--- title: "What the tape said: set at signing" url: https://lendriskanalytics.com/insights/set-at-signing.html publisher: LendRisk Analytics published: 2026-10-02 kind: Article description: "Used cars are worth about a third more than before the pandemic. Subprime lenders recover about eleven points less on repossessions than they did then. The vintage explains it: recovery is fixed by the price paid when the loan was written, not by the auction market when the car is sold. Every figure traced to a filing or a Fitch document." html: https://lendriskanalytics.com/insights/set-at-signing.html --- # What the tape said: set at signing [← All insights](https://lendriskanalytics.com/articles.html) What the Tape Said · Issue 15 Research brief · 13 min read Subprime auto · Recoveries · Vintage # Set at signing. Public record through October 2, 2026 · LendRisk Analytics A used car is worth about a third more today than it was at the end of 2019. A subprime lender that repossesses one gets back about eleven points less of the balance than it did then. Both of those are true at the same time, and the usual desk shortcut, that strong auction prices protect loss severity, cannot hold them together. The number that does is the year the loan was written. Recovery on a repossessed car is fixed by what the borrower paid for it, not by what the auction pays for it later, and the loans written at the top of the market are the ones coming back now. +35% Manheim index, September 2026 against December 2019 32.6% Fitch subprime recovery index, end of 2025, against a 43.7% pre-pandemic average 22% vs 47% CPS recovery on 2022-vintage repossessions against 2025-vintage, same quarter 257.7 Manheim index at the end of 2021, when the 2022 vintage was being priced ## Bottom line Three numbers from three sources, none of them ours. First, the auction market. Cox Automotive's Manheim Used Vehicle Value Index stood at 206.2 in mid-September 2026. Cox's own December 2023 release put the index at 204.0 and described it as "about 33% higher than at the end of 2019," which places the pre-pandemic year-end near 153. Used cars are still expensive by any pre-2020 standard. Second, the recovery index. Fitch's subprime auto ABS recovery index ended 2025 at 32.64%, up from 31.84% a year earlier and, in Fitch's words, "well below the pre-pandemic average of 43.73%." Eleven points of balance that used to come back at auction and now does not. Third, the vintage. On its second-quarter 2026 call, Consumer Portfolio Services' chief operating officer, Mike Lavin, gave recovery rates by the year the loan was written: 22% for 2022, 25% for 2023, 37.5% for 2024, 47.1% for 2025. Same lender, same quarter, same auctions. The difference is more than two to one, and it runs exactly opposite to the auction market, which was at its all-time high when the 2022 paper was written and has fallen about a fifth since. The read Recovery is a ratio. The numerator is what the car sells for, and the auction market sets that. The denominator is the balance still owed, and that was set at signing by the price paid, the amount financed and the term. A loan written against a car priced at the 2021 peak carries a denominator the 2026 auction cannot reach, however firm the bid. **The improvement CPS is now reporting is the 2022 and 2023 paper leaving the book, not the market getting better,** and that distinction decides what happens to severities if wholesale prices keep softening from here. ## I · The paradox, drawn Chart 1 · Wholesale prices stayed high. Subprime recoveries did not follow. Manheim Used Vehicle Value Index above, Fitch subprime auto ABS recovery index below, December 2019 to September 2026. Only published readings are plotted; the lines between them are connections, not data. Manheim: Cox Automotive releases for each date; the December 2019 value is derived from Cox's statement that December 2023 (204.0) stood "about 33% higher than at the end of 2019," and is marked hollow for that reason. Fitch recovery: year-end 2024 and 2025 and March 2026 from the Fitch and Black Book Vehicle Depreciation Report 2026; June and July 2026 from Fitch's August 19 commentary as reported by Auto Remarketing. Fitch revised the index's inclusion criteria from the July 2026 reading, so the hollow July point is on a broadened basis. Linear scales, 7.16 pixels per month. The top panel is the market everyone quotes. The bottom panel is what a subprime lender gets. Between the pandemic peak and now, the top line fell about a fifth and settled a third above its old floor. The bottom line never got back within ten points of its old average, and the spring 2026 recovery to 37.48%, then 39.5% in June, is the ordinary tax-season bounce that Fitch's own report expects to be "short lived." Fitch names part of the mechanism itself. In the same report: "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." The Manheim index