Subprime Recovery by Make: Old Toyota Beats New BMW
Repossess a subprime Toyota and 54 cents of every dollar written off comes back. Repossess a BMW and you get 36. That is an eighteen point spread on identical credit, and the obvious objection is that it has nothing to do with the badge: subprime lenders finance Toyotas nearly new and BMWs nearly worn out, so the league table might be measuring odometers.
The objection is half right. Vehicle age explains about a third of the Toyota-to-BMW gap. The badge keeps the other two thirds.
The raw league table
We pulled every subprime loan in the securitized data, 2017 through 2023 vintages, origination FICO under 660, loan trusts only with leases excluded. Recovery is recovered amount over gross charge-off principal on the same cohort, measured across months on book 1 to 36. Makes qualify with at least 8,000 subprime loans and $100M of original balance behind them.
Toyota leads at 54%, Honda follows at 50%, and the bottom four are German and British: BMW 36%, Audi 36%, Mini 37%, Mercedes 37%, with Lincoln level at 37%. The subprime average across these makes is 45%.
Everyone who has priced a subprime pool has a story for this, and most of the stories are about resale demand. Those stories are probably right, but they are untested until you deal with what the lenders actually financed.
The metal really is older
Weighting each make's charge-off dollars by the vehicle's age at origination gives you the collateral profile behind the number:
| Make | Age at origination | Raw recovery |
|---|---|---|
| Mitsubishi | 1.2 years | 44.4% |
| Dodge | 2.4 | 48.3% |
| Jeep | 2.6 | 48.4% |
| Toyota | 2.9 | 54.2% |
| Hyundai | 3.4 | 41.9% |
| Ford | 3.9 | 43.3% |
| Honda | 4.2 | 49.7% |
| Cadillac | 5.0 | 38.4% |
| Mercedes-Benz | 5.1 | 37.4% |
| BMW | 5.3 | 35.9% |
| Lexus | 5.5 | 47.7% |
A subprime BMW is 2.4 years older at origination than a subprime Toyota. That is a real difference and it points the way the skeptics expect.
Age is also the strongest single thing in this dataset. Pooled across all makes:
| Vehicle age at origination | Recovery | Share of charge-off dollars |
|---|---|---|
| Current model year | 51.5% | 22.3% |
| 1-2 years | 49.0% | 20.6% |
| 3-5 years | 43.1% | 38.0% |
| 6-8 years | 35.4% | 15.5% |
| 9 years or more | 28.9% | 3.6% |
That is a 22.6 point range, wider than the spread between any two makes on the raw chart. Nothing else we cut moves severity that far.
So the question is whether the make ranking survives holding age constant.
Same age mix, different answer
Direct standardisation is the clean way to do it. Take each make's recovery rate within each vehicle age band, then reweight those rates to the charge-off age mix of the entire subprime pool. Every make gets scored as though it had financed the same distribution of vehicle ages.
Toyota barely moves, from 54.2% to 53.5%. BMW climbs from 35.9% to 41.0%. The gap narrows from 18.3 points to 12.5, so vehicle age accounts for 32% of it. Toyota still finishes first and BMW still finishes in the bottom quarter, but BMW goes from dead last of 26 makes to 23rd, which is a different sentence about the same collateral.
The makes that gain most are exactly the ones the objection predicted:
| Make | Raw | Age-adjusted | Shift |
|---|---|---|---|
| Audi | 36.4% | 43.4% | +7.0 |
| Mini | 37.0% | 43.4% | +6.4 |
| Acura | 44.1% | 49.3% | +5.2 |
| BMW | 35.9% | 41.0% | +5.1 |
| Mercedes-Benz | 37.4% | 41.7% | +4.3 |
| Infiniti | 37.9% | 42.1% | +4.2 |
| Cadillac | 38.4% | 41.5% | +3.1 |
Audi and Mini move from the bottom three of the raw chart into the middle of the pack. If you have been applying a haircut to European collateral in subprime pools, some of that haircut is paying for vehicles the lender chose to finance old, not for the marque.
Outside the screened universe the effect gets larger still. Porsche runs 36.8% raw and 45.7% adjusted, a nine point swing, though that sits on 2,135 loans and I would not price anything off it. Volvo gains 8.3 points on 5,695 loans. Both are thin enough to be indicative rather than usable.
Mitsubishi is the one to worry about
The adjustment cuts the other way too, and the biggest loser is the most interesting number in this exercise.
Mitsubishi prints 44.4% raw, eighth of 26 makes, comfortably mid-table. Age-adjusted it drops to 39.6%, second to last. The reason is in the first table: its subprime book averages 1.2 years old at origination, by a wide margin the newest collateral any make brings. Strip that advantage out and there is nothing underneath it.
