Inclusion criterion
A facility is included when the amount the borrower may draw is limited by a borrowing
base computed over eligible accounts receivable and/or eligible inventory at stated advance
rates, less reserves — irrespective of form. Revolving credit facilities, term loans and
FILO tranches all qualify when this structure governs availability, and the form is recorded
with each verdict. The label “asset-based” plays no role in classification: it is
neither necessary (many executed ABL agreements never use it) nor sufficient (it saturates
asset-backed securities material).
Facilities stated in another currency
Twenty-one facilities state their commitment in Canadian dollars, euro, sterling or
Australian dollars. The size column shows them converted to US dollars so that the column
sorts and totals, marked with the currency they came from; open the record and the original
amount is shown as the document states it, followed by the approximate dollar equivalent, the
rate used and the date that rate was published. Rates are the European Central Bank’s daily
euro reference rates, and each facility is converted at the rate for its own agreement date
rather than today’s — a euro facility signed in 2002 converted at a 2026 rate would be wrong
by about a third. Reference rates are not dealt rates, which is why every converted figure is
labelled approximate. Two facilities carry no usable agreement date, so they are shown in
their original currency and left out of the total.
What the committed total means
The figure beside the row count adds the first-stated dollar commitment of every
selected facility that states one, which for a multi-tranche agreement is the revolver rather
than an accordion ceiling. It is a floor twice over: roughly half of agreements state no
figure at all, and a facility contributes the amount written when it was signed, not what was
drawn or what later amendments changed it to. Selecting a year range also drops facilities
whose agreement could not be dated, so the total then matches the vintage series above.
Without a year range it additionally includes facilities with no usable date, among which a
borrower whose undated documents could not be chained may appear as more than one facility.
Corpus construction
The corpus is enumerated from SEC EDGAR full-text search over the structural term
"borrowing base", 2001 to present — documents across
companies, complete in every year. Full-text search indexes
exhibits, so an executed credit agreement filed as EX-10.x is found directly
rather than through its parent form.
The anchor was audited with a supplementary sweep of six product-name phrases
(“asset-based lending”, “asset-based revolving credit facility”, …):
55,024 hits, 29,907 documents the structural term had not surfaced. Of these, 99.6% contain
no borrowing base of any kind, and the audit recovered one borrower the anchor had missed —
a measured miss rate of 1 company in 4,645.
Screening
A deterministic screen separates working-capital ABL from the other lending products that
compute a borrowing base — reserve-based oil & gas lending, commercial-real-estate pools,
fund-finance and BDC portfolio facilities, mortgage warehouses, securitisations — before any
language model is invoked. Across the full corpus, 80% of borrowing-base language is not
working-capital ABL. Documents the screen excluded:
Extraction and verification
Documents that survive the screen are read by a dedicated extraction agent, one agent per
document, against a targeted excerpt of the definitions article and availability covenant.
Every extracted value must be supported by a verbatim quotation from the filed document. Each
quotation is mechanically matched back against the source; claims whose quotations cannot be
located are discarded and counted rather than shown — quotations
located to date. Quotations marked fragmented matched
clause-by-clause in document order rather than as one contiguous span;
weak marks a short span that recurs within the document and
therefore fixes its value less precisely.
Sample and measurement notes
The term distributions count each borrower once.
executed agreements — the strongest per company —
inform the charts, drawn from ABL documents read.
Quarterly restatements and superseded amendments are read and shown in company
timelines but excluded from the distributions: a facility that lives ten years
files one agreement, several amendments and forty restatements, and counting
them all would weight terms by longevity and amendment habit.
Advance rates below % are excluded as clause
misreads. Dilution reserves and permitted-reduction provisions parse as
percentages but are not advance rates; observations were removed
under this rule. Genuinely low rates are retained — tiered agreements set investment-grade
receivables in the low sixties.
