The insolvency rate has a denominator problem, and ECCTA is about to make it bigger

In UK lender and large trade creditor circles, the monthly insolvency release comes up in a very particular way. Someone has read the headline, formed a view, and wants to know whether it changes anything. Almost nobody has the release wired into a model. It sits in the “context we’re aware of” pile rather than the “input we act on” pile.

July’s numbers, published recently, are a good illustration of why that gap costs you something.

What the release actually says

There were 1,931 company insolvencies in England and Wales in July 2026. That is 5% higher than June and 5% lower than July 2025. The 12-month rolling rate fell to 50.3 per 10,000 companies, or one in 199, down from 52.5 a year earlier. Read at that level, the story is “gently improving.”

Now look at the composition. Creditors’ voluntary liquidations rose 9% on the month to 1,497 and accounted for 78% of all cases. Administrations fell 33% on the month and 19% on the year, to 124.

Those two processes describe different companies. A CVL is usually a small company whose directors have concluded there is nothing left to rescue. An administration usually involves a business large enough, or with enough residual value, that somebody thinks a rescue or a going-concern sale is worth attempting. So the mix in July says small-company failure held up while the larger-company rescue route thinned out. That is a different sentence from “insolvencies fell 5% year on year,” and it is the sentence that matters if your exposure is concentrated in mid-market names.

The bit I’d like to flag

The rolling rate is the figure people reach for when they want to sound rigorous. It is also the least stable number in the release.

It is a ratio. The numerator is insolvencies. The denominator is the number of companies on the effective register. That denominator has been growing for fifteen years, and it is the main reason the rate looks benign while volumes sit close to their post-2009 highs. The Insolvency Service says so directly: the 2008-09 peak rate was 113.1 per 10,000, more than double today’s, and the register has more than doubled in size since then.

Look at what the denominator is actually made of. At the end of June 2026 there were 5,516,377 companies on the register. In the April to June quarter alone, Companies House recorded 192,287 incorporations and 156,515 strike-offs and dissolutions, a net gain of 37,339, or 0.7%. Around 150,000 companies leave the register every quarter and slightly more join it. This is not a stable population being measured. It is a high-turnover flow with a long tail of entities that have never traded, never filed anything meaningful, and were never going to appear in an insolvency statistic in the first place.

Then there is ECCTA. Identity verification became a legal requirement on 18 November 2025, with a twelve-month transition period that ends this November. After it closes, directors and PSCs who have not verified face financial penalties and cannot make filings for their company. It is a deliberate integrity measure and, in my view, a good one. It is also, mechanically, a denominator event. If a meaningful share of the register does not verify, the register shrinks for reasons entirely unconnected to credit conditions, and the published insolvency rate rises without one additional company failing.

I do not know how large that effect will be. Nobody does yet. But if your credit strategy contains a threshold phrased as “when the national insolvency rate exceeds X,” you should know that X is measured against a base that a piece of company law is about to move, and that the move will look like deterioration when it is nothing of the kind.

Four levels of using this data effectively

Most organisations I speak to are at level one or two. The work of getting to three and four is not glamorous, but it is where the release stops being commentary and starts being an input.

Level 1: Read the headline

You know the direction of travel. You cannot act on it, because a national aggregate says nothing about your book. This is where most people stop, and it is fine as long as nobody mistakes it for analysis.

Level 2: Benchmark against the sector table

The 12-month sector figures to July 2026 are construction at 3,841 cases (17% of the total), wholesale and retail at 3,422 (15%), accommodation and food service at 3,221 (14%), and administrative and support services at 2,212 (10%). If your portfolio is 40% construction, the national average was never your base rate. Re-weighting the national figures to your own sector mix takes an afternoon and immediately tells you whether your loss experience is better or worse than the environment you are actually lending into.

Level 3: Build a denominator you trust

Strip the entities from your comparison base that could never have been credit decisions in the first place. Dormant companies, non-filers, shells incorporated and dissolved inside eighteen months. What you want is a rate expressed over trading companies of a comparable size and sector to the ones you underwrite. That number will be materially higher than 50.3 per 10,000, and it will be the one that reconciles to your own experience. It also insulates you from the ECCTA effect, because you were never counting the entities most likely to fall off the register.

Level 4: Treat the release as ground truth, not as signal

By the time a company appears in these statistics, it has already failed. The release is excellent for backtesting and calibration. Use it to check whether your early-warning system was actually early. It is useless as a warning in itself. The warning has to come at entity level and ahead of the event, from financial-health scoring like the Company Watch H-Score® and from court-filed distress signals such as CCJs and winding-up petitions, run continuously across the whole book rather than pulled one company at a time. If that data only arrives when someone requests a report, you have built a system that confirms what the Insolvency Service will tell you for free in six months.

If I were leading this

I would spend the next quarter on level three, and I would treat it as a data engineering problem rather than a credit one, because that is what it is. Build the re-weighted base rate, get it into whatever your analysts actually query, and stop bench-marking a specialist portfolio against a national average built on a register that carries a large non-trading tail and is about to be reshaped by identity verification.

The awkward part is usually not the analysis. It is that the entity-level data sits in a vendor portal, the exposure data sits in a warehouse, and joining them is a manual export. That is the activation gap, and it is the reason so many teams read the monthly release as commentary. Data delivered through an API, into BigQuery, or through MCP so an analyst can ask a question in plain language, is what turns it into something you can run against 40,000 counte-rparties on a schedule.

November will be an interesting month for anyone who has not done this work. If the register contracts and the published rate ticks up, I would rather be the team that can explain exactly why than the one revising its appetite on the strength of an artefact.