehouse bank's surveillance team eventually reads it, except **caught early, while there is still time to act.** It is here to show what the analysis looks like, nothing more. One synthetic book run through five tools: the dashboard, the dealer scorecard, the vintage static pool, the vehicle-to-term matcher, and the covenant projector. Read it top to bottom. It moves from "this book looks fine" to the exact dealers, the exact month, and the three moves that keep the line open. 01 · We open the tape On paper, you look completely fine. Tool · Portfolio Dashboard Walking the book Start with the headline numbers, the same five at the top of this page. Net charge-offs running **9.6% annualised**, 90+ delinquency at **4.2%**, weighted FICO **574**. The NCO covenant cap is 13.5%. So today the book sits almost four full points under its cap. If we stopped right here, you would feel fine. Your bank would feel fine on the snapshot. The note would end here. Most reviews stop here, which is exactly why most problems are found by the lender's bank instead of the lender. *A book is a movie, not a photograph.* The number that matters isn't where you are. It's how fast you're moving, and in which direction. 02 · The slope, not the snapshot Watch the column nobody puts on the dashboard. Tool · Covenant Projector · trend read What I'm pointing at Look at the "Monthly Δ" column. Your NCO is not parked at 9.6%, it is climbing **0.65 points every month**. Your 90+ DPD is climbing 0.30. The level is comfortable. The trajectory is not. That slope is a straight line I fit through your last six months of readings. Carry it forward and the comfortable gap to your cap closes faster than anyone in the room expects. | Metric | Current | Monthly Δ | Covenant cap | Headroom | Status | |---|---|---|---|---|---| | NCO (annualised) | 9.60% | +0.65 pp | 13.50% | 3.90 pp | ⚠ 6 months | | 90+ DPD | 4.20% | +0.30 pp | 6.00% | 1.80 pp | ⚠ 6 months | | WA FICO | 574 | −1.2 / mo | 560 floor | 14 pts | ✓ Watch | | 60+ DPD | 7.10% | +0.40 pp | n/a | n/a | ⚠ Elevated | | Advance rate (blended) | 112% LTV | stable | n/a | n/a | ✓ In policy | Benchmark: Fitch subprime auto ABS, Jan 2026, NCO 9.81% · 60+ DPD 6.65%. You are roughly at market on the level, and above market on the rate of change. At this slope, your book *breaches its NCO cap in six months.* That puts it in October. 03 · Who is doing the damage Is the whole house on fire, or three rooms? Tool · Dealer Scorecard The question that changes the plan This is the question you ask the second you see the slope: is my whole book deteriorating, or is a handful of dealers dragging the average. Because the answer decides whether you reach for a sledgehammer or a scalpel. So we attribute. Every dealer, ranked by what they are actually costing you, not by volume, by loss and by how early their paper goes bad. | Dealer | Loans | % of book | Net loss | EPD rate | Severe 60+ | Flag | |---|---|---|---|---|---|---| | Dealer N | 92 | 4.6% | 16.4% | 29.3% | 38.1% | Toxic | | Dealer H | 148 | 7.4% | 15.1% | 27.0% | 35.8% | Toxic | | Dealer C | 200 | 10.0% | 14.8% | 27.8% | 34.2% | Toxic | | Dealer F | 118 | 5.9% | 9.4% | 9.1% | 18.3% | Watch | | Dealer M | 95 | 4.8% | 8.8% | 8.4% | 17.1% | Watch | | 13 remaining dealers | 967 | n/a | 5.1% | 6.8% | 13.2% | Clean | 18 dealers total · EPD network median 7.2% · the three toxic dealers run 3.9× the network EPD · HHI 1,840 (moderate concentration) Reading it back to you Three names do almost all of the work: **Dealers C, H, and N.** Together they are 22% of your originations and **51% of your severe delinquency.** Their early-payment-default rate is around 28%, nearly four times your network median. That is the tell. EPD this high means these loans were not good loans that went bad. They were bad the day they were written. Meanwhile your 13 clean dealers are running 5.1% loss, comfortably under market. You do not have a book problem. You have a three-dealer problem hiding inside a book-level average. Twenty-two percent of the book. *Fifty-one percent of the damage.* 04 · When it started The damage has a date stamp. Tool · Vintage Static Pool Why timing matters Knowing who is half the answer. Knowing when tells you whether it is still happening or already behind you. So we line up every origination quarter as its own static pool and watch each one age at the same number of months on book. That strips out the "newer loans look better because they are younger" illusion. | Vintage | Loans | Orig $ | Avg MOB | 60+ DPD | Charged off | Cum net loss | |---|---|---|---|---|---|---| | 2023 Q1 | 371 | $5.9M | 39 | 6.1% | 9.4% | 9.8% | | 2023 Q2 | 368 | $5.8M | 36 | 6.4% | 8.9% | 9.3% | | 2023 Q3 ▲ | 374 | $5.9M | 33 | 8.8% | 11.2% | 12.4% | | 2023 Q4 ▲ | 369 | $5.8M | 30 | 9.1% | 10.8% | 11.9% | | 2024 Q1 | 377 | $6.0M | 27 | 5.9% | 5.2% | 6.1% | | 2024 Q2 | 141 | $2.1M | 24 | 4.8% | 3.1% | 3.4% | ▲ Flagged cohorts: 2023 Q3 and Q4 exceed 11% cumulative loss at month 30-33, about 1.4× what 2023 Q1-Q2 did at the same seasoning. Common factor underneath: 72-month terms, 8-10 year vehicles, concentrated in Dealers C, H and N. Reading it back to you The break is the **back half of 2023.** Your Q3 and Q4 pools are running 12% cumulative loss at month 30, while your older 2022 and early-2023 paper sat near 9% at the same age. Same dealers underneath, same loan structure. And notice 2024 is already coming in cleaner, that tells me whatever changed in late 2023 has partly self-corrected, but the bad vintage is still on your book bleeding. 05 · Why those loans go bad You didn't underwrite a borrower. You underwrote a car. Tool · Vehicle-to-Term Matcher The mechanism, so you can defend the finding Here is the why, because when the bank asks, you want to say you understand the cause, not just that you spotted the symptom. The flagged paper is **72-month terms on 8 to 10 year old vehicles.** Put the depreciation curve next to the amortization curve and the problem draws itself. Loan balance vs. vehicle value · 72-month term, 9-year-old vehicle (illustrative) Loan balance (72-mo amortization) Vehicle value (depreciation) Where defaults cluster Reading it back to you You advanced above the car's value on day one, that is your 112% blended LTV. On a 9-year-old vehicle the value falls off a cliff while a 72-month note barely moves in the early years. The negative-equity gap is widest right where these loans actually default, **months 18 to 36.