is mix-adjusted to a market the subprime repo lot does not resemble. That explains why the two lines decoupled. It does not explain why they decoupled by vintage, which is the part that matters for anyone holding a book. ## II · Same lender, same quarter, same auctions Chart 2 · CPS recovery rate by origination year, second quarter 2026 Recovery on repossessed vehicles sold in the quarter, grouped by the year the contract was written, as stated by CPS management on the Q2 2026 call. The Manheim level at each vintage's pricing point is annotated beneath. Vintage recovery rates as spoken by COO Mike Lavin on the Q2 2026 earnings call (August 2026); they do not appear in the 10-Q. CPS defines its recovery rate in its releases as "wholesale auction liquidation amounts (net of expenses) as a percentage of the account balance at the time of sale." Manheim annotations are Cox year-end readings, with the all-time high of 257.7 dated by Cox to the end of 2021. Read the bottom row against the bars. The vintage priced at the peak recovers least. The vintages priced after the index gave back a fifth recover most, and the 2025 paper, written against essentially the same auction market the 2024 paper was, recovers ten points better still, which is where underwriting enters: Fitch records lenders tightening from late 2022 "through higher minimum FICO thresholds, lower loan to value caps, and enhanced income verification." Lower LTV at signing is a smaller denominator. The chart is price paid and advance rate, in that order. **Inference** The CPS numbers are one lender's, spoken rather than filed, and they are the only public vintage split of subprime recoveries this brief could find. They are used here because the direction is large, consistent with Fitch's description of 2022 and 2023 as the weakest cohorts, and consistent with simple arithmetic. They are not used to put a number on the industry. The fact that the cited sources describe the mechanism rather than demonstrate it is stated plainly in the Limits. ## III · The improvement is burnout, not market CPS is the best-disclosed subprime issuer on this question because it publishes both halves of the story every quarter: the recovery rate and the repossession inventory as a share of the portfolio. Chart 3 · CPS: recoveries up, repo inventory down, eight quarters Quarterly recovery rate above, repossession inventory as a share of the total portfolio below, Q3 2024 to Q2 2026, all from CPS earnings releases filed as 8-K exhibits. CPS earnings releases, 8-K Exhibit 99.1, for each quarter. Full-year 2025 recovery rate 28.8% against 30.1% in 2024. The repo bucket in the 10-Q delinquency table fell from $111,104 thousand at December 31, 2025 to $94,310 thousand at June 30, 2026 while the contractual portfolio grew from $3,778,127 thousand to $4,306,511 thousand. From the fourth quarter of 2025 to the second quarter of 2026, CPS's recovery rate rose from 28.5% to 33.3% and its repossession inventory fell from 2.94% of the portfolio to 2.19%, from $111.1 million to $94.3 million in dollars while the book grew half a billion. Over the same two quarters the Manheim index went from 206.0 in mid-December to a tax-season peak of 215.3 in March and back to 208.2 by August, a round trip to roughly where it started. Lavin's explanation on the call was the vintage, with the older cohorts set to "flush out" of the portfolio and recovery rates expected to trend higher toward year-end as they do. He also said repossessions "were down over the first quarter and the second quarter" and that "extensions as a percentage of the portfolio were slightly up quarter over quarter." Both of those belong in the record. Fewer repossessions and more extensions in the same two quarters is consistent with cars being kept on the road rather than taken back, and [Issue 8](https://lendriskanalytics.com/insights/current-on-paper.html) covered what that does to a delinquency line. It is also consistent with a tax season that cured accounts. The inventory decline has more than one parent. **Inference** If the aggregate improvement is vintage runoff, two things follow that an operator can act on. CPS's recovery rate should keep rising through the second half of 2026 even if Manheim keeps softening, for as long as 2022 and 2023 paper keeps leaving the book. And any lender with a heavier 2022 and 2023 concentration should be showing worse severities than peers at the same auction prices, which is a comparison worth running on your own servicing data before a lender or a trust tells you the market did it. ## IV · Three stories the data does not support 1 **"Repossessions are surging, recoveries are falling."