A statistic that survives a subeditor has usually passed three tests nobody writes down. Here they are, along with the sources that keep passing them.
What kills a statistic before publication?
Not being wrong. Being untraceable.
A figure that cannot be followed back to a named document with a date on it will get cut, not because anyone has disproved it but because nobody can defend it if a reader complains. That is the practical filter operating in most newsrooms, and it is stricter than the accuracy filter.
The second killer is a mismatch between the number and the sentence it is being used to support. A figure about account holders being used to describe active traders will not survive anyone who reads carefully, and the correction is embarrassing in a way that getting a decimal place wrong is not.
Which sources clear the bar reliably?
The regulator, first. FCA market studies, policy statements and the Financial Lives survey are all published, dated, methodologically documented and free. Nobody has ever been criticised for citing them. They are slow, they lag the market, and they answer the questions the regulator was interested in rather than the ones a reporter has, but they are unimpeachable.
Why company filings are underexploited
Then company filings. Listed brokers publish annual reports containing client counts, revenue per client, onboarding numbers and geographic splits, and those are audited numbers that the company has a legal interest in getting right. They are also underexploited, because reading them is dull and most people would rather quote a press release about them.
What about official statistics beyond the regulator?
HMRC for anything tax-shaped, and it is more useful than people expect.
ISA subscription and market-value data, capital gains statistics and compliance activity are all published on a schedule. The April 2026 release put the adult ISA count at 22.6 million against a market of around GBP 872 billion, which is the kind of figure that anchors a piece without anyone being able to argue with it.
Who publishes the market structure numbers?
The Bank of England and the Bank for International Settlements cover market structure. The BIS Triennial Survey put UK average daily foreign exchange turnover at USD 4.745 trillion in April 2025, 37.8% of the global market. That is a number with a methodology paper attached, which is exactly what you want when an editor asks where it came from.
| Source type |
Best for |
Lag |
Watch out for |
| FCA publications |
participation, harm, conduct |
12-24 months |
answers their question, not yours |
| Company annual reports |
client counts, revenue |
3-6 months |
definitions of ‘active’ vary |
| HMRC releases |
ISAs, CGT, compliance |
6-12 months |
accounts are not people |
| BIS and Bank of England |
market structure, FX |
triennial or quarterly |
wholesale, not retail |
| Academic research |
behaviour, outcomes |
years |
often not UK data |
| Industry surveys |
trader counts, sentiment |
months |
methodology often unpublished |
| Press releases |
almost nothing |
immediate |
the number exists to be quoted |
Seven source types ranked by how much defending they need. The lag is inversely related to the reliability, which is the central frustration of the beat.
Why do the same three statistics keep circulating?
Because citation is cheaper than sourcing. A figure that has appeared in a national title acquires a kind of laundered authority, and the next writer cites the article rather than the underlying document. Do that four times and the original context, definition and date have all fallen away while the number keeps travelling.
The result is a small number of statistics doing an enormous amount of work, long after they were current. Anyone who has traced a widely quoted figure back to source has usually found it is older than expected, narrower than implied, or measuring something adjacent to what it is being used for.
What is the single most useful habit?
Clicking through. Not to the article that quoted the number, to the document the article quoted.
It takes a couple of minutes, it catches most errors, and it occasionally produces a better story than the one you started with because the original document contains context the secondary coverage stripped out.
The corollary is never citing anything you have not opened. That rule sounds obvious and is broken constantly, including by people who would be genuinely embarrassed to admit it. It is the difference between a piece that survives scrutiny and one that survives until somebody checks.
Where do research sites fit?
Usefully, when they are doing primary work, and as a hazard when they are not. A site that has compiled figures from filings and states which filing each came from has saved you genuine labour. A site presenting an aggregate with no underlying sources named has added a layer between you and the evidence.
The test is whether the compilation is checkable, and it is a test you can put The Investors Centre to directly: its UK trading statistics pages are built from HMRC, FCA and Ombudsman Service releases alongside the audited annual reports of the listed brokers, so any given figure either traces back to a named document or it does not.
Where it does not, treat it as a lead rather than a citation.
What about proprietary indices?
Handle carefully. A site aggregating the published loss disclosures of fourteen brokers into a mean has done real work, and the resulting number is genuinely informative. It is also that site’s own calculation, not an official statistic, and it should be attributed that way rather than laundered into something that sounds regulatory.
The distinction matters more than it seems. Analysis by a research site, April 2026 is an honest attribution. FCA data showing is not, if the FCA did not publish the aggregate. The underlying disclosures being regulatory does not make the aggregation regulatory, and conflating the two is one of the commonest sourcing errors in this field.
What is the uncomfortable part of all this?
That research produced partly in order to be cited will tend to measure the things that get cited. A site that benefits from coverage has an incentive to generate quotable aggregates rather than dull, heavily caveated ones, and that incentive shapes what gets counted even when everyone involved is honest.
This applies to every organisation publishing data for public consumption, including regulators with policy positions and trade bodies with members.
It is not a reason to distrust compiled research. It is a reason to prefer the ones that publish their definitions, their dates and their limitations, and to be suspicious of any dataset in which every finding happens to be interesting.
How should you handle a number you cannot verify?
Attribute it precisely and let the reader judge, or leave it out. Those are the only two defensible options. Estimates put it at is a construction that survives because it names nobody, and it should not survive, because a reader has no way to weigh an estimate whose author is unnamed.
If a figure is genuinely important and genuinely unverifiable, that is itself worth reporting. Nobody publishes a reliable count of X is a legitimate finding and occasionally a more interesting one than whatever number you were chasing. Write that sentence instead of the number, and you have told the reader something true, which is the whole job.
