Loss to Liquidation by FICO: Six Bands Above Range

Of every dollar of principal that left the 2025 Q1 auto cohort during its first twelve months, loans written to borrowers scored 670 to 689 at origination gave back 5.22 cents as net loss. The 24 prior quarterly cohorts in that same FICO band ran between 2.16 and 5.00 cents. The newest one is the first to clear five.

That is not the band anyone watches. It sits in the middle of the credit ladder, and five other bands next to it are doing the same thing. Neither end is.

What loss to liquidation actually measures

Loss to liquidation is a ratio of two cumulative dollar amounts, both taken from loan-level ABS-EE filings.

The numerator is cumulative net loss: charged-off principal, less whatever came back as recoveries. The denominator is cumulative principal reduction, meaning every dollar of principal that has left the pool by any route. Scheduled amortization, prepayment, payoff at trade-in, charge-off, all of it.

So the question it answers is narrow and useful: of the principal that has already exited this cohort, what share exited as loss?

A worked version. Take a cohort that originated $1.0B. Twelve months later, $300M of principal has left the pool and $21M of that departure was net loss rather than cash. Loss to liquidation is 7.0%. The cumulative net loss rate over original balance would read 2.1% for the same cohort, because it divides by the full $1.0B including the 70% that has not gone anywhere yet.

Those two numbers describe the same cohort and behave very differently. The cumulative net loss rate is dragged around by how fast a pool runs off. A cohort with heavy prepayment shows a low loss rate early because the denominator never shrinks, not because its credit is better. Loss to liquidation removes that by only counting principal that has resolved.

What it is not is loss given default. Loss given default asks what a single defaulted loan recovers. Loss to liquidation blends two things: how often loans default, and how much is lost when they do, both scaled against the pace of ordinary runoff. A cohort can move on this metric because defaults rose, because auction proceeds fell, or because healthy borrowers stopped prepaying. Reading it means holding all three possibilities open.

I covered the segment-level version of this metric in the loss-to-liquidation triangles earlier this year. This post cuts it by FICO band instead.

Why a corridor instead of a level

Plotting loss to liquidation against FICO produces a line that falls from left to right, and that is the whole content. Subprime loses more than prime. Nobody needed loan-level data to learn that.

The comparison that carries information is each band against its own record. So for every FICO band I take the prior quarterly cohorts at the same age, compute the 15th and 85th percentile of their loss to liquidation, and shade the space between. That band of shading is where roughly seven of ten historical cohorts landed. Then the newest cohort goes on as a single line across all eleven bands.

Now the chart reads as position rather than level. A band sitting above its own shading is running worse than that credit tier normally does at that age, whether its absolute number is 20% or 0.3%.

Six bands out of eleven

At twelve months on book, the 2025 Q1 cohort sits above the 85th percentile of its own history in six consecutive FICO bands.

The 630-649 band is the one gap in the run, at 9.45% against an 85th percentile of 10.01%.

Both ends of the ladder are unremarkable. The sub-550 band reads 20.48% against a historical middle of 15.24% to 22.65%, which places it just above its own median and nowhere near its record. The 720-759 band sits below its 85th percentile at 1.19%. The 760-plus band reads 0.27% against a range of 0.20% to 0.32%.

That shape is the finding. The deterioration is concentrated between 570 and 719, and the deep subprime tail that usually leads a credit turn is behaving normally. Two of the six bands, 670-689 and 590-609, are at or past the worst reading any prior cohort produced, though the 590-609 margin is two basis points and I would not lean on it.

How much a twelve-month reading is worth

A cohort measured at twelve months is not finished. Charge-off recognition lags the liquidation that produced it, and recoveries arrive for months afterward, so the number keeps moving. The question is how much.

I took every band and quarterly cohort with enough reporting history and compared its reading at each age against its own reading at 24 months on book.

So a twelve-month reading is carrying about four fifths of where it settles. The curve flattens closer to 18 months than to 12, and it tends to crest near 24 before easing slightly as late recoveries land.

The more practical test is whether the corridor flag itself carries over. Of the cohort-band cells that printed above their own 85th percentile at twelve months, roughly six in ten were still above it at eighteen. Four in ten had dropped back inside. The reverse error is about as common: a similar share of the eighteen-month flags were not visible at twelve months at all.

Those flips are small in magnitude. The cells that dropped back inside landed a median of 3% below their own 85th percentile, and every one of them was still above its band's median. Falling back inside the corridor means the reading softened from elevated to ordinary, not that it was wrong about direction.

Margin at twelve months separates the durable flags from the fragile ones. Cells that cleared their 85th percentile by more than 15% held every time. Cells that cleared it by less than 5% held about six times in ten.

The pattern that produced a false signal

One cohort in the record is worth studying before trusting any twelve-month read.

The 2024 Q1 cohort lit up ten of eleven FICO bands at twelve months. By eighteen months, nine of them were back inside their corridors. That single cohort accounts for most of the disagreement between the two ages in the entire history.

The mechanism is visible in the numbers. Its own loss to liquidation barely moved between the two ages. The sub-550 band went from 22.65% to 23.18%, and the 650-669 band went from 7.60% to 7.61%. What changed is that the comparison cohorts kept climbing and passed it. 2024 Q1 did not improve. It stood still while the reference set caught up, which is what a recognition-timing artifact looks like rather than a credit event.

The tell was in the margins. 2024 Q1 cleared its 85th percentile by 2% to 10% in every band at once. A cohort that flags everywhere by a narrow margin is a different object from one that flags in six bands by wider margins while three other bands sit clearly inside, which is what 2025 Q1 does.

Method and limits

The metric is cumulative net loss divided by cumulative principal reduction, computed from loan-level filings across every reporting issuer, balance weighted, loans only with leases excluded.

A quarterly cohort enters the chart only once all three of its monthly vintages have reported through fourteen months on book, so no reading is taken inside the two-month window where recoveries are still arriving. Cells holding under $25M of original balance are dropped.

The COVID handling deserves a note, because the obvious version of it is not enough. Cohorts originated between 2019 Q3 and 2021 Q2 liquidated into the 2020-21 used-car price spike and produced loss-to-liquidation readings nobody will see again, so they are excluded. But excluding by origination quarter alone still leaves a problem: a 2018 Q1 cohort originates well outside that window and its 24-month and 36-month readings still land inside the price spike. For the seasoning tests above I required the entire measurement window, origination through the age being measured, to sit outside the distorted calendar period.

That produces an honest baseline and an awkward one. The clean history becomes a barbell of 2017 through 2018 and 2022 H2 through 2024, with nothing between. Those are different credit environments treated as one reference set, and no additional data fixes it, because undistorted liquidations genuinely do not exist for that stretch. Every percentile in this analysis rests on that assumption.

Rebuilding the corridor from that stricter clean history changes none of the six flags, which is the strongest robustness check the data supports.

Loan-level loss and recovery fields, cohort curves, and the underlying filings are available through the ABS-EE dataset and remittance data. Plans are on the pricing page.