# Auto Credit Intelligence · What it means

how to read filers and deals: every measure, its limits, the masking arithmetic.

Live page: https://getbookiq.com/intel/#meaning

---

How to read this desk. Two censuses share the room. The Filers side reads 26 SEC filers that hold a consumer credit book, one quarter at a time, from the XBRL facts they tag themselves; every ratio is that filer's own definition, and the bottom of every Filer Read shows which concept resolved. The Deals side reads 44 registered auto securitizations one month at a time from the raw loan tape (form ABS-EE, exhibit EX-102), so the numbers are computed from every loan rather than reported by the servicer. A statement tells you what management concluded; a tape tells you what the borrowers did. Each page opens with a verdict, which is rule-based: the archetype is chosen from the flags in a fixed priority order and the numbers are interpolated from the latest row. Below the verdict, the read is the flag list with its evidence, the numbers are the latest row with a peer percentile, and the charts put the filer or deal against its median. A flag is a question, not an accusation. The composite score is the sum of active flag weights, capped at 10, and it is not a probability.
Filers, question one: is the allowance keeping up?
ALLOWANCE / GROSS
hist alw_ratio = allowance / receivables_gross x 100, quarter end
MEANS
The share of the gross book already written down against expected loss. Under CECL it is the lifetime expected loss on today's loans; under the older incurred model it was losses already probable.
WHEN HIGH
Deep-subprime paper, or a fresh true-up after an under-reserved stretch. Car-Mart's restated allowance of $436.1M was 28.9% of receivables. Credit Acceptance runs above 30% because its book is dealer-holdback paper by design.
WHEN LOW
A prime book, or an under-reserved one. The ratio alone cannot tell you which; provision against charge-offs can.
VS PEERS
Percentile against the filer's basis (its own model where enough filers tag the ratio, otherwise all filers). Low percentile is flagged hot because a thin allowance is the risk.
OUR LINE
coverage_falling fires when alw_chg_yoy is at or below -0.25 points (weight 1, 3 filers now). It has no standalone read line; it is the ingredient in scissors and growth_thin.
LIMITS
Fair-value books carry no allowance at all, so the ratio is blank where the fair_value_book screen fires. A filer that changes its allowance concept mid-history can show a step that is not a credit event; the one-concept rule exists to stop that.
PROVISION / NCO
hist prov_nco = provision / nco, both quarterly flows differenced from fiscal year-to-date spans
MEANS
Whether the quarter's expense refilled what losses consumed. Above 1.00x the allowance is being built; below it the allowance is being spent.
WHEN HIGH
Reserve building, usually ahead of expected deterioration or a growing book. A sudden spike is often a catch-up for prior under-reserving.
WHEN LOW
The allowance is absorbing losses it is not being refilled for. A shrinking book can legitimately run below 1.00x for a while; a growing one cannot.
VS PEERS
Percentile against basis; low is hot.
OUR LINE
reserve_bleed fires at 2 or 3 consecutive quarters under 1.00x (weight 2); reserve_bleed_deep replaces it at 4 or more (weight 3, 1 now). Car-Mart ran 13 consecutive quarters below before restating (latest 2026-04-30: $91.9M provision against $139.9M NCO).
LIMITS
Quarterly differencing amplifies tagging errors; where NCO comes out negative or implausible the screens blank the ratio rather than publish it.
RESERVE COVER, QUARTERS
hist reserve_qtrs = allowance / nco (one quarter's net charge-offs)
MEANS
How many quarters of the current loss run-rate the allowance could absorb with no further provision.
WHEN HIGH
A deep cushion, or a very low current loss quarter flattering the denominator.
WHEN LOW
The allowance is thin against what is actually being lost right now.
VS PEERS
Not percentiled; read against the flag line.
OUR LINE
reserve_thin fires under 3.0 quarters (weight 2, 3 now).
LIMITS
One quarter's NCO is noisy; a seasonal loss quarter can move this a full quarter of cover.
THE SCISSORS
alw_chg_yoy at or below -0.25 points AND nco_chg_yoy at or above +0.50 points, year over year
