Motorcycle ABS Has a 670 Credit Score Floor
Across every loan Harley-Davidson has ever put into a public securitization, 332,572 loans and $7.09 billion of original balance filed monthly with the SEC since May 2019, the minimum credit score is 670. Not roughly 670. Exactly 670, with a 1st percentile of 672 and a median of 755. That is an eligibility floor, not a thin tail.
What is actually in public motorcycle ABS
Two issuers file motorcycle collateral under Reg AB-II. Harley-Davidson Motorcycle Trust is the only dedicated motorcycle shelf. The other is BMW, whose Vehicle Owner Trust filings carry 22,511 motorcycle loans ($400 million of original balance) mixed in with the cars, identifiable by the filed vehicle type code. I scanned every other issuer in our coverage for motorcycle collateral by vehicle type and make; there is none.
The BMW book reaches lower, with a minimum score of 600, but not by much: 9.1% of its motorcycle loans sit below 670 and 4.5% below 650, on a median score of 767. So the two books together give you roughly 355,000 loans of loan-level motorcycle performance history, and effectively all of it is prime.
The floor is a fact about the securitized collateral, not a fact about motorcycle lending overall: what reaches the trusts starts at 670, and the filings cannot say what stays off them.
Within the observed range, the book has a clean gradient. I banded Harley's collateral at 670-699, 700-724, 725-749, 750-774, and 775+:
| Band | Share of loans | WA APR | WA term (mo) | Avg loan-to-value* |
|---|---|---|---|---|
| 670-699 | 12.3% | 11.03% | 75.5 | 104.7% |
| 700-724 | 15.8% | 9.52% | 74.2 | 105.3% |
| 725-749 | 18.0% | 8.33% | 73.0 | 103.6% |
| 750-774 | 17.8% | 7.30% | 71.8 | 101.2% |
| 775+ | 36.2% | 6.51% | 68.7 | 91.6% |
*Loan amount over the filed vehicle value at origination.
More than a third of the book scores 775 or better. The collateral is young: 51.8% of loans finance the current model year and 68.6% a bike two years old or newer, with 59.1% of units sold new.
Delinquency against autos, band by band
Comparing motorcycle ABS to auto ABS at the pool level gets the sign wrong. In June 2026 the Harley pool shows 2.59% of active balance 30 or more days past due (DPD) while our auto-loan benchmark pool shows 6.25%. That gap is composition: the securitized auto pool carries the large subprime shelves, and the motorcycle pool starts at 670. Hold the score band constant and the comparison flips.
30+ DPD as a share of active balance, June 2026 collection period:
| Band | Harley motorcycles | Auto benchmark |
|---|---|---|
| 670-699 | 5.87% | 4.84% |
| 700-724 | 4.00% | 2.67% |
| 725-749 | 2.74% | 1.57% |
| 750-774 | 1.57% | 0.89% |
| 775+ | 0.65% | 0.28% |
The auto benchmark is every other issuer in our coverage, retail loan trusts only, measured with identical definitions on the same filing months.
The June snapshot overstates how one-sided this has been. At 700 and above, the motorcycle bands have run above their matched auto bands in roughly two-thirds to five-sixths of months since 2020. The 670-699 band behaves differently: it ran below the matched auto band for most of 2020 through 2024, and has been above it in every month since September 2025. June-over-June, Harley's 670-699 band went from 2.99% (2024) to 3.86% (2025) to 5.87% (2026), while the matched auto band moved from 3.96% to 4.33% to 4.84%. The lowest motorcycle band is deteriorating faster than its auto counterpart, from a lower starting point.
Charge-off counts tell the same story with a growth caveat. Annual charge-off cohorts grew from 55 loans (2019) to 1,901 (2024) and 2,693 (2025), with 1,325 already in the first half of 2026. The filed book grew over that period too, so raw counts overstate the shift, but the vintage curves below, which are normalized to cohort balance, move the same direction.
A filing quirk that produces 100% recovery rates
Before any loss number, a finding about the filings themselves, and a worked example of why loan-level data needs cleaning before it needs charting.
The first pass at recovery rates divided cash recoveries by the filed charged-off amount, which is the obvious calculation and the one the raw fields invite. It returned 101% for the completed 2019-2021 charge-off cohorts. A rate above 100% is either a data problem or a story, so I audited every charged-off loan's monthly record. The first suspect, recoveries filed as running totals and double-counted across months, was ruled out quickly: the typical charged-off loan shows one or two recovery months (ninetieth percentile: three), which is the shape of a collateral sale, not a re-filed cumulative field.