** The best-disclosed subprime issuer shows the reverse through mid-2026: inventory and charge-offs down, recoveries up. Fitch's recovery index rose from year-end into June. The widely circulated loan-level repo series carries a structural break: Bill Ploog's September 2025 analysis of ABS-EE filings, which put the July 2025 below-prime first-time repossession rate at 0.74%, "68% (30 bp) above July-2024," discloses in its own footnote that Exeter "ceased reporting nearly all repossessions on form ABS-EE exhibit 102" in November 2022 and that the chart "excludes Exeter's reporting of nearly 86K repossessions in Mar-2025, but it does include their repossessions from May forward." A July 2025 reading that includes Exeter against a July 2024 reading that largely does not is not a clean year-over-year comparison. The honest version of that series excludes Exeter's trusts or starts in May 2025, and this brief has not yet re-run it. 2 **"Tariffs lifted used-car prices and therefore recoveries."** Fitch says the opposite for subprime, in the sentence quoted above. Manheim's 2026 peak of 215.3 landed in March, tax-refund season, and by mid-September the index was 0.4% below a year earlier, "its first year-over-year decline of 2026" in Cox's words. Nothing in the public record isolates a tariff effect on subprime recoveries. 3 **"The time from delinquency to repossession is lengthening."** Nobody can measure this from public data after 2022. The CFPB's only repossession dataset, published January 2025, covers nine lenders from 2018 through December 2022: 0.75% of outstanding loans assigned to repossession in December 2022 against 0.61% in December 2019, forwarder use up from 31% to 66%, average deficiency above $11,000. It stops there. The ABS-EE loan-level schema has no repossession date field, so the lag can only be approximated from delinquency status transitions, which is a different measurement. ## V · What is on the record around it Context, not evidence, and labeled as such. Fitch's 60+ day subprime delinquency index hit a record 6.90% at the January 2026 reading, eased to 6.11% in March on tax refunds and printed 6.13% in July, with annualized net losses at 8.42% at mid-year. Prime recoveries in the same index were 64.0% in June and 59.0% in July, twenty points above subprime at the same auctions, which is the collateral-mix point in one line. The New York Fed's second-quarter household report put auto balances at $1.71 trillion and the annualized flow into 90-day delinquency at 3.00%, against 2.93% a year earlier. GM Financial, a prime and near-prime captive, recovered $281 million against $535 million of charge-offs in the second quarter, 52.5%, down from 55.3% a year earlier, and told investors in its third-quarter 2025 deck that it expects "used vehicle prices to trend lower over time, driving higher off-lease return rates and lower recovery rates on repossessed vehicles." Credit Acceptance's forecast collections slipped $34.2 million in the fourth quarter of 2025 and $39.1 million in the second quarter of 2026, both 0.3%. America's Car-Mart's net charge-offs ran 9.5% in its quarter to July 2026 against 6.6% a year earlier, with retail units down 81.9% as it ran short of capital to buy inventory; [Issue 9](https://lendriskanalytics.com/insights/the-cash-problem.html) covers how it got there. The supervisory backdrop moved in the other direction from the data. The CFPB withdrew 67 guidance documents on May 12, 2025, among them Bulletin 2022-04, "Mitigating Harm from Repossession of Automobiles." On February 4, 2026, Senator Warren sent letters to twelve lenders, dealers and trade associations asking for repossession error-rate data, on the stated ground that no public review of it exists. That is correct. There is no public measure of wrongful repossession, and there is no public measure of repossession timing after 2022. What is public, and underused, is the recovery side, which is what this brief is about. ## VI · What this does not show 1 **The mechanism is inferred, not demonstrated.** Fitch, Cox and CPS each supply a verified number. None of them shows that loan balance relative to vehicle value at signing is the cause. The direct test is loan-level: repossession proceeds over balance at default, grouped by origination year, alongside original loan amount over vehicle value, in a public ABS-EE shelf such as Santander's. That test has not been run here. When it is, it belongs in this series. 2 **The vintage split is spoken, single-issuer, and single-quarter.** It comes from a transcript, not a filing, and one transcript renders the first bucket as "2020 to 2022 vintage." The figures are quoted as the COO gave them and attributed as such. 3 **Fitch's recovery figures exist in more than one version.** The Fitch and Black Book report gives 32.64% for end-2025 and 31.84% for end-2024. Auto Remarketing's March 2026 reporting of Fitch's December index gives 32.86% and 35.25%. The sets differ by index date and, from July 2026, by a revision that brought previously excluded shelves into the index. This brief cites the Fitch-authored document and notes the other. The gap to the pre-pandemic average is about eleven points on either set. 4 **Recovery is defined differently by every source.