MEANS
Coverage falling while the loss rate rises: losing more, reserving less, at the same time.
WHEN HIGH
Not a level; a pattern. When it fires the book is in the geometry that preceded the Car-Mart restatement.
WHEN LOW
Not applicable.
VS PEERS
Year-over-year comparisons use the prior observation nearest 365 days back within a 300 to 430 day window, so fiscal-year filers compare like with like.
OUR LINE
scissors weight 3, read severity DANGER, 0 filers now.
LIMITS
Requires both ratios to be tagged in both years; filers with suppressed NCO cannot fire it.
Question two: are losses moving?
NCO, ANNUALIZED
hist nco_rate = (writeoffs minus recoveries) x 4 / receivables_gross x 100
MEANS
The realized loss rate on the book, net of what came back.
WHEN HIGH
Subprime paper: the highest annualized rate on this tape is 30.21%. SGC's BHPH benchmarks put bad debt at 21% in 2022, 24% in 2023 and 28% in 2024, the worst year since 1999.
WHEN LOW
Prime paper, or a filer whose write-off policy is slow (losses arrive later as bigger charge-offs).
VS PEERS
Percentile against basis; high is hot.
OUR LINE
losses_rising fires when nco_chg_yoy is at or above +0.50 points (weight 1, 1 now).
LIMITS
Each filer's write-off timing is its own. Where the concept resolves to gross write-offs only or the flow comes out negative, the screens blank it.
PAST DUE
hist dq_total_r = dq_total / receivables_gross x 100; dq90_r for 90+ only; the board shows dq_total_r with dq90_r as fallback
MEANS
The share of the book behind on payment, as the filer itself aggregates it.
WHEN HIGH
Stress arriving that has not yet become a charge-off. Delinquency leads loss by one to two quarters.
WHEN LOW
A clean book, or a servicer that extends, re-ages or writes off fast enough to keep the number down. The tape side of this desk exists to test that.
VS PEERS
Percentile against basis; high is hot.
OUR LINE
dq_rising fires when the past-due share is up 0.50 points or more year over year (dqtot_chg_yoy, falling back to dq90_chg_yoy), weight 2. dq_vs_reserve fires in addition when coverage also fell 0.25 points, weight 3.
LIMITS
Delinquency bucket tables live in dimensioned XBRL the companyfacts feed does not carry; only filers tagging an undimensioned aggregate show a number here. Where the ratio exceeds 100% the past_due_ratio_impossible screen blanks it.
NONACCRUAL SHARE
hist nonacc_ratio = nonaccrual / receivables_gross x 100
MEANS
Loans the filer has stopped accruing interest on: management's own admission the loan is impaired.
WHEN HIGH
Recognized impairment building. Bank-style filers tag this; most finance companies do not.
WHEN LOW
Either a clean book or a filer that does not use the concept.
VS PEERS
Not percentiled.
OUR LINE
nonaccrual_rising fires when nonacc_chg_yoy is at or above +0.50 points (weight 1, read severity WATCH).
LIMITS
Sparse coverage across this tape.
Question three: is growth outrunning reserves?
BOOK GROWTH
hist growth_yoy = receivables_gross vs the observation one year back, percent
MEANS
How fast the gross book is growing. New loans have not yet had time to default, so fast growth dilutes every delinquency and loss ratio for 12 to 24 months.
WHEN HIGH
Either real demand or an origination push into weaker borrowers. The Fed's May 2026 FEDS note put the average subprime auto loan at $15,402 over 55 months at 25.39% APR.
WHEN LOW
Run-off, a tightened credit box, or a book being sold.
VS PEERS
Percentile against basis; high is hot because growth hides loss.
OUR LINE
growth_thin fires when growth exceeds 10% AND alw_chg_yoy is at or below -0.25 points (weight 2, 0 now).
LIMITS
Acquisitions and consolidation changes show up as growth.
Question four: can the balance sheet take it?
EQUITY / ASSETS
hist equity_ratio = equity / assets x 100; leverage = debt / equity
MEANS
The cushion between the creditors and the losses. For a warehouse-funded lender this is the number the guarantor and the bank both watch.
WHEN HIGH
A thickly capitalized book, or one whose assets were just written down.
WHEN LOW
Creditors own the outcome. Both Chapter 11 filings on this tape ran through here.
VS PEERS
Percentile against basis; low is hot.
OUR LINE