** A borrower who hits a bump there cannot sell or refinance their way out, the car is worth thousands less than they owe, so the keys come back instead. The term outran the metal. *The loan runs to 72; the car was gone by 36.* 06 · Where this ends Now we put it on the calendar. Tool · Covenant Breach Projector The part the bank cares about most This is the slide your warehouse bank will care about more than any other, because these are their covenants. I take the slopes we measured and run them to the line. Same math they will run, just earlier. NCO Covenant · Breach projected 6 months Current 9.6% · cap 13.5% · slope +0.65 pp/mo Projected breach: **October 2026** 2 of 6 mo consumed 90+ DPD Trigger · Breach projected 6 months Current 4.2% · trigger 6.0% · slope +0.30 pp/mo Projected breach: **October 2026** 2 of 6 mo consumed FICO Floor · Compliant 14 pt headroom Current 574 · floor 560 · drift −1.2 pts/mo Projected floor: **12 months out** ~2 of 14 pts consumed Post-intervention · NCO slope 18+ months After pausing Dealers C, H, N: slope +0.22 pp/mo Breach pushed beyond 18-month horizon Line preserved Reading it back to you NCO and 90+ DPD both break in **month six. The same month. October.** A breach trips your cash sweep, the sweep halts new originations, and you lose access to the line at the exact moment you would need it to grow out of the problem. Your FICO floor is not the worry. That is a year out. The fourth card is the one I like ending on, because that is this same book after we act. Hold that thought. 07 · The fix A scalpel, not a sledgehammer. Tool · Dealer Scorecard + Underwriting policy Why not just pull back The instinct under covenant pressure is a book-wide pullback, cut everyone. That is the panic move, and it punishes your 13 clean dealers for the sins of three. It also kills the yield you need to recover. Because we attributed the loss to named dealers, we can do something far more precise. Intervention outcome · modelled Pause new originations from **Dealers C, H and N** only. New submissions on hold, existing paper stays on the book under enhanced monitoring. Nothing else changes. The remaining 15 dealers keep funding without interruption. +0.65 → +0.22 Monthly NCO slope ~140 bps Yield recovered ~$1.8M Covenant headroom restored Reading it back to you That one move flattens your slope from 0.65 to 0.22 a month, which pushes the breach from October out past the 18-month horizon. You recover roughly 140 basis points of yield and about 1.8 million of covenant headroom versus the book-wide cut, and your healthy dealers never feel a thing. That fourth covenant card from the last section, the green one, that is this scenario. Three dealers paused. *The line preserved. Fifteen dealers never interrupted.* 08 · What you do Monday Three actions. One is immediate. Output · the page you keep What the analysis points to Here is where the numbers lead. Short enough to act on in a week, documented enough to stand up on a bank's next surveillance call. 1 Immediate · This week Pause new originations from Dealers C, H, and N Notify all three that new submissions are on hold pending performance review. Do not terminate the relationship, existing paper stays in the book under enhanced monitoring. The pause alone flattens your monthly NCO slope to +0.22 pp and removes the October breach risk. Document the decision and the data behind it for the bank's next surveillance call. 2 30 days · Underwriting policy Cap term at 60 months on vehicles aged 8 years or older 72-month terms on 8-10 year vehicles is the structural driver of the 2023-H2 deterioration. The collateral falls below the balance early and stays there through the default window. Capping term at 60 months on this vehicle profile cuts expected loss by roughly 35% on new originations in the segment. Apply it to all dealers, not just the three flagged. 3 Ongoing · Monthly Run the dealer scorecard monthly; watch Dealers F and M Dealers F and M sit above the network median (9.4% and 8.8% loss) but below the toxic threshold. Both bear watching the next two cycles. If either crosses 11% NCO or 20% EPD, apply the same pause protocol. A monthly refresh catches the slope before it ever reaches the aggregate, which is the whole point: you find it here, not on the bank's call. **Method & lineage.** Net loss = (charge-off − recovery) ÷ originated balance, computed per loan and aggregated by vintage and dealer. Vintages group by origination quarter. MOB = months from origination to as-of date (April 30, 2026). 60+ DPD counts loans in 60/90/repo buckets over active loans (paid-off and charged-off excluded from denominator). EPD = first default ≤3 months on book. Covenant slope derived from linear regression over the prior 6 months of monthly NCO readings. Dealer median excludes the top 3 outliers to avoid upward skew. The balance-versus-value chart in section 05 is illustrative of the structure, not a per-loan plot. This was prepared on a synthetic composite portfolio for illustration; the method is the same one a risk desk would apply to a live book. LendRisk Analytics Independent market research the research desk LendRisk Analytics · lendriskanalytics.com Worked Example · May 2026 Synthetic data · Illustrative only The point The aggregate hides the answer. A book like this reads compliant right up until the month it does not. The slope, the three dealers, and the breach date were all sitting inside an average that looked fine. Attribution is what turns a calm snapshot into a decision you can actually make. --- title: "When the term outlives the car" url: https://lendriskanalytics.com/term-matcher.html publisher: LendRisk Analytics kind: Page description: "A method note on matching loan term to vehicle useful life: straight-line depreciation versus amortization, the negative-equity window, and why long terms on aged cars carry a structural path to default." html: https://lendriskanalytics.com/term-matcher.html --- # When the term outlives the car [← All insights](https://lendriskanalytics.com/articles.html) Methods · Note 10 Method note · 6 min read Method note · Collateral & term # When the term outlives the car. LendRisk