Dodge loses 2.2 points, KIA 1.6, Jeep 1.5, Nissan 1.4. These are the makes whose raw numbers are flattered by young books. Nobody is going to reprice a pool over 1.5 points, but Mitsubishi's 4.8 is large enough to matter, and it is the sort of thing that only shows up when you force the comparison.
Japanese versus European is the wrong split
The framing that motivated this analysis was Japanese against European. It holds, but it is not where the action is. Recovery by origin region, within each age band:
| Region | New | 1-2yr | 3-5yr | 6-8yr | 9yr+ |
|---|---|---|---|---|---|
| Japanese | 54% | 50% | 45% | 38% | 31% |
| Domestic | 50% | 49% | 43% | 36% | 30% |
| European | 52% | 46% | 39% | 32% | 25% |
| Korean | 52% | 46% | 40% | 29% | 22% |
Japanese collateral beats European in every band, by two points when new and six once the vehicle is past three years. That is a real and consistent edge.
But look at the bottom right. Korean collateral starts even with European and ends four points below it. Hyundai loses 25 points of recovery between the 1-2 year band and the 9 year band, KIA loses 24, against 18 for Toyota and 19 for BMW. Korean vehicles depreciate off a cliff in the back half, and because subprime lenders finance them young, the raw league table never shows it. KIA and Hyundai look average at 42% raw. On old metal they are the worst collateral in the dataset.
That is the finding I did not expect. The European penalty is mostly a mix effect. The Korean penalty is a genuine curve, and it is hidden.
What survives, and what it means for pricing
Three things hold up after the adjustment.
Toyota and Honda are genuinely better collateral, not better-financed collateral. Toyota gives up only 0.7 points to standardisation and still leads. The cleanest way to see it is a single cross-comparison that needs no adjustment at all, because it holds age constant by construction: a six-to-eight year old Toyota recovers 47 cents, and a one-to-two year old BMW recovers 43.
Age dominates within any make. Toyota itself runs 61% on current model year collateral and 38% on nine-year-old cars. A pool's vehicle age distribution tells you more about severity than its make distribution does, and age at origination is a field you can read straight off the ABS-EE tape.
And the badge is worth roughly 12 points at the extremes once age is controlled, down from 18. That is still a lot. It is just not 18.
The practical version of this is a sensitivity, not a forecast. Every band above is indexed to vehicle age at origination, so a pool's exposure is set the day the loans are written. What the Korean curve says is that the cost of drifting older is not the same for every make. Move a book of Toyotas from three-to-five year old collateral into six-to-eight and recovery falls five points. Make the same move in KIA and it falls eleven, then another six if the book slides past nine years. KIA and Hyundai look average at 42% raw, and that average is a statement about what lenders currently choose to finance rather than about the collateral. It holds only as long as the buying stays young.
How we built this, and what is not controlled
Recovery is recovered amount over gross charge-off principal, both accumulated over months on book 1 to 36, on loans observed from within three months of origination. Vehicle age is model year against origination year, taken from the gold reporting attributes table where make normalisation already merges the TRUCK and TRUCKS filer suffixes into the base make. The standard population is the charge-off dollar mix of the whole subprime pool: 22.3% current model year, 20.6% one to two years, 38.0% three to five, 15.5% six to eight, 3.6% nine or more. Makes need at least four qualifying age bands to appear.
Two things are not controlled, and both could move the residual.
The recovery window is fixed at 36 months on book, not measured from each loan's own charge-off date. A loan that charges off at month 34 contributes its full principal to the denominator and about two months of collections to the numerator. If makes differ in when their loans default, that biases the ratio for reasons that have nothing to do with collateral. This is the same censoring trap that distorts loss given default when you let observation time vary, which we wrote about for prime severity, wearing a different costume.
The second is the issuer. Make and lender are heavily confounded in this data. CarMax's book is used Japanese metal; the German captives finance their own new cars. Repossession speed, auction channel, and collection effort vary enormously between servicers, and some unknown share of that remaining 12.5 points is execution rather than steel. A within-issuer cut would separate them and it is the next thing on my list.
Neither caveat touches the old-Toyota-versus-new-BMW comparison, which is why I lead with it.
The warehouse behind this holds 35.4 million classified loan contracts across 20 issuer families, with model year, model, new/used, vehicle value, and origination LTV on the loan record. If you want to run the make cut against your own pools, or slice recovery by issuer and age together, the ABS-EE dataset and plans are the place to start.