equity_thin fires between 0 and 10% (weight 1); equity_negative below zero (weight 4, 1 now); equity_eroding when eq_chg_yoy is at or below -4 points (weight 2, 6 now).
LIMITS
Buybacks and fair-value marks move equity without moving credit. The Fed survey found 81% of subprime auto lenders warehouse-funded with guarantors and 65% asset-based; the warehouse covenant, not this ratio, is where the line is actually enforced. Cross-reference: loans to nonbank financial institutions across all FDIC banks reached $1.52T at 2026-06, up from $238B in 2015.
Events: what each one legally means
NON-RELIANCE 8-K (ITEM 4.02)
events[].kind = nonreliance; flag nonreliance fires when the event is dated on or before the quarter end and within 400 days of it, weight 3
MEANS
The company or its auditor has concluded that previously issued financial statements should no longer be relied upon. It is the filer telling you its own history was wrong.
WHEN HIGH
Not a level. Every event here was verified by reading the filed document, because EDGAR item codes are wrong often enough to matter.
WHEN LOW
Not applicable.
VS PEERS
Not percentiled.
OUR LINE
Fires from the event, not from any ratio.
LIMITS
The 400-day window means a two-year-old restatement no longer scores; the banner on the Filer Read stays.
LATE FILING (NT 10-K / NT 10-Q)
events[].kind = late_filing; flag late_filing weight 1
MEANS
A Form 12b-25 notice that the periodic report will miss its deadline. It buys 15 calendar days on a 10-K and 5 on a 10-Q.
WHEN HIGH
Usually the allowance, a covenant calculation or a going-concern question is being argued with the auditor.
WHEN LOW
Not applicable.
VS PEERS
Not percentiled.
OUR LINE
Fires from the event.
LIMITS
A missed deadline is itself a default under most credit facilities; the notice tells you the clock is running, not why.
BANKRUPTCY, DELISTING, DEREGISTRATION
events[].kind = bankruptcy (Item 1.03), delisting (Form 25), deregistration (Form 15); status = departed when no 10-K or 10-Q follows
MEANS
Bankruptcy is a court filing under Chapter 7 or 11. A Form 25 removes the securities from an exchange; a Form 15 ends the reporting obligation. Only the last two, with no periodic report after them, make a filer departed here.
WHEN HIGH
Not a level.
WHEN LOW
Not applicable.
VS PEERS
The System tab compares the peak score of departed filers in their final eight quarters against live filers.
OUR LINE
Departed is a status, not a flag; it carries no weight. A Form 25 or 15 with a later 10-K or 10-Q is a single-class delisting and is ignored.
LIMITS
Regional Management's 2024 filing tagged Item 1.03 is a credit-facility amendment, which is why item codes are never scored unread.
The data-quality screens
WHY SOME CELLS ARE BLANK
dq[] on each company: fair_value_book, nco_negative, nco_rate_implausible, past_due_ratio_impossible, alw_ratio_implausible, writeoffs_tagged_zero
MEANS
Each screen names a reason a ratio would be wrong if published: the book is carried at fair value (no allowance exists), differenced NCO came out negative, the annualized loss rate is outside a plausible range, past due exceeds the book, the allowance ratio is outside a plausible range, or write-offs are tagged as exactly zero on a book that clearly has losses.
WHEN HIGH
Not a level: a screen either fires or does not.
WHEN LOW
Not applicable.
VS PEERS
A filer is compared against its own model (card-retail, personal-installment, auto-indirect, lease-to-own) when enough filers of that model tag the ratio to make a median meaningful; otherwise against all filers. The header of every Filer Read says which basis is in use, and the peer percentile uses the same basis.
OUR LINE
The screens are applied before flags, so a suppressed ratio cannot fire a flag either. The Ratios suppressed screen on the Screens tab lists every filer affected.
LIMITS
A blank cell is a decision, not missing data. The definitions block at the bottom of each Filer Read shows what the filer actually tagged.
Deals: loss against seasoning
CUMULATIVE NET LOSS AND THE BAND
month cnl = running sum of (charged-off principal minus recoveries) / snap.init_balance x 100; pf = active balance / initial pool; band = A.band[age].cnl, the median of deals at the same age in months
MEANS