Analytics · Method note · 2026 This note publishes the full method behind a simple underwriting question: given a vehicle's age, mileage, and reliability class, what is the longest loan term that pays off *before* the vehicle reaches the end of its useful life? The market cares because the opposite structure, an aged car on a long term, is one of the most repeatable default patterns in subprime auto. Every model assumption is stated below. Every worked number is synthetic and labeled as such. ## Two curves, one race Every used-car loan is a race between two curves. The collateral curve falls: the method models wholesale value as straight-line depreciation from today's wholesale value down to a **$1,000 scrap floor**, spread over the vehicle's remaining useful life (floored at one year). Monthly depreciation is simply *(wholesale value − $1,000) ÷ (12 × years of useful life remaining)*. The balance curve also falls, but on a different schedule: a standard fully amortizing payment, where the outstanding balance after *m* months is the compounded principal minus the compounded value of payments made. The **underwater point** is the first month where projected balance exceeds projected wholesale value. At subprime rates in the low twenties, early payments are mostly interest, so the balance curve starts flat while the collateral curve falls at full speed. That asymmetry is the entire mechanism. And when the amount financed already sits above wholesale value at signing, because of rolled negative equity or add-ons in the note, the loan is underwater from month one and the race is over before it starts. ## How long a car actually lasts The method groups vehicles into six reliability tiers, each with a useful-life ceiling defined as **a mileage limit or an age limit, whichever comes first**, a reliability multiplier applied to repair hazard, and an average major-repair cost. The tier ordering and its ceilings are not invented and are not drawn from any proprietary or vehicle-history dataset: they are calibrated to **public, published data**. The [iSeeCars Longest-Lasting Cars Study (2025)](https://lendriskanalytics.com/data.html), built on roughly 400 million vehicles, ranks Toyota, Lexus, Honda, and Acura as the only brands whose models clear the 4.8% industry-average odds of reaching 250,000 miles, which is why they anchor Tier 1. The age ceilings track the **U.S. Bureau of Transportation Statistics** average-age-of-vehicles-in-operation series, which puts the typical light vehicle on the road well past a decade old, and the repair-cost column follows the brand ordering in **RepairPal** and **Consumer Reports** cost-of-ownership data, where European luxury marques run roughly two-to-four times mainstream annual repair cost. Sources are listed in full under [Data & sources](https://lendriskanalytics.com/data.html). | Tier | Representative makes | Useful life | Hazard multiplier | Avg major repair | |---|---|---|---|---| | 1 | Toyota, Honda, Mazda | 250k mi / 18 yr | 0.70× | $1,800 | | 2 | Hyundai, Kia, Subaru, Lexus, Nissan | 220k mi / 16 yr | 0.85× | $2,000 | | 3 | Ford, Chevrolet, Ram pickups | 220k mi / 16 yr | 0.95× | $2,200 | | 4 | Ford, Chevrolet sedans and SUVs | 190k mi / 14 yr | 1.00× | $2,100 | | 5 | BMW, Mercedes, Audi | 150k mi / 12 yr | 1.20× | $4,000 | | 6 | Land Rover, Jaguar, older Volvo | 130k mi / 10 yr | 1.35× | $4,500 | Reliability tiers as used in the model, calibrated to public data: iSeeCars Longest-Lasting Cars Study (2025, ~400M vehicles), U.S. Bureau of Transportation Statistics average-vehicle-age series, and RepairPal / Consumer Reports repair-cost-by-brand figures. Indicative tier averages, not vehicle-specific and not from any loan or vehicle-history dataset. Major-repair risk is modeled as a per-year hazard rate that steps up with the vehicle's age at the start of each loan year, then gets multiplied by the tier's hazard multiplier and capped at 85% per year: **5%** under age 5, **15%** at ages 5-7, **30%** at 8-10, **45%** at 11-13, and **60%** at 14 and older. Cumulative probability over the term is one minus the product of the yearly survival rates, and expected repair burden is that cumulative probability times the tier's average repair cost. **Inference**The hazard schedule is why term length dominates APR in this failure mode. A loan written across ages 10 through 15 stacks six high-hazard years inside one contract. No rate adjustment changes the mechanical odds that the collateral needs a repair the borrower cannot fund; it only changes how slowly the balance amortizes while those odds compound. ## The recommended-max-term rule The rule is one line. Compute the vehicle's remaining runway two ways: by age (useful-life years minus current age) and by mileage (remaining miles divided by expected annual mileage, defaulting to roughly **13,500 miles a year** per Federal Highway Administration figures unless the applicant's own usage says otherwise). Take the smaller of the two, convert to months, and subtract a **12-month buffer** so the loan retires while the borrower still has a functioning car and trade-out options. The result is clamped between 12 and 84 months. The same runway also splits the term into condition zones: the first 55% of remaining useful life is treated as healthy, 55-85% as aging, and everything beyond 85% as end-of-life. The verdict bands follow directly. **Matched:** the proposed term is at or under the recommended maximum, and the vehicle stays inside both its age and mileage ceilings through maturity. **Risky:** the term overshoots the recommended maximum by 12 months or less but the vehicle still finishes inside its useful life. **Mismatched:** anything longer, or any structure where projected age or mileage at maturity exceeds the ceiling, meaning the loan is scheduled to outlive the car. Where a deal lands risky or mismatched, the arithmetic offers three repairs: shorten the term to the recommended maximum, add roughly 15% of the amount financed as additional money down (with a $500 minimum) so a shorter term carries the same payment, or put the borrower in a vehicle with more runway. ## Worked example: a ten-year-old sedan on 72 months All numbers in this section are **synthetic and illustrative**, generated from the formulas above. Take a tier-4 domestic