How the pricing benchmark is built
A firm that sells debt advice publishing a benchmark on the cost of debt has an obvious conflict. The only useful answer to it is to make the whole thing checkable, so this page states every gate, prints the count each one removed, and names what is excluded. If you disagree with a decision here you can see exactly which one it is and what it cost.
The source is the public record, and only that.
Two Companies House surfaces, both free and both public. The register of charges says which companies have granted security, when, and to whom, which is how a company enters this universe at all: a company with no registered charge is not borrowing against security and has no place in a benchmark of secured corporate debt. The filed accounts carry the financial disclosures used in the benchmark.
Nothing here comes from a survey, a lender, a broker panel, or any private source. No borrower was contacted and no client information is used. Every figure is re-derivable by anyone with the same two public sources and the method below.
Source: Companies House filed accounts and register of charges. Companies House bulk register, July 2026 snapshot; accounts read September 2026. Input file hash bbe2dbdc1f54044b, 41,250 rows, generated 6 September 2026. A re-run producing different figures against the same hash is a code change, not a data change.
Series one: the realised cost of debt.
For each company, the finance cost reported for the period, divided by the average of its interest-bearing debt at the opening and closing balance sheet dates. Both figures relate to the same reporting period and entity.
Why the average and not the closing balance
A year’s interest is paid on the debt carried through the year, not on the balance standing at the end of it. A company that repaid half its debt in month eleven shows a small closing balance against a full year of interest, and dividing one by the other produces a rate that is not a rate. Using the average of opening and closing fixes the common case. Where the balance moved by more than 60% the average is no longer a fair representation of the year, and the company is excluded rather than smoothed.
What is in the numerator
Where a company tags interest on bank loans and overdrafts specifically, that is used. Where it tags only the total finance cost line, that is used instead and identifiable non-debt interest is netted off: finance lease interest, interest paid to group undertakings, interest on directors’ and shareholder loans, and net interest on defined benefit pension obligations. What remains still includes arrangement fees amortised over the facility term and any borrowing struck in a higher-rate year, which is why this series reads above a quoted margin plus the reference rate. It is a realised all-in cost of the stock, and it is labelled that way everywhere it appears.
Placing a rate against its own year
A cost of debt from a year ending in 2023 and one from a year ending in 2026 were struck against different Bank Rates. Where the spread over Bank Rate is published, each observation is measured against the average Bank Rate over its own accounting period, integrated across the Monetary Policy Committee’s actual step dates, rather than against today’s 3.75%. Without that, the series would measure the rate cycle instead of the borrower.
Series two: the margin a borrowings note discloses.
Some companies state their price in words. The text of every readable filing is scanned for rate-shaped statements, and each hit is classified from the two hundred characters either side of it before it is allowed anywhere near a statistic. The classifier is deliberately strict, and an unclassifiable hit is discarded rather than guessed at.
What the classifier throws away, and why
Filed accounts are full of percentages that look exactly like a margin and are not one. A naive scan of the same documents produces a benchmark that is mostly gross profit margins and directors’ loan interest. Across the documents read so far the classifier rejected 4,014 rate-shaped statements for these reasons:
- A rate with no loan, facility, borrowing or lender anywhere near it. Usually a tax rate, a discount rate or a percentage of something else entirely.2,335
- Interest on a director's loan, a group balance or a shareholder loan. Real interest, but not the price of third-party debt, and typically set for tax reasons rather than by a credit decision.809
- Corporation tax rates, deferred tax rates, discount rates and weighted average costs of capital.296
- Ownership percentages, voting rights and shareholdings.231
- The word margin doing different work: gross margin, profit margin, operating margin. The single largest source of false positives, and the reason a naive text scan of accounts produces nonsense.203
- Net interest on a defined benefit obligation, which is an actuarial figure.92
- rate_out_of_band33
- A margin over a named reference rate outside 0.25% to 9%, which is not a senior margin in this market.15
What survives has to have a loan, facility, borrowing, debenture, mortgage, overdraft or lender in the same sentence, must not mention a director, a group undertaking or a related party, and must fall inside a plausible band. A margin over a named floating reference is kept as a margin; an absolute rate with no reference named is kept separately as an all-in rate and never blended into the margin series.
Reference rates are recorded as filed. A margin quoted over base rate and one quoted over SONIA are not interchangeable and are never pooled: they are published as separate rows, because the same numeral is a different price depending on what it sits on top of.
Coverage of the benchmark.
Inclusion depends on the availability and comparability of financial disclosures. The benchmark is a sample of corporate borrowing, not a census.
- Companies reviewed41,250
- Published realised cost observations631
- Companies contributing a disclosed rate1,798
The rules for publishing a figure.
- Minimum cell size 30. No median, quartile or range is published off fewer than 30 observations. A cut below the floor appears as a named absence with its count, never as a statistic.
- Trimmed at the 2.5th and 97.5th percentiles. The extremes are pulled to those bounds before percentiles are taken, so a single distressed borrower cannot move a headline. The trim is applied to every published cell and to none of the counts.
- Rates outside 0.5% to 25% are excluded. Below the floor the figure is a fee artefact or a mis-tag; above the ceiling it is not the price of ordinary corporate debt.
- Average debt below £250,000 is excluded. On a small balance, arrangement fees and rounding dominate the computed rate.
- Medians, not means. Percentiles are reported throughout. A mean cost of debt is dragged upward by the tail of distressed borrowers, and the question this benchmark answers is what a typical company pays.
- Both series stay separate. A disclosed margin and a realised cost are different measures of different things and are never pooled into a single headline.
What this is not.
It is not a census. These figures are counted from filed accounts, not surveyed. They are floors on a prioritised sample of UK companies that carry a registered charge, not a census of UK business borrowing. Inclusion depends on the availability and comparability of financial disclosures. Small and micro-entity companies and the largest groups are under-represented.
Both ends of the market are under-represented. Micro and small companies frequently file accounts that tag neither a finance cost nor a debt balance. Large audited groups often file scanned documents carrying no tags at all, and 9,796 of the companies read here fell into that category. The readable middle of the register is therefore over-weighted relative to its share of borrowing, and every figure should be read as what the readable part shows.
It is a stock, not a flow. The realised cost series measures what companies are paying on debt they already have, most of which was priced in an earlier year. It is not the rate available to a borrower today, and it will lag a fall in Bank Rate by roughly the average remaining life of the outstanding facilities.
It cannot price a specific credit. Two companies at the same rate can be differently levered, secured and covenanted. The distribution says where a company stands among its peers. It does not say what that company should be paying, and no figure here should be used as a target, a valuation input or a negotiating anchor without the credit behind it.
It is published by an interested party. Solon advises borrowers on raising and refinancing debt, so a wide published dispersion is commercially convenient for us. That is exactly why the gates, the counts and the input hash are on this page: the way to check the claim is to re-run it, and the way to disagree is to name the gate.
Using the figures.
The benchmark is free to reuse, including commercially, with attribution to Solon Corporate Finance and a link to the benchmark page. Quote the as-at date with the figure: a price with no date is not a fact about anything. The full set of published cells, every cut and every percentile, is available by request:request the dataset.
If you are writing about this and want the cut behind a figure, or a breakdown that is not published because it sits below the cell floor, ask and we will send it with the counts attached. Corrections are welcome and are made in public: if a figure here is wrong, the method above is enough to show us where.
Comparator figures on the benchmark page are the Bank of England’s, not ours: the effective interest rate on new SME lending, 6.16% at April 2026, and a spread of 234 basis points over Bank Rate, both published in Money and Credit. They are reproduced for comparison and are not part of this dataset.