How much of the original pool has been lost for good, and how much of the pool is left to lose more from. The band is computed at equal seasoning, never by calendar date.
WHEN HIGH
A subprime pool two years in: the published read of SDART-2023-6 had cumulative net loss at 9.56% of a $1.06B initial pool at month 30 with a 31% pool factor (the live tape here has since been backfilled, so its current row differs slightly).
WHEN LOW
Prime paper, or a young pool. A pool with a calendar gap in its tape also reads low, which is why gapped deals are rejected before publication.
VS PEERS
Above the dashed line on the chart is worse than peers at that age; a band exists only where two or more deals reached the age.
OUR LINE
cnl_above_band fires when cnl exceeds 1.1 x the band median (weight 2, 19 now).
LIMITS
No presale expected-loss ranges are loaded, so this is a peer comparison only. A fast pool factor concentrates remaining loss in fewer loans.
REPORTED VS EXTENSION-ADJUSTED 60+
month dq60r = balance 60+ days past due / active balance; dq60a adds back loans current today but extended within the trailing 6 months; gap60 = (dq60a minus dq60r) in basis points
MEANS
The flagship metric. An extension moves a delinquent loan to current without a payment catching up; the adjusted series refuses to count that as a cure for six months.
WHEN HIGH
The servicer is leaning on extensions. SDART-2023-6: 10.16% reported against 21.73% adjusted, an 1,157bp gap. The Philadelphia Fed's April 2026 CFI report found the same understatement in headline auto delinquency.
WHEN LOW
Reported and adjusted agree: the delinquency line is what it looks like. Prime pools sit near zero.
VS PEERS
dq_above_band fires when reported 60+ exceeds 1.1 x the band median (weight 1, 18 now).
OUR LINE
masking_severe fires above 200bp (weight 3, 17 now); masking between 75 and 200bp (weight 2); a WATCH line is written at 75bp or under.
LIMITS
Extensions are detected as a period-over-period rise in each loan's cumulative paymentExtendedNumber; a servicer that does not populate the field would read clean by omission.
EXTENSION STOCK, NEW EXTENSIONS, REDEFAULT
month exts = balance of loans extended in the trailing 6 months / active balance; extn = balance newly extended this month / active balance; redefault = share of a month's extended loans that are 60+ again within 6 months (A.redefault)
MEANS
The stock is how much of the pool is currently being carried by the tool; the flow is how hard the tool is being used this month; redefault is whether it worked.
WHEN HIGH
Stock rising is a servicer under pressure. Redefault of 17-22% within six months, the SDART-2023-6 reading, means roughly one extended loan in five comes straight back.
WHEN LOW
Extensions rare and mostly curing: the tool is being used the way the pooling documents intend.
VS PEERS
The Market tab draws prime and subprime extension stock side by side.
OUR LINE
ext_stock_rising fires when the latest 6-month stock exceeds 1.5 x its value 6 months earlier AND exceeds 5% of balance, with at least 7 months of data (weight 2).
LIMITS
Redefault cohorts need six months of observation, so the newest extensions cannot be judged yet.
RECOVERIES
month rec_pct = cumulative recoveries / cumulative charged-off principal x 100
MEANS
How much of what was charged off came back through repossession sale, deficiency collection or insurance.
WHEN HIGH
Strong collateral values or aggressive deficiency collection. SDART-2023-6 recovered 45% of charged-off principal at month 30.
WHEN LOW
Weak used-car values, slow repossession, or a young pool where recoveries have not yet caught up with charge-offs.
VS PEERS
No band overlay on the Deal Read; compare across the Deals table.
OUR LINE
recovery_weak fires under 33% (weight 1, 12 now).
LIMITS
On a prime pool with under 1% cumulative loss the denominator is tiny and the ratio swings; read the verdict's hedge.
ROLL RATES
roll[].c_to_30 = share of current balance at month t that is 30+ or charged off at t+1; b60_co = share of 60+ balance that charges off; b30_worse, b30_cure, b60_cure likewise; balance-weighted, loans on both consecutive tapes
MEANS
The month-over-month migration between delinquency buckets. Current-to-30+ is the earliest stress signal on any tape, because an extended loan re-enters it the next month.
WHEN HIGH