sedan: 10 years old, 120,000 miles, $6,000 wholesale value, driven 12,000 miles a year. Proposed structure: $7,500 financed over 72 months at 24% APR, payment about $197. The runway math: 4 years left by age, 5.8 by mileage, so 4 years governs; 48 months minus the 12-month buffer gives a **recommended maximum term of 36 months**. The proposed 72-month term doubles it, and the car crosses its 14-year ceiling at month 48 with two years of loan left. 48 mo Projected useful life remaining (synthetic example) 36 mo Recommended max term after 12-month buffer 98% Modeled probability of a major repair during the 72-month term M1 Underwater point: financed above wholesale at signing | Month | Projected balance | Projected wholesale value | Equity gap | |---|---|---|---| | M0 | $7,500 | $6,000 | −$1,500 | | M12 | $6,864 | $4,750 | −$2,114 | | M24 | $6,056 | $3,500 | −$2,556 | | M36 | $5,033 | $2,250 | −$2,783 | | M48 | $3,735 | $1,000 (scrap floor) | −$2,735 | | M60 | $2,088 | $1,000 | −$1,088 | | M72 | $0 | $1,000 | +$1,000 | Synthetic worked example, tier-4 model assumptions. The loan is underwater from signing, the gap peaks near $2,800 around month 36, and the vehicle hits the model's end of useful life at month 48 with $3,735 still owed. 72-month term vs. vehicle runway (synthetic example) Half the term sits inside the recommended maximum. A third of it runs past the vehicle's projected end of life. **Inside recommended max **Buffer consumed (M36-48) **Past end of useful life Synthetic example. Zone boundaries from the tier-4 useful-life model: 14 years or 190,000 miles, whichever comes first. At the recommended 36-month term the same $7,500 amortizes at about $294 a month, roughly $97 more than the 72-month payment. That is the honest trade on the table: the long term does not make the deal affordable, it moves the cost from the payment line into a 98% chance of a repair-versus-default decision on a car the borrower cannot exit. ## The negative-equity window What matters is not just *whether* the loan goes underwater but *how long it stays there*, because that window is when a breakdown converts into a default. The method reads the trade-out gap, balance minus value, in four bands. More than $2,000 underwater: the borrower is stuck; no trade-in covers the shortfall, and no rational replacement loan absorbs it. Up to $2,000 underwater: barely movable, and only by rolling the gap into a worse next loan. Thin equity under $2,000: one major repair wipes it out. Real equity beyond that: the borrower can trade, refinance, or absorb the repair. In the worked example the borrower sits in the stuck band from roughly month 6 through month 52, which is most of the contract. **Inference**Inside that window, the borrower's most rational economic move after a major failure is often to stop paying: the repair can cost more than the car is worth, and the note holds them above water on nothing. Read this way, a cluster of month-30-to-50 defaults on aged-vehicle, long-term paper is not a borrower-quality story. It is a deal-structure story that was visible at signing. The read The default is written into the structure the day the deal is signed. A ten-year-old car on a 72-month note is a race the collateral cannot win: **the loan must die before the car does, or the car's death becomes the loan's.** Sizing term to runway, with a 12-month buffer, is the whole method. ## Limits This method cannot see the individual unit. A meticulously maintained ten-year-old car and an abused one carry the same tier, age, and mileage inputs; the hazard schedule prices the class average, not the vehicle in front of you. Straight-line depreciation ignores used-market cycles, so in a year when wholesale values swing sharply, the value curve will be wrong in whichever direction the market moved. The tier ceilings compress wide model-level variance into six buckets, the $1,000 scrap floor is an assumption, and annual mileage is whatever the applicant reports. Most importantly, the hazard rates and useful-life thresholds are modeled calibrations to public reliability studies, not parameters fitted to loan-level outcome data. The method flags a structural mismatch between term and collateral; it does not predict any specific loan's fate, and it says nothing about income shocks, which end loans regardless of equity position. **Sources & notes** Useful-life tiers, age and mileage ceilings, and repair-hazard calibrations are modeled relationships anchored to **public, non-proprietary data**: the [iSeeCars Longest-Lasting Cars Study](https://www.iseecars.com/longest-lasting-cars-study) (2025, ~400 million vehicles analysed) for model longevity and the odds of reaching 250,000 miles; the [U.S. Bureau of Transportation Statistics](https://www.bts.gov/) average-age-of-vehicles-in-operation series for fleet age; the [Federal Highway Administration](https://www.fhwa.dot.gov/policyinformation/statistics/2023/vm1.cfm) Highway Statistics (annual vehicle-miles) for the ~13,500-mile default; and [RepairPal](https://repairpal.com/reliability) and [Consumer Reports](https://www.consumerreports.org/cars/car-maintenance/the-cost-of-car-ownership-a1854979198/) cost-of-ownership data for repair-cost ordering by brand. None of these are loan-level, borrower-level, or vehicle-history data. The tiers reflect the rankings those public studies produce; they are not fitted to any private dataset, so treat all outputs as indicative. All worked examples in this note, including the ten-year-old-sedan table and the term bar, are **synthetic and illustrative**, generated from the stated formulas, and describe no actual borrower, dealer, or loan. Full source list under [Data & sources](https://lendriskanalytics.com/data.html). LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned. Not investment, legal, or accounting advice. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). LR LendRisk Analytics Independent market research Continue reading [Methods · Note 01 Reading a *loan tape*: vintage curves, aging, and attribution](https://lendriskanalytics.com/analyze.html) [Methods · Note 02 The segments *carrying the loss*](https://lendriskanalytics.com/loss-drivers.html) --- title: "The composite read: three numbers against warehouse bands" url: https://lendriskanalytics.com/tool.html publisher: LendRisk Analytics kind: Page description: "The full arithmetic of a three-input composite read for auto lending books: annualised loss rate, 90-plus-day delinquency, and weighted-average FICO scored 0-100 against standard warehouse covenant bands, with every threshold and weight published as a reference table." html: https://lendriskanalytics.com/tool.html --- # The composite read: three numbers against warehouse bands [← All insights](https://lendriskanalytics.com/articles.html) Methods · Note 05 Method note · 6 min read Method note · Composite scoring # The composite read: three numbers against warehouse bands. LendRisk Analytics · Method note · 2026 A warehouse facility does not read a book the way its owner does. It reads a short list of covenant tests: a cap on annualised losses, a trigger on severe delinquency, a floor under borrower quality. This note publishes the full arithmetic of a 60-second composite read built on exactly those three numbers, scored 0 to 100 against the bands standard facilities write. This page previously hosted an interactive version of the calculation; the method is the useful part, so the method is now the page. 40 pts Weight on loss rate (NCO), the realised economic loss 30 pts Weight on severe lates (90+ DPD), the leading indicator 30 pts Weight on borrower quality (weighted-average FICO) 70 Composite level where the read turns to at-risk ## Three numbers, one score The composite takes the three numbers every auto lender already tracks and blends them into a single 0-100 read. **Loss rate** is annualised net charge-offs: for every $100 lent, the dollars that are permanently gone after the vehicle was repossessed, auctioned, and came up short. A 9% NCO means $91 recovered and $9 destroyed. Fitch's subprime auto index printed 9.81% annualised in January 2026, already inside the 8-10% cap range that standard warehouse facilities write. **Severe lates** is loans 90 or more days past due as a share of the active book, the leading indicator that more losses are coming. TransUnion's industry benchmark was 3.8% in Q4 2024; warehouse trigger levels commonly sit at 4-6%. **Borrower quality** is the weighted-average credit score across the active book; standard facility terms put the floor somewhere between 560 and 590. ## The arithmetic, in the open The score is one line of algebra. Each input is normalised to a 0-1 scale against a fixed denominator, multiplied by its weight, and summed: score = 40 × (NCO ÷ 12) + 30 × (DPD ÷ 10) + 30 × ((700 − FICO) ÷ 200) Each ratio is capped at 1, and the total is clamped to 0-100 and rounded. The weighting follows covenant priority order: NCO is the primary economic loss metric, DPD is the forward indicator, and FICO is the credit-quality floor. | Input | Weight | Normalisation | Safe | Watch | Breach-level | |---|---|---|---|---|---| | Loss rate (NCO, annualised) | 40 pts | NCO ÷ 12% | The full parameter set of the composite. Per-metric bands are calibrated to the Fitch NCO print (9.81%), standard warehouse caps (8-10% NCO), the TransUnion delinquency benchmark (3.8%), warehouse delinquency triggers (4-6%), and standard FICO floors (560-590). **Inference**The denominators are themselves judgments. Scaling NCO against 12% says that a book running at the January 2026 Fitch index level has already consumed roughly 82% of the loss dimension of the scale. Measuring FICO as distance below 700 across a 200-point band treats a move from 690 to 680 the same as a move from 570 to 560, which real credit risk does not. The normalisation is defensible as a covenant-distance measure; it is not a risk model. ## Three synthetic books, scored The mechanics are easiest to see worked through. All three books below are **synthetic and illustrative**; the stressed book uses the default preset values from the original interactive version. | Synthetic book | NCO | 90+ DPD | WA FICO | Component points | Composite | Tier | |---|---|---|---|---|---|---| | Clean book | 3.50% | 2.50% | 610 | 11.7 + 7.5 + 13.5 | 33 | Covenant safe | | Drifting book | 6.00% | 4.00% | 585 | 20.0 + 12.0 + 17.3 | 49 | Watchlist | | Stressed book | 9.30% | 6.90% | 578 | 31.0 + 20.7 + 18.3 | 70 | At risk | Synthetic worked examples. Note the drifting book: no single input is at breach level, yet it lands mid-watchlist. The composite is designed to catch books that look tolerable metric by metric but are deteriorating on all three axes at once. ## Reading the bands The composite maps to three tiers. Below 40, the book is comfortably inside the covenant bands. From 40 to 69, one or more metrics is drifting toward a threshold: watchlist territory, where direction matters more than level. At 70 and above, at least one dimension has typically crossed a line that facilities write into covenants, and the economics of the book are under direct pressure. Composite tier bands, 0-100 The three tiers of the published scoring arithmetic, with the 56 mark where the original diagnostic began assuming forward stress projections. Tier boundaries from the published arithmetic: safe below 40, watchlist 40-69, at-risk 70 and above. Two disclosures about the original interactive version, for completeness. First, it carried a soft internal flag at a composite of 56: above that level, the diagnostic assumed a warehouse bank was already running forward stress projections on the tape. Second, it displayed three benchmark-styled variants of the score, labeled Fitch, S&P Global, and TransUnion, computed as fixed multiples of the same composite (0.92×, 1.05×, and a delinquency-only 1.1× respectively). Those were presentational restylings of one number, not independent data feeds, and they should be read that way. **Inference**The 56 flag encodes a market observation rather than a covenant: banks model the slope, not just the point. A book that moves from 45 to 56 over three quarters is, in this framing, already the subject of internal forward projections even though no covenant has tripped. That is an interpretive claim about lender behaviour, consistent with how facility monitoring generally works, but it is not written in any facility document. ## What three numbers are enough for Three numbers are enough to answer one question well: *which side of the bands does this book sit on, and how much room is left?