The actual mechanism shows up when you split the 936 charged-off loans from those completed cohorts by how large their recoveries are relative to the filed charge-off amount, then reconcile each group against the loan balance entering the charge-off month:
| Recoveries vs filed charge-off amount | Share of loans | Filed amount / balance (median) | Filed amount + recoveries / balance (median) |
|---|---|---|---|
| At or below 1x | 54.6% | 1.00 | 1.45 |
| 1x to 2x | 19.3% | 0.69 | 1.49 |
| Above 2x | 26.1% | 0.24 | 1.05 |
The top group files the full balance as the charged-off amount; recoveries then come in on top of it. The bottom group files something very different: an amount around a quarter of the balance, and adding its recoveries back reconstructs the balance almost exactly. That is a charge-off filed net of expected sale proceeds, with the proceeds then filed again as recoveries. Divide recoveries by that netted amount and you have counted the sale twice, which is where the 101% came from, and why net losses computed the same way printed near zero or negative.
The fix is a unified basis: gross loss is the loan balance entering the first charge-off month, and net loss is that balance minus all recoveries collected. For the loans that file gross, the two bases are identical. On that basis the completed cohorts' recovery rate drops from 101% to 62-66%, and every loss figure below uses it.
Harley is not the only issuer that files this way. Running the same signature test across the other nineteen issuers in our coverage shows the large subprime shelves file the full balance cleanly, while several prime captives file partial, netted, or near-zero charge-off amounts on most of their charged-off loans. The auto benchmark figures below are therefore computed on the same balance basis, both sides of every comparison. Nothing in the pool-level reports would surface any of this; it is only visible when each loan's charge-off row is reconciled against its own balance history.
Losses, and what the metal brings back
Cumulative net loss (CNL) here is the balance charged off minus recoveries, cash basis, as a percentage of cohort original balance. At the 670-699 band, month-36 CNL by origination vintage:
| Vintage | Harley 670-699 | Auto 670-699 |
|---|---|---|
| 2019 | 0.63% | 1.03% |
| 2020 | 1.23% | 0.90% |
| 2021 | 2.10% | 1.54% |
| 2022 | 2.02% | 2.53% |
The 2019 and 2022 vintages came in below the matched auto band; 2020 and 2021 came in above. In the middle of the ladder the comparison tilts against the motorcycles: Harley's 2021-vintage CNL runs roughly twice the matched auto band at 700-724 through 750-774 (700-724: 2.18% vs 0.94%), while at 775+ the two are identical at 0.26%. The absolute levels stay low because this is a 670-plus book, but the mid-band direction matches the delinquency tables: at the same score, motorcycle loans miss more payments and, in most of these cells, lose more.
What keeps the loss numbers as low as they are is the collateral. Against $157.5 million of balance charged off across 8,940 loans, Harley trusts have collected $72.9 million of recoveries to date, 46.3%, against 42.4% for the auto benchmark pool on the same basis (2019-2025 vintages). The cohort view is more informative than the blend:
| Charge-off year | Loans | Recoveries / balance charged off | Share of recovery cash in months 0-3 |
|---|---|---|---|
| 2019 | 55 | 62.0% | 75.6% |
| 2020 | 400 | 66.1% | 78.0% |
| 2021 | 524 | 65.6% | 82.5% |
| 2022 | 862 | 52.6% (accruing) | 72.1% |
| 2023 | 1,180 | 46.9% (accruing) | 67.9% |
Completed cohorts recovered 62-66% of the balance charged off, well above the auto pool's rate, and three-quarters or more of the recovery cash arrived within three months of the charge-off: collateral sales, not decade-long deficiency collection. The caveat is the trend. The 2019-2021 cohorts sold into a strong used-vehicle market; the 2022-2025 cohorts are tracking in the mid-40s to low-50s, closer to the auto pool's rate, and where they settle is the number I would watch. The recovery rate also rises with score, from 43.5% at 670-699 to 53.2% at 775+ across all cohorts to date.
Two more behaviors are worth knowing. First, this collateral pays off fast: the trailing-twelve-month conditional prepayment rate (CPR) on the Harley book averages 20.8%, reached 25.3% in June 2026, and loans representing roughly half of a vintage's original balance have fully paid off by month 36. Second, late-stage delinquency resolves quickly: on a trailing-twelve-month basis, 40.3% of the 90+ DPD bucket rolls to charge-off each month, while 92.7% of current loans stay current month to month.
Methodology notes
Source is the loan-level ABS-EE data filed with the SEC, which we parse monthly; pool-level remittance filings cannot support any of the cuts above. Delinquency is the filed days-past-due status measured against active balance (loans with balance above zero). Charge-off events are the first month with a zero-balance code of 4 or a filed charged-off amount; gross loss is the balance entering that month on both the motorcycle and auto sides, which also captures the 137 Harley charged-off loans (1.5%) that filed no amount at all. Loans enter the data at trust cutoff, typically six to fourteen months seasoned depending on vintage, so pre-securitization performance is unobserved on both sides; the vintage comparisons here use the 2019-2022 cohorts, the ones with a full 36-month observation window. Recent-month cells can revise as issuers re-file, and a small share of the auto pool's June 2026 filings was still pending when I pulled this.
The full cut set behind this post, delinquency, loss curves, prepayment, recoveries, and roll rates by band, month, vintage, and month-on-book, is the kind of extraction we build for enterprise clients against any slice of the filed universe.