Borrowers newly failing to pay. A rising current-to-30+ roll with flat reported delinquency is extension masking seen from the other side.
WHEN LOW
A stable book, or one where nearly everyone who could fail already has.
VS PEERS
The Market tab shows the median deal's current-to-30+ roll by tier.
OUR LINE
No flag; read the charts.
LIMITS
Small pools late in life have few current loans left, so the roll gets noisy as the pool factor falls.
NEVER-DELINQUENT SHARE
snap.neverdq.n30, n60 = share of loans never 30+ (60+) across their whole life on tape
MEANS
How many borrowers simply paid. Once a loan crosses 60+, failure odds are severe in every tier; what separates prime from subprime is how many loans ever get there.
WHEN HIGH
Prime paper.
WHEN LOW
Subprime paper, or an old pool where nearly every marginal borrower has had a chance to slip.
VS PEERS
The table on the Cohorts tab lists every deal.
OUR LINE
No flag.
LIMITS
A pool that has amortized to a low factor is dominated by survivors, so the share drifts up with age.
VINTAGE DRIFT
drift[] by origination year: wa_term, gt72 and gt84 (share of origination balance over 72 and 84 months), wa_amount, wa_score; each loan counted once at first appearance, cohorts under 100 loans dropped
MEANS
Whether the sponsor wrote later cohorts longer, larger or weaker than earlier ones. A pool is a selected slice, so this is drift in what was securitized, which usually tracks what was originated.
WHEN HIGH
Term stretching past 72 months and loan size rising with score flat is the classic affordability squeeze: the payment is held down by pushing the term out.
WHEN LOW
Stable underwriting, or a sponsor that tightened.
VS PEERS
The Market tab's drift table shows first against latest cohort for every deal.
OUR LINE
No flag.
LIMITS
Reg AB II carries no dealer identifier, so drift cannot be attributed to a source.
THE FIRST-60 CURVE
first60.byYear: for loans that ever hit 60+, the end state by the year they first crossed (charged off, still active, paid, removed); first60.curves: cumulative share failed by months since first 60+
MEANS
What happens after a borrower first goes seriously delinquent. Red is charge-off, blue still active, green paid, grey removed from the tape.
WHEN HIGH
Recent years read low on failure only because their story is not over: the blue is where the red comes from.
WHEN LOW
Not applicable.
VS PEERS
Cohorts under 50 loans are dropped from the bars, under 200 from the curves.
OUR LINE
No flag.
LIMITS
Removed (zero-balance codes 2 and 3) hides repurchases and substitutions that can flatter a pool.
PRIME VS SUBPRIME MARKET LINES
A.market.prime and A.market.subprime: every loan on every tape in this room, balance-weighted by calendar month; fields dq60r, dq60a, exts, extn, co3_ann, repo_pct, c_to_30_med
MEANS
The desk's own version of the public segment trackers, built from loans rather than remittance summaries, so every line decomposes to the deal and the bucket underneath it.
WHEN HIGH
The gap between the tiers is the level story; the gap between subprime reported and adjusted is the masking story.
WHEN LOW
Not applicable.
VS PEERS
These lines describe the deals on tape, not the whole market; the number of deals behind every point is shown.
OUR LINE
No flag.
LIMITS
Coverage thickens as shelves are added; a month with one deal in a tier is that deal, not the market.
THE 144A PERIMETER
A.registry[].reg: true when the shelf files ABS-EE loan tapes on EDGAR; false for 144A shelves, verified by full-text search on 2026-08-30
MEANS
Reg AB II forces the loan tape into the open only for SEC-registered deals. Registration buys a monthly loan-level file for every deal on the shelf; a 144A deal files nothing, by design.
WHEN HIGH
Not a level.
WHEN LOW
Not applicable.
VS PEERS
Registered and visible: Santander DRIVE, AmeriCredit, Exeter, Carvana, World Omni, Ally, CarMax, GM Financial. Dark: Tricolor (failed September 2025), DriveTime/Bridgecrest, GLS, Flagship, Westlake, CPS.
OUR LINE
No flag; the Perimeter tab lists both sides.
LIMITS
CPSS is SEC-reporting as a company while its collateral is dark, so its Filer Read exists and its deals cannot. The lenders most likely to fail choose 144A for exactly this reason.
Four filers you will meet
THE RESTATED BHPH LENDER