* That is genuinely useful. It is the difference between preparing for a covenant conversation this quarter and discovering one has been scheduled for you. A book owner who walks into that conversation with their own vintage curves and roll rates is in a different position than one who waits for the bank to arrive with theirs. The read A three-number composite is a **triage instrument**. It tells you where a book stands relative to the covenant bands in about a minute. It does not tell you **why** the book is there, **which direction** it is moving, or **what is driving** the drift. The honest use is sequencing: the score tells you whether pulling the vintage curves and roll-rate tables is a today problem or a this-quarter problem. ## Limits What this method cannot see, stated plainly. It has no vintage view: cohort-level deterioration surfaces six to nine months before it moves the portfolio aggregate, and a blended NCO hides it entirely. It has no roll rates: the 30-day bucket today predicts the 90+ bucket in two months, and the composite reads neither. It is denominator-blind: an annualised NCO on a fast-growing book understates true vintage losses, because new originations dilute the base before they have had time to default. Two structural limits matter most. First, **the bands here are calibration midpoints, not anyone's contract.** Actual covenant levels vary facility by facility; a real threshold is whatever the facility documents say it is. Second, **facilities test covenants one at a time, and a weighted blend can mask a single-metric breach.** A book at FICO 555 with pristine losses can score under 40 on this composite while sitting below a hard floor. Near any line, the per-metric status matters more than the blend. And the score itself is a weighted heuristic calibrated to public benchmarks, not a model fitted to loan-level data: an indicative read, not a measured probability. The anchors are date-stamped prints, Fitch January 2026, TransUnion Q4 2024, and they drift. **Sources & notes** Calibration anchors are those stated on the original calculator: the Fitch subprime auto ABS index (net charge-offs of 9.81% annualised, January 2026 print), the TransUnion industry 90+ day delinquency benchmark (3.8%, Q4 2024), the S&P Global rated-deal NCO series, and standard warehouse facility terms (8-10% NCO caps, 4-6% delinquency triggers, 560-590 FICO floors). The composite is a weighted heuristic calibrated to those public benchmarks, not a model fitted to a proprietary loan dataset. All worked examples in this note are synthetic and illustrative; they describe no actual lender. LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned. Not investment, legal, or accounting advice. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). LR LendRisk Analytics Independent market research Continue reading [Methods · Note 06 Recovery decay: *what a day of delay costs*](https://lendriskanalytics.com/repo-timing.html) [Methods · Note 07 The four levers of recovery, *state by state*](https://lendriskanalytics.com/repo-map.html) --- title: "Methods" url: https://lendriskanalytics.com/tools.html publisher: LendRisk Analytics kind: Page description: "Ten plain-English method notes on how a subprime auto book gets read, vintage curves, covenant runway, recovery, dealer channel, plus worked examples on synthetic books. Independent market research." html: https://lendriskanalytics.com/tools.html --- # Methods Methods · Independent research # How a subprime book gets read. These are method notes. Each one walks through an analysis a risk desk or a warehouse bank actually runs on a subprime auto book, in plain English, with the formulas, the thresholds, and a worked synthetic example on the page. Every number is either cited to a named public source or labeled synthetic. The write-ups are the whole point. Question 01 "Something is dragging the book. *Where is it coming from?*" [Note 01 · Loan tape Reading a loan tape: vintage curves, aging, and attribution How a loan-level tape turns into static-pool vintage curves, delinquency aging, and channel attribution. Those three views show deterioration months before the aggregate moves. Walked through on a synthetic tape. Read the note →](https://lendriskanalytics.com/analyze.html) [Note 02 · Segments The segments carrying the loss A book can read fine on the aggregate while two or three segment combinations quietly carry it down: deep-subprime on 84-month paper, or aged luxury collateral. How a cross-cut scan across nine risk dimensions finds them. Read the note →](https://lendriskanalytics.com/loss-drivers.html) [Note 03 · Dealer channel Reading the dealer channel How dealer-level loss rates, severe lates, and early payment defaults separate a soft channel from a bad one, and why the worst dealer in a book rarely looks bad in any single month. Read the note →](https://lendriskanalytics.com/dealers.html) Question 02 "The bank call is coming. *Where do I stand, and when does a trigger break?*" [Note 04 · Covenant runway Covenant runway: the month the trigger breaks Take current metrics and their monthly rate of change, project each warehouse trigger forward, and the runway falls out: the month a slope that reads "fine" turns into a cash sweep. Worked on a synthetic covenant package. Read the note →](https://lendriskanalytics.com/covenant.html) [Note 05 · Composite read The composite read: three numbers against warehouse bands Loss rate, severe delinquency, and weighted average credit score, scored against typical warehouse covenant bands. A sixty-second read on where a book stands before anyone opens the tape. Read the note →](https://lendriskanalytics.com/tool.html) Question 03 "A loan just went bad. *What comes back, and how fast do I have to move?*" [Note 06 · Recovery decay Recovery decay: what a day of delay costs The car barely depreciates day to day; the odds of getting it back fall off a cliff. A decay curve that prices the gap between "stage the file" and "go now," worked on a synthetic BHPH account. Read the note →](https://lendriskanalytics.com/repo-timing.html) [Note 07 · State law The four levers of recovery, state by state Self-help repossession, right to cure, deficiency judgments, and wage garnishment. Those four statutory levers decide what a defaulted loan actually returns, mapped across all 50 states. Read the note →](https://lendriskanalytics.com/repo-map.html) Question 04 "*Who gets in the door?