Auto ABS Loss to Liquidation: 2022-23 Owns the Peak
Of every dollar of principal that has liquidated out of the subprime auto cohort originated in Q3 2023, 19.1 cents left as net loss. That is the highest loss-to-liquidation reading for any quarterly cohort in the securitized data going back to 2017, and it belongs to a vintage that is only 30 months old.
Loss to liquidation is my favorite severity lens because it ignores the noise that distorts annualized loss rates. It asks one question: of the principal that has actually exited the pool, through payoff, prepayment, or charge-off, what share exited as loss? We compute it from loan-level ABS-EE filings across every issuer, and this month we added two new views of it to the State of Auto deck. Both are below.
The whole cycle in one table
The cleanest way to read nine years of loss history is the triangle: one row per origination quarter, one column per age checkpoint, every cell the cohort's cumulative loss-to-liquidation at that age. Color is a rank against every other cohort, so green means cool for its age and red means hot, wherever it sits in the table.
Three regimes are visible without squinting.
The 2017 through mid-2019 cohorts season in the 9% to 16% range. That was the old normal. The cohorts from Q3 2019 through Q2 2021 form a green valley, with the 2020 originations bottoming between 3.4% and 5.1%. Underwriting did not get better in 2020. Those loans liquidated into the 2020-21 used-car price spike, when repossessed collateral sold for numbers nobody will see again. We show those rows but exclude them from the color scale for exactly that reason; letting them define "good" would make everything else look artificially hot.
Then the crest. Every quarter from Q1 2022 through Q4 2023 runs 16% to 19% at maturity, and 2022-Q2 has sat pinned at 17.3% to 17.4% from month 24 all the way through month 48. Cohorts passing through a rough patch bend back down at some point. This one found its level and stayed.
Reading down the MOB 24 column makes the sequence concrete:
| Cohort | LTL at MOB 24 | Read |
|---|---|---|
| 2017-Q1 | 16.0% | old-normal high end |
| 2018-Q2 | 14.9% | old normal |
| 2019-Q2 | 8.2% | entering the COVID valley |
| 2020-Q2 | 4.3% | the used-car price gift |
| 2022-Q2 | 17.4% | the crest |
| 2023-Q3 | 18.4% | on its way to the 19.1% peak |
| 2024-Q2 | 16.3% | newest cohort with a MOB 24 print |
The newest cohorts are high, not peak
The triangle shows where every cohort sits. The curve view shows where the newest ones are heading. Here the full history runs in slate and the last four measurable quarters get color and a label.
The 2024-Q2 through 2025-Q1 cohorts are running 14.7% to 16.3% at their current ages, which puts them in the top quartile of history but visibly below the 2022-23 curves at the same age. Two things matter in how you read those labels.
First, LTL curves rise into roughly month 30 before recoveries and clean payoffs bend them back down, so every one of those numbers is still climbing. Early prints flatter you. The 2024-Q3 cohort looked relatively cool at 12.0% at month 12, then added three full points in the next six months.
Second, the ordering within the four is informative. 2024-Q2 is the hottest at 16.3% and tracking just under the 2023 curves. 2025-Q1 opened at 14.8% at month 12, hotter than three of the four 2024 quarters at the same age. Whatever cooling happened in mid-2024 originations, the newest paper is not extending it.
Prime runs the identical cycle at one tenth the amplitude. The prime triangle peaks at 1.9% for 2022-Q2, bottoms at 0.4% for the 2019-Q4 and 2020 cohorts, and has the newest four quarters at 1.5% to 1.6%. Same cycle, smaller numbers. If you only watch subprime you would call this a subprime story. It is a collateral value story that subprime feels ten times harder.
| Segment | COVID valley floor | Peak (cohort) | Newest four cohorts |
|---|---|---|---|
| Subprime (FICO <660) | 3.4% | 19.1% (2023-Q3) | 14.7% to 16.3% |
| Prime (FICO 660+) | 0.4% | 1.9% (2022-Q2) | 1.5% to 1.6% |
Severity is doing the work
Loss to liquidation moves for two reasons: more loans fail, or failed loans recover less. The second lever has done unusual work this cycle.
We sampled resolved charge-offs from the loan tape, 250 loans per year per segment, and computed loss given default per loan: one minus recoveries collected over charged-off principal. The medians by charge-off year tell the story. Subprime LGD ran 44.5% for 2021 charge-offs, 46.9% for 2022, 56.9% for 2023, and 66.4% for 2024. Prime is the sharper version: 27.3% in 2021, when a two-year-old repossessed car often covered most of the balance, converging to 55.6% by 2023, essentially indistinguishable from subprime that year, before settling at 54.2% in 2024 while subprime kept climbing.
That recovery decay is what pinned the 2022-23 cohorts at the top of the triangle. They originated at peak vehicle prices, so borrowers started with thin equity, and their defaults liquidated into a normalizing used-car market. Both legs of the LTL ratio moved against them at once. You can watch the same dynamic monthly in servicer reports on the Form 10-D side of the dataset, but the loan-level view is what lets you split it by cohort and credit tier.
What to watch from here
Three specific things over the next two quarters of remittance data.
The 2024 cohorts hit months 24 to 30, the age where LTL curves historically peak. If 2024-Q2 rolls over below 17%, the peak stays with 2023-Q3 and the "past peak" narrative survives. If it pushes through, the stress didn't crest in 2023, it plateaued.
2025-Q1's next few prints. Its 14.8% month-12 open was the sixth-hottest of 33 cohorts at that age. One more quarter of data tells us whether that was mix or momentum.
Prime's newest rows. The 2023-Q4 and 2024-Q1 prime cohorts are printing 1.2% to 1.4% through month 18, a genuine step down from 2022's 1.6% to 1.9%. If those rows stay amber instead of red as they age, that is the earliest confirmation the loss cycle is decaying rather than resting.
How we build this
Loss to liquidation is cumulative net losses divided by cumulative principal reduction, built from loan-level ABS-EE filings across all issuers, balance-weighted, loan trusts only. Prime splits at FICO 660 at origination. Each quarterly cohort uses a fixed issuer panel: only issuers whose trusts report through the cohort's display horizon, holding at least 70% of its month-12 balance, so curves never bend because a trust got called. Curves start at month 12, because before that liquidation accrues from month one while charge-off recognition lags the repossession pipeline, and every curve would show the same meaningless ramp. The last two reported months are dropped while recoveries settle.
Both charts ship monthly in the State of Auto deck, and the loan-level data behind them, 9.5 million loans and counting, is what we sell. If you want to run this cut yourself, or a different one, the ABS-EE dataset and plans are the place to start.