Prime Auto Loses Almost as Much per Default as Subprime
When a prime auto loan charged off in 2023, the lender recovered 35 cents on the dollar. When a subprime loan charged off the same year, it recovered 39. Prime lost more.
That is not the relationship most people assume, and it held for more than one year. Measured on identical recovery windows, prime and subprime auto severity have been within a handful of points of each other for most of the last seven years. Prime paper is genuinely safer, but the safety is in how often loans default, not in what happens when they do.
The severity gap is small, and 2021 was the exception
We sampled resolved charge-offs from the loan tape, 400 loans per year per segment, and computed loss given default on each: principal charged off, less recoveries, over principal charged off. Every loan gets exactly twelve months of recovery observation measured from its own charge-off date, so the years are comparable.
| Charge-off year | Prime LGD | Subprime LGD | Prime edge |
|---|---|---|---|
| 2019 | 51.4% | 55.4% | 4.0 pts |
| 2020 | 55.8% | 50.1% | none |
| 2021 | 33.7% | 50.0% | 16.3 pts |
| 2022 | 54.5% | 51.0% | none |
| 2023 | 65.0% | 60.7% | none |
| 2024 | 61.2% | 68.1% | 6.9 pts |
| 2025 Q1 | 67.4% | 66.5% | none |
Before COVID, prime's severity edge was four points. In three of the seven years prime lost more per default than subprime. The one year that matches the conventional story is 2021, where prime LGD dropped to 33.7% while subprime stayed at 50%.
That gap has an explanation, and it is not underwriting. A prime borrower who defaulted in 2021 typically had a two or three year old vehicle that could be repossessed and sold into the most extreme used-car market on record. Subprime collateral was older, higher mileage, and further down the depreciation curve, so it never caught the same lift. The 2021 prime number describes a collateral market, not a borrower.
Both segments now sit near 67%. Prime severity is up 16 points from its 2019 level, and it has not been meaningfully below subprime since 2021.
This is worth stating plainly because the arithmetic runs the other way from intuition. If you price prime paper assuming recoveries will cushion a default, the last seven years say otherwise. What prime buys you is a smaller probability of getting there at all.
Frequency is where prime actually wins
Loss to liquidation is my preferred pool-level view. It asks what share of the principal that actually left the pool, whether through payoff or default, left as loss. We compute it from loan-level ABS-EE filings, splitting prime at FICO 660 and above at origination.
Prime's entire table runs between 0.38% and 1.93%. The subprime version of the same chart peaks above 19%. That is the real gap between the segments, and it is an order of magnitude, driven almost entirely by how rarely prime loans reach charge-off.
Reading down the 24-month column shows the cycle:
| Vintage | LTL at MOB 24 |
|---|---|
| 2018 Q2 | 1.29% |
| 2019 Q2 | 0.89% |
| 2020 Q2 | 0.52% |
| 2021 Q2 | 1.05% |
| 2022 Q2 | 1.91% |
| 2023 Q2 | 1.61% |
| 2024 Q2 | 1.64% |
The worst cell anywhere in prime is 1.93%, and it belongs to 2022 Q2. Cohorts written since are running below it, so prime is past its peak.
Past the peak is not the same as normalized, though, and the pre-COVID baseline is the comparison that matters. The 2019 Q2 cohort sat at 0.89% at the same age. The 2024 cohorts are at 1.64%, about 80% higher. The 0.52% readings from 2020 were never a baseline at all; those cohorts liquidated into the same price spike that produced the 33.7% severity number, which is why we hold the 2019 Q3 to 2021 Q2 window out when setting the color scale on that table.
The curve view says the same thing on the seasoning dimension. The newest four quarters run 1.5% to 1.6%, sitting in the upper half of the historical envelope but below the 2022 curves at every comparable age.
"Prime" hides more than it reveals
Segment averages flatten large differences between shelves. Book-level loss to liquidation as of March 2026, across the prime-dominant captives:
| Issuer | Book LTL |
|---|---|
| BMW | 0.13% |
| Honda | 0.27% |
| Ford | 0.31% |
| Nissan | 0.44% |
| Toyota | 0.52% |
| Ally | 0.74% |
| Hyundai | 1.06% |
| Volkswagen | 1.45% |
BMW's book loses about one ninth of what Volkswagen's does. Both are shelves an allocator files under the same heading. Widen to the mixed books and the spread grows: GM at 2.27%, Mercedes-Benz at 3.14%, CarMax at 3.68%. Those are full books rather than prime-only slices, so some of that is non-prime paper. The point survives the caveat.
Direction differs too, and you can see it without any forecasting model. Set each issuer's four newest vintages against its own history at the same age: GM's recent cohorts run at 0.7% where its historical median is 2.6%, a book that should improve as it turns over. CarMax's newest cohorts sit above its own history. Same segment label, opposite trajectories. The monthly version shows up in servicer reports on the Form 10-D side of the dataset, but the vintage cut needs loan-level data.
What to watch
Three things over the next two quarters of remittance data.
Whether prime severity keeps rising. Prime LGD has moved from 51% to 67% since 2019 with no sign of reverting, and prime has been at or above subprime in four of the last six years. If that persists, prime's loss profile is a frequency story and nothing else.
Whether the 2024 and 2025 prime cohorts hold below 1.6% as they season past month 24. LTL curves keep climbing into roughly month 30 before recoveries bend them down, so today's readings on the youngest cohorts will rise before they settle. The question is where they settle against 2022's 1.91%.
Dispersion between the captives. A widening gap between BMW at 0.13% and Volkswagen at 1.45% means the segment label is losing its diagnostic value, and pricing prime paper off segment averages leaves money on the table in both directions.
How we build this
Loss to liquidation is cumulative net losses divided by cumulative principal reduction, from loan-level ABS-EE filings across all issuers, balance-weighted, loan trusts only, prime defined as FICO 660+ at origination. Each quarterly cohort uses a fixed issuer panel, so curves never bend because a trust got called: only issuers reporting through the cohort's display horizon and holding at least 70% of its month-12 balance are included. Curves start at month 12, because before that liquidation accrues from month one while charge-off recognition lags the repossession pipeline, and every cohort would show the same meaningless ramp. The last two reported months are dropped while recoveries settle.
Loss given default uses a fixed twelve-month recovery window measured from each loan's own charge-off date, sampled 400 loans per year per segment, restricted to charge-offs old enough to have that full window observed. This matters more than it sounds. Summing recoveries over each loan's entire life instead, which is the easier query, gives older charge-offs years of collection and recent ones only months. That single choice moves prime's 2021 figure from 33.7% to 29.6% and its 2022 figure from 54.5% to 43.3%, and it manufactures a severity trend out of nothing but observation time. Loans without a credit score at origination are excluded.
Both LTL charts ship monthly in the State of Auto deck. The loan-level data underneath, 9.5 million loans and counting, is what we sell. If you want to run the prime cut yourself, or slice it by issuer, term, or LTV band, the ABS-EE dataset and plans are where to start.