*" [Note 09 · Deal economics The economics of a single deal One contract taken apart: acquisition cost, expected loss, recovery, servicing drag, and discount. The arithmetic that decides whether a deal ever had a margin to begin with. Worked on a synthetic contract. Read the note →](https://lendriskanalytics.com/underwriter.html) [Note 10 · Term vs. collateral When the term outlives the car Long terms on old collateral push the payoff date past the collateral's useful life. Where the crossover sits, why negative equity concentrates there, and how to read term against vehicle age. Read the note →](https://lendriskanalytics.com/term-matcher.html) Worked examples · Synthetic books Two blinded synthetic books, each walked through the methods above end to end. Every figure on those pages is synthetic and labeled as such. [Worked example · Indirect How to read a subprime book, indirect One blinded synthetic book of dealer-originated paper, walked through the methods end to end: vintage curves, dealer attribution, covenant projection, and the read that falls out. Read the walkthrough →](https://lendriskanalytics.com/sample-report.html) [Worked example · Direct How to read a subprime book, direct One blinded synthetic book of direct-to-borrower paper, walked through the methods end to end: vintage curves, borrower segments, covenant projection, and the read that falls out. Read the walkthrough →](https://lendriskanalytics.com/sample-report-direct.html) Market briefs The borrowing-base certificate as the single point of failure in BHPH warehouse funding, one problem, read from both seats. [Market brief · Operator's seat The certificate problem, the operator's seat The monthly borrowing-base certificate is self-reported, and the Tricolor record shows what happens when the market cannot tell a clean one from a fabricated one. What that problem looks like from the seat that files it, read against the Fed note and the public record. Read the brief →](https://lendriskanalytics.com/borrowing-base.html) [Market brief · Lender's seat The certificate problem, the lender's seat The same certificate from the other side of the table: what a warehouse lender can and cannot see in a self-reported borrowing base, and how the sector re-rated after Tricolor. Read against the Fed note and the public record. Read the brief →](https://lendriskanalytics.com/borrowing-base-lender.html) --- title: "The economics of a single deal" url: https://lendriskanalytics.com/underwriter.html publisher: LendRisk Analytics kind: Page description: "A method note on single-deal auto-loan economics: how probability of default, loss given default, expected loss, and lifetime ROA combine into a fund, counter, or decline read, with benchmark assumptions and a worked synthetic example." html: https://lendriskanalytics.com/underwriter.html --- # The economics of a single deal [← All insights](https://lendriskanalytics.com/articles.html) Methods · Note 09 Method note · 5 min read Methods · Deal economics # The economics of a single deal. LendRisk Analytics · Method note · 2026 Every funded auto loan is a small standalone business: it earns interest on a shrinking balance, pays for its funding and servicing, and carries a probability-weighted loss. This note lays out the four-line arithmetic that turns a borrower, a vehicle, and a set of terms into a *fund, counter, or decline* read, and shows, on a synthetic example, when a counter-offer rescues a marginal deal and when nothing does. 12% 12-month base default rate, 600-649 score band +25% PD multiplier per 10 LTV points above 100 ≈0.55× Average balance on an amortizing loan, vs amount financed 50% / <0 Decline lines: lifetime PD, or expected ROA ## The question the method answers Large lenders answer "should this loan be written?" with a scored model and a credit team. Small independent lenders and buy-here-pay-here operators answer it at a desk, in minutes, usually on instinct. The method here is the middle path: an explicit, checkable chain from deal inputs to an economic verdict. It does not predict any individual borrower's behavior. It asks a narrower question: **given typical loss behavior for this profile, does the deal's expected lifetime economics clear the lender's own hurdle?** ## Four lines of arithmetic **Line 1, probability of default (PD).** Start with a 12-month base rate by credit-score band: roughly 1.5% at 800+, 4% at 700-749, 12% at 600-649, 20% at 550-599, 30% at 500-549, and 40% below 500. Then multiply by deal-shape adjustments: each 10 points of loan-to-value above 100 adds 25% to PD; debt-to-income above 45 adds 30-55%; each year of vehicle age past five adds 4%; and longer terms raise the per-period rate (about 0.85× at 36 months, rising ~0.12 per additional year). Finally, convert 12-month PD to lifetime PD with a term multiplier, about 1.8× at 36 months, 2.5× at 60, 2.85× at 84, capped at 85%. **Line 2, loss given default (LGD).** Severity is driven by the collateral path, not the borrower. A base LGD by vehicle age (35% for near-new, 55% at four to six years, 75% past ten) is scaled by LTV at the time of default, approximated as origination LTV less about 15% of balance paydown, and bounded between 20% and 95%. **Line 3, expected loss (EL).** PD × LGD × amount financed. This is the reserve-sized dollar figure the deal must earn back before it earns anything. **Line 4, lifetime ROA.