** Fitch measures recoveries against gross losses. CPS measures auction proceeds net of expenses against balance at sale. GM Financial's recoveries include all post-charge-off collections. Levels are not comparable across the three. Direction is, and only direction is used. 5 **The Manheim pre-pandemic level is derived.** Cox rebased the index to January 1997 = 100 with its January 2023 release and restated history, so older press-release levels do not line up with current ones. The figure near 153 for December 2019 is backed out of Cox's own December 2023 statement rather than read from a table, and is marked as such wherever it appears. Falsifiable The reading fails if CPS's recovery rate falls back below 30% in the third or fourth quarter of 2026 while the Manheim index holds near 206, because that would mean the market, not the vintage, was driving it. It also fails if Fitch's subprime recovery index closes within three points of its 43.73% pre-pandemic average by the January 2027 reading without the 2022 and 2023 cohorts having amortized out of the index. **Both are checkable from the next two CPS releases and the next Fitch annual report.** ## Proof: every figure, traced Verified means the figure was read at the named primary document and matches. Derived means calculated here from verified figures, with the arithmetic shown. Spoken means taken from an earnings-call transcript rather than a filing, and attributed to the speaker. | Claim as stated | Source | Status | |---|---|---| | Fitch subprime recovery index ended 2025 at 32.64%, from 31.84% at end-2024, "well below the pre-pandemic average of 43.73%"; TTM 37.02% from 37.92% | Fitch Ratings and Black Book, Vehicle Depreciation Report 2026 , text extracted from the PDF | Verified | | March 2026: delinquencies 6.11%, annualized net losses 8.80%, recoveries 37.48%, improvement "expected to be short lived" | Same report | Verified | | 60+ day delinquency record 6.90% at the January 2026 index date, from 6.45%; annualized net losses 9.81% | Same report | Verified | | "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" | Same report, verbatim | Verified | | Lenders tightened from late 2022 "through higher minimum FICO thresholds, lower loan to value caps, and enhanced income verification"; 2022 and 2023 vintages the weakest | Same report, verbatim | Verified | | Fitch mid-2026: recoveries peaked in April, 39.5% June, 38.0% July; July 60+ delinquency 6.13%; annualized net losses 8.42%; prime recoveries 64.0% June, 59.0% July; inclusion criteria revised ("previously excluded shelves were incorporated") | Auto Remarketing, August 24, 2026 , reporting Fitch commentary of August 19, 2026 | Verified | | Alternative Fitch reading: December 2025 recovery 32.86% against 35.25% a year earlier | Auto Remarketing, March 12, 2026 ; disclosed in Limits | Verified | | Eleven-point gap: 43.73 less 32.64 = 11.09 | Arithmetic on the Fitch figures | Derived | | Manheim all-time high 257.7 at the end of 2021; December 2024 204.8 | Cox Automotive, December 2024 release | Verified | | Manheim December 2023 at 204.0, "about 33% higher than at the end of 2019" | Cox Automotive, December 2023 release | Verified | | Manheim December 2019 near 153 | 204.0 divided by 1.33 = 153.4, from the Cox statement above | Derived | | Manheim mid-December 2025 at 206.0 (+0.6%); March 2026 at 215.3 (+6.2%, +1.4% month on month); June 212.9; August 208.2 (+0.4%); mid-September 206.2 (down 0.4%, "its first year-over-year decline of 2026"); Jonathan Gregory quotation | Cox Automotive releases: mid-December 2025 , March 2026 , Q2 2026 , August 2026 , mid-September 2026 | Verified | | Manheim rebased to January 1997 = 100 as of the January 2023 release, history restated | Cox Automotive methodology notes | Verified | | +35% September 2026 against December 2019: 206.2 over 153.4; down about a fifth from the peak: 206.2 over 257.7 = 0.80 | Arithmetic on the Cox figures | Derived | | CPS recovery by vintage, Q2 2026: 22% (2022), 25% (2023), 37.5% (2024), 47.1% (2025); older vintages to "flush out"; repossessions "down over the first quarter and the second quarter"; extensions "slightly up quarter over quarter" | CPS Q2 2026 earnings call transcript , Mike Lavin, Chief Operating Officer | Spoken | | CPS recovery rate Q2 2026 33.3% against 30.4%; repo inventory 2.19% against 2.64%; net charge-offs 7.28% against 7.45%; 31+ delinquency including repo 12.16% against 13.14%; portfolio $4,306.66M; recovery rate defined as "wholesale auction liquidation amounts (net of expenses) as a percentage of the account balance at the time of sale" | CPS 8-K Exhibit 99.1, Q2 2026 | Verified | | CPS Q1 2026: recovery 31.1% against 27.7%; repo inventory 2.63% against 2.60%; net charge-offs 8.57% against 7.54% | CPS Q1 2026 earnings release | Verified | | CPS Q4 2025: recovery 28.5% against 27.2%; repo inventory 2.94% against 2.74%; full-year recovery 28.8% against 30.1%; full-year net charge-offs 7.76% against 7.62% | CPS 8-K Exhibit 99.1, Q4 2025 | Verified | | CPS Q3 2025: recovery 28.7% against 29.1%; repo inventory 2.84% against 2.79%; net charge-offs 8.01% against 7.32% | CPS Q3 2025 earnings release | Verified | | CPS repo bucket $94,310 thousand at June 30, 2026 against $111,104 thousand at December 31, 2025; contractual portfolio $4,306,511 thousand against $3,778,127 thousand | CPS Form 10-Q, Q2 2026 , finance receivables note | Verified | | GM Financial Q2 2026 charge-offs $535M, recoveries $281M, net $253M; Q2 2025 $488M, $270M, $217M; six months $1,076M / $551M and $967M / $520M; annualized net charge-offs 1.3% against 1.1% | GM Financial 8-K, Q2 2026 | Verified | | GM Financial recovery ratios 52.5% and 55.3%; 51.2% and 53.8% | 281/535, 270/488, 551/1,076, 520/967 | Derived | | GM Financial: "Expect used vehicle prices to trend lower over time, driving higher off-lease return rates and lower recovery rates on repossessed vehicles" | GM Financial 3Q25 earnings presentation , slide 13, text extracted from the PDF | Verified | | Ploog, September 20, 2025: July 2025 below-prime first-time repo rate 0.74%, "68% (30 bp) above July-2024"; Bridgecrest 1.03%, Exeter 0.93%, Santander 0.79%; 7.6 million loans, $138 billion; Exeter footnote on the November 2022 reporting stop and the March 2025 86K | CUCollector, September 20, 2025 | Verified | | CFPB: 0.75% of outstanding loans assigned to repossession in December 2022 against 0.61% in December 2019; forwarder use 31% to 66%; average deficiency above $10,000 then above $11,000; nine lenders, 2018 to 2022 | CFPB, Repossession in Auto Finance, January 2025 | Verified | | New York Fed Q2 2026: auto balances $1.71 trillion, up $28 billion; annualized flow into 90+ day delinquency 3.00% against 2.93% | New York Fed, August 11, 2026 | Verified | | Credit Acceptance: forecasted net cash flows down $34.2M (0.3%) in Q4 2025 from $58.6M (0.5%) in Q3; down $39.1M (0.3%) in Q2 2026 against $55.8M (0.5%) a year earlier | Credit Acceptance Q4 2025 release ; Q2 2026 release | Verified | | America's Car-Mart, quarter to July 31, 2026: net charge-offs 9.5% against 6.6%; retail units down 81.9%; revenue down 57.3%; loss $69 million | Car-Mart 8-K Exhibit 99.1, September 9, 2026 | Verified | | CFPB withdrew 67 guidance documents on May 12, 2025, including Bulletin 2022-04, "Mitigating Harm from Repossession of Automobiles" | CFPB withdrawn guidance page ; Federal Register notice | Verified | | Senator Warren letters of February 4, 2026 to twelve recipients requesting repossession error-rate data | Senate Banking Committee, minority release | Verified | Removed from the source memo before publication because they could not be placed at a primary document, or because the memo itself flagged them as unsafe: a count of 1.73 million repossessions in 2024 (an industry estimate, not a measured series); KBRA's July 2026 index figures (secondary only); a going-concern characterization of Car-Mart; any claim about repossession timing after 2022; any tariff attribution; and the "2025 vintage" attribution for Credit Acceptance's Q2 decline, which the release does not state. **Sources & notes** **Indices.** Fitch Ratings and Black Book, Vehicle Depreciation Report 2026 (May 2026), for the recovery, delinquency and loss index readings through March 2026 and the collateral-mix statement. Fitch mid-year commentary (August 19, 2026) as reported by Auto Remarketing (August 24, 2026) for June and July readings and the inclusion revision. Auto Remarketing (March 12, 2026) for the alternative December reading. Cox Automotive Manheim Used Vehicle Value Index releases for every Manheim figure. **Issuers.** Consumer Portfolio Services earnings releases (8-K Exhibit 99.1) for Q3 2024 through Q2 2026, the Q2 2026 Form 10-Q, and the Q2 2026 earnings call transcript. GM Financial Q2 2026 earnings release (8-K) and 3Q25 earnings presentation. Credit Acceptance Q4 2025 and Q2 2026 earnings releases. America's Car-Mart Q1 FY2027 release (September 9, 2026). **Other.** CFPB, Repossession in Auto Finance (January 2025). CFPB withdrawn-guidance notice (May 12, 2025). Federal Reserve Bank of New York, Household Debt and Credit Report, Q2 2026. Senate Banking Committee minority release (February 2026). Bill Ploog, CUCollector (September 20, 2025), for the ABS-EE repo series and its Exeter footnote. **Companion issues.** Issue 8 covers extension mechanics and what they do to a delinquency line. Issue 9 covers the 2023 to 2026 failure roster, including Car-Mart and Tricolor, as cash events. Where this brief reasons beyond what a document states, it is labeled as an inference. Point-in-time reading of the public record through October 2, 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)