An integrated buy-here-pay-here operator that sells the car and originates the note. Provision runs below net charge-offs for quarters on end, coverage drifts down while the loss rate climbs, then a non-reliance 8-K resets the allowance in one line. The tape shows the mechanics for two years before the filing does.

verdict: a restated book · flags: nonreliance, reserve_bleed_deep, reserve_thin, losses_rising
THE RESERVE BLEEDER

Not restated yet. Provision under 1.00x for two, three, four quarters while the allowance ratio slides and reserve cover falls under three quarters of the loss burn. The warehouse bank reads this before the equity market does; on the Issue 9 timeline every failure was funding, not demand.

verdict: a reserve bleed · flags: reserve_bleed or reserve_bleed_deep, coverage_falling
THE ERODING BALANCE SHEET

Equity share of assets down four points or more in a year, sometimes through zero. Both Chapter 11 filings on this tape ran through here, and the departed cohort's peak scores cluster in this screen. The final filings are often the calmest, which is why the read uses the peak quarter.

verdict: a balance sheet under pressure · flags: equity_eroding, equity_thin, equity_negative
THE STEADY PRIME BOOK

Loss rate under 3%, coverage flat, nothing fires. Its value is as the denominator: the same ratios on the same axes let a subprime filer's numbers be read as distance from normal rather than as absolutes.

verdict: a prime reference or a steady book · flags: none
Four deals you will meet
THE EXTENSION-MASKED SUBPRIME POOL

Reported 60+ under 10%, adjusted 60+ above 20%, a quarter of the pool touched by an extension in six months, one extended loan in five back at 60+ within six months. Cumulative loss above the band and rising as the extensions age out.

verdict: an extension-masked pool · flags: masking_severe, cnl_above_band, dq_above_band
THE POOL LOSING ABOVE THE BAND

Cumulative net loss 10% or more above the median at the same age, with delinquency in line. Either the sponsor's credit box was wider than peers or recoveries are weak; the recovery rate tells you which.

verdict: a pool losing above the band · flags: cnl_above_band
THE PRIME REFERENCE

Cumulative loss under 1% at month 30, reported and adjusted delinquency within a few basis points, extensions near zero, pool factor falling fast on prepayments. The risk here is speed, not credit.

verdict: a pool seasoning as expected · flags: none, or recovery_weak on a tiny loss base
TOO YOUNG TO READ

Under six monthly tapes. Cumulative loss near zero by construction, no band at this age, no redefault cohorts observed yet. The current-to-30+ roll is the only early tell.

verdict: too young to read
What this desk cannot tell you
Dealer identity. Reg AB II carries no dealer identifier, so no loan on any tape can be traced to the store that wrote it, and no filer breaks its book out by source. Servicer practice. The tape records that an extension happened, not why, not whether a payment was demanded, not what the collector said. Anything inside 144A collateral. Tricolor, DriveTime, GLS, Flagship, Westlake and CPS issue 144A; their deals do not exist here, and the desk's prime and subprime lines describe registered paper only. Post-filing events. A quarter that has not been filed and a month whose tape has not landed are blank; a failure between filings is invisible until the next one, and a departed filer's tape ends with its last report. Definitions. Every filer defines its own past due, its own charge-off timing and its own allowance concept, and those definitions travel with the number; the desk normalizes the field names, never the accounting. On the credit union side of the house, purchased vehicle participations reached $13.5B and participation delinquency ran above the overall book in 2025, which is the same story told from the buyer's side; it is not on this desk either.