** Gross interest is the APR applied to the average amortizing balance (≈0.55× the amount financed) over the term. Subtract cost of funds on the same average balance, per-loan operating cost, and expected loss. Divide the remainder by the amount financed. The verdict follows mechanically: **decline** if lifetime PD is 50% or higher or expected ROA is negative; **fund** if ROA clears the lender's target; **counter** in between. **Inference**Because the PD adjustments are multiplicative, risk factors compound rather than add. A 600-score borrower at 115 LTV on a 72-month note is not "three notches worse" than the base case, the multipliers stack to several times the base rate. In this framework, most decline verdicts are reached through the LTV multiplier, not the score band alone. ## Two lenders, two benchmark sets The same arithmetic prices very different books depending on the lender's own economics. Two benchmark profiles anchor the method, an independent dealer or finance company writing mid-subprime paper, and a buy-here-pay-here operator self-financing older vehicles at deeper subprime scores. | Benchmark assumption | Independent / finco | BHPH operator | |---|---|---| | Cost of capital | 7.5% | 9.5% | | Operating cost per loan | $200 | $350 | | Target lifetime ROA | 1.5% | 3.0% | | Typical credit score | 615 | 540 | | Typical APR | 18% | 22% | | Amount financed | $15,500 | $10,500 | | Vehicle wholesale value | $14,500 | $9,000 | | Down payment | $1,500 | $1,500 | | Term | 60 mo | 36 mo | | Vehicle age | 5 yr | 9 yr | Benchmark profiles used by the method. The BHPH set trades a shorter term and smaller advance against deeper scores, older collateral, and a higher hurdle, because the operator's own funding and servicing cost more per loan. ## A worked deal, walked to a verdict A synthetic deal, deliberately marginal: a 615-score borrower at 38% DTI, a five-year-old vehicle worth $13,500 wholesale, $15,500 financed at 18% APR for 60 months, against the independent-lender economics above. | Step | Value | Read | |---|---|---| | LTV at funding | 114.8% | Negative equity from day one | | Lifetime PD | 49.3% | Just under the 50% decline line | | LGD | ~54% | Mid-age collateral, near-100 LTV at default | | Expected loss | $4,101 | 26% of the amount financed | | Gross lifetime interest | $7,673 | 18% on the average balance, 5 years | | Cost of funds | −$3,197 | 7.5% on the same average balance | | Operating cost | −$200 | Per-loan servicing assumption | | Net profit / lifetime ROA | $175 · 1.1% | Positive, but below the 1.5% target → COUNTER | Synthetic example computed with the arithmetic above. Not a real borrower, vehicle, or lender. Where the interest dollar goes · worked deal Of $7,673 in expected lifetime interest, expected credit loss and funding cost consume almost all of it. The lender's margin is the sliver on the right. **Cost of funds **Operating cost **Expected loss **Net profit Synthetic worked example. Values computed from the stated model assumptions, not observed data. The read Single-deal underwriting is a race between **interest margin and expected loss**, and on marginal subprime paper the two run nearly even. The verdict is rarely about whether the deal makes money in the good outcome, it is about whether the margin survives the probability-weighted bad one. **A deal that "pencils" at zero losses is not a priced deal.** ## When a counter rescues the deal, and when it can't A counter verdict means the deal is under-priced for its risk, not unprofitable. The method searches for the single smallest change that clears the target, in order: a rate bump (half-point steps, capped at 29% APR), additional down payment ($250 steps), a shorter term (six-month steps, floor of 24), or a smaller advance. On the worked deal, the rescues are strikingly cheap. A **half-point APR increase to 18.5%** adds about $213 of lifetime interest on the average balance and lifts ROA from 1.1% to roughly 2.5%, clearing the target. So does **$250 of additional down payment**, which trims LTV, PD, and expected loss simultaneously. The term lever, by contrast, fails on this deal at every step. Shortening from 60 to 48 or 36 months does cut lifetime PD meaningfully, but it cuts lifetime interest faster, and ROA never reaches the target. In this arithmetic, term reduction de-risks the deal and de-profits it at the same time; price and advance are the levers that actually move the verdict. **Inference**The model treats a rate bump as pure margin: APR appears in the revenue line but not in the PD line. Real borrowers are not so obliging, a higher payment raises default risk, especially at stretched DTI. Treat rate-based counters as an upper bound on the rescue, and prefer the down-payment lever when the borrower can reach it: it is the only counter that improves both sides of the ledger. ## Limits This is a decision-support heuristic, not a credit model. Its parameters are stylized calibrations to the shape of published industry default curves, not fitted coefficients, and a lender's own portfolio history should replace them wherever it exists. The score-band base rates step in cliffs, 599 versus 600 moves the base rate from 20% to 12%, where reality is smooth. The average-balance approximation (0.55×) and the fixed default-timing assumption (~15% paydown at default) are conveniences, not observations. **Inference**The multiplicative stacking is calibrated to the independent-lender case and turns harsh at the deep-subprime end: the BHPH benchmark profile itself, 540 score, 117 LTV, nine-year-old collateral, stacks to a lifetime PD above the 50% decline line. BHPH economics survive in practice on levers this framework does not price: relationship collection, repossess-and-re-sell cycles, and down payments that exceed vehicle cost basis. For that segment, read the output as a stress reading, not a verdict. Finally, lifetime ROA here is net profit over the amount financed across the whole term, not an annualized figure, comparable across deals of similar term, less so across a 36-month and a 72-month note. None of this changes the core discipline the arithmetic enforces: no deal is priced until expected loss has been subtracted. **Sources & notes** All model parameters in this note, score-band base rates, the LTV, DTI, term, and vehicle-age multipliers, LGD assumptions, the average-balance approximation, benchmark profiles, and decision thresholds, are the method's own stylized assumptions, calibrated to the general shape of published industry auto default curves rather than to any single dataset. All deals, borrowers, and lender profiles shown are **synthetic**. LendRisk Analytics is an independent research publication with no position in, and no affiliation with, any company mentioned. Not investment, legal, or accounting advice. Have a question about the market, or a different view? [Send it through →](https://lendriskanalytics.com/contact.html). LR LendRisk Analytics Independent market research Continue reading [Methods · Note 10 When the *term outlives* the car](https://lendriskanalytics.com/term-matcher.html)