Administrative agents are consolidated by institution. Filing-name
variants are collapsed case-insensitively so one lender occupies one row; distinct
affiliates, such as a bank and its capital-finance subsidiary, are kept separate. The
eight most frequent agents arrange % of the sample read.
Industry is the filer’s own classification, not ours.
Each borrower is placed in the SIC division of the code the Commission assigned it,
read from its EDGAR submissions record — borrowers
carry one. Manufacturing is split at the standard non-durables/durables line rather
than left as a single division holding a third of the sample. Filers with no code, and
the handful the endpoint does not recognise, are shown as Unclassified rather
than assigned a division by inference from the company name.
Lender type describes the named administrative agent.
A charter test on the agent’s name — bank, banc, national association,
savings, trust company — is combined with a hand-built list of institutions the
test cannot see: bank-owned asset-based units that carry no bank token
(Wells Fargo Foothill, Congress Financial, LaSalle Business Credit), and
finance companies, credit funds and captives that are not banks whatever their name
suggests (GE Capital, the CIT Group, MidCap, Gordon Brothers).
facilities resolve; the rest are
Unclassified and stay that way. An unrecognised name is never defaulted to
non-bank, because that would inflate the one figure this classification exists to
measure. The label is about the agent, not about the ultimate source of funds: a
chartered trust bank acting only as agent reads as a bank, and a loan-agency business
doing the same job reads as a non-bank.
No time series is drawn yet. With the tier-one census complete —
a median of agreements per filing year — time variation in
terms is now measurable in principle, but agreements enter the sample at their filing
date, so year-over-year comparisons mix vintage effects with composition; a
properly controlled view is future work rather than a chart withheld.
Coverage limits
Private borrowers are invisible. Companies with no SEC filing
obligation cannot appear. Industry survey estimates place a large share of ABL — near
40% — in bilateral loans to private companies, outside the reach of any filings-based
method.
Benchmarked against Loan Pricing Corporation. Compared with LSEG
DealScan, the industry's record of syndicated lending: of LPC’s 617 US asset-based
borrowers with a stock ticker since 2001, 89% appear in this registry; 9% were found
and excluded by the structural definition (LPC’s “asset-based” tag also
covers cash-flow revolvers and receivables securitisations); 2% have no borrowing-base
document on EDGAR. No registrant with a borrowing-base filing was missed. 81% of
LPC’s asset-based borrowers have no ticker at all — the private market a
filings-based registry cannot see. Where both record the same facility, agreement dates
agree within a week in 95% of cases; the stated facility size matches LPC exactly in 65%
of pairs and within 5% in 68% — the remainder is largely this registry reporting one
tranche of a package that LPC records whole. No LPC data is published here.
2001 onward only. EDGAR full-text search does not index earlier
filings.
Terms reflect the documents read. A company row leads with terms from
its strongest executed agreement; its other ABL filings are listed in the row’s
timeline and reflected once read. Amendments in unread filings are not yet shown.
Detection identifies facilities. The registry says nothing about
credit quality, deterioration or performance.
Getting the data out
Export CSV , above the table, writes whatever the table is currently
showing to a file — every row the filters left, not just the page on screen, in the
columns the view displays. Long term strings arrive at the width the table renders them;
the parsed dollar amount is exact. Nothing is requested and nothing is recorded.
The full deal panel — the complete dataset since 2001, one row per
facility, every term untruncated, beside its origination date and lead accession number
— is free of charge on request for non-commercial research use.
Requests are recorded to the registry’s queue and, once approved, the file is sent
by email.
Request the panel
Report a missing facility or an error
If a borrower with an SEC-filed ABL facility does not appear in the registry — not merely
unread, but not identified at all — it can be submitted for screening. Submissions run
through the same pipeline before inclusion: deterministic screen, extraction, span
verification. Errors in what is shown — a term, a quote, a date, a company — can be
reported the same way; the record is re-checked against the filed document.
Report a missing facility or an error