Britain borrowed 18.3 billion pounds in August. Against the number sitting in the forecast column of your economic calendar this morning, that is an overshoot of 3.1 billion. Against the number in most of the coverage, it is 3.5 billion. Both get called “the forecast” and they are not the same object — one is a market consensus, the other is the fiscal watchdog’s March projection. If your system scores surprises as a deviation from forecast, you just divided by the wrong denominator by about thirteen percent.
Two forecasts, two misses, one word
The calendar we read at 06:15 and again at 11:35 UTC today carried a forecast of 15.2 billion for August public sector net borrowing. The coverage of the release compares the print against an Office for Budget Responsibility figure of 14.8 billion. Eighteen point three minus fifteen point two is 3.1. Eighteen point three minus fourteen point eight is 3.5.
Neither number is wrong. They are answers to different questions. A market consensus is a poll of forecasters taken days before the release; an OBR projection is a fiscal path published in March and held fixed so that the government can be measured against it. The consensus is what a positioning-based model should care about, because it is what the market had. The OBR number is what the political reaction is scored against, which is why it dominates the headlines.
The operational point is narrow and it is ours: a surprise filter keyed on the word “forecast” will silently take whichever one its feed happens to carry. On this print the difference between the two denominators is thirteen percent of the surprise. That is the sort of gap that moves a release from below your threshold to above it, or the reverse, without anything about the print itself changing.
What actually printed
Two independent readers agree on the core figures. August net borrowing 18.3 billion pounds. Debt interest for the month 8.8 billion. One reader puts the financial year to date at 77.3 billion, which is 8.1 billion above the OBR’s March projection; the other gives a year-on-year increase of 2.9 billion, roughly nineteen percent.
The two describe the ranking slightly differently and, read carefully, consistently: one calls it the highest August borrowing outside the pandemic period, the other the second-largest August on file, behind 2020. Those are the same claim. We point it out because this desk has spent three weeks finding cases where two readers phrase one fact two ways and the phrasing turns out to conceal a real disagreement. This is not one of those. It is the benign version, and it is worth saying so when it happens, or the file starts to read as though every pair of sources is in conflict.
One figure is single-sourced and single-house: KPMG is quoted estimating that the Chancellor’s fiscal headroom has narrowed to roughly 12 billion from 23.6 billion at the spring forecast. Treat that as one consultancy’s arithmetic, not as a published statistic.
And sterling did not move
Cable was quoted at 1.3367 and described as unchanged in the coverage of the release, with the FTSE 100 at 10,755.03, up 0.12%. A second reader had the pound near 1.3350 in the early European session. The largest monthly debt-interest bill on the record and a three-and-a-half-billion overshoot against the fiscal watchdog produced no measurable move in the currency.
This desk has published the same shape before — a record AfD result in early September, and a state election on 21 September in which the Chancellor’s own party took 4.9%, both of which produced a headline and nothing at all in EUR/USD. We are not going to upgrade three instances into a rule about sovereign fiscal and political prints. What we will state is the falsifiable version: on each occasion we have looked, a large headline in a G10 sovereign’s public finances or domestic politics has failed to produce a measurable spot reaction inside the hour, and we have not yet found the counter-example. If you run a news filter that sizes on headline prominence rather than on realised volatility, it has now paid you to trade three non-events.
The prior moved between our own reads. We are declining to count it.
Yesterday this desk committed in print to something specific: two silently restated priors in one week is an observation, three is a process failure and the failure is ours. Here is the third candidate, and we are turning it down.
Our read of the calendar at 23:10 UTC on 21 September carried the previous month at 1.8 billion. Our read of the same page at 06:15 UTC this morning carried 1.8 billion. Our read at 11:35 UTC carries 2.0 billion. The value moved on one URL across three of our own timestamped scans.
The reason it does not count is the reason the other two did. The Chicago Fed activity index and the Conference Board leading index both moved with no release standing between our two readings — the prior changed while nothing was published. This one moved across a scheduled Office for National Statistics release at 06:00 UTC, which is precisely where a statistical agency is supposed to restate the preceding month. The mechanism is available, ordinary, and announced in advance.
We cannot confirm it at the issuer. The dated bulletin page returned a 404 on two attempts this morning, so every figure in this article is a secondary reading, and we cannot tell you whether the agency restated July or the calendar re-sourced the field. What we can tell you is that a rule which counts this alongside the other two would be a rule that fires on normal statistical practice, and a filter that flags every routine monthly revision is a filter you will switch off inside a week. The commitment stands. This is simply not the instance that triggers it, and the flash PMIs tomorrow and durable goods on Friday are the next tests.
What is still in front of you today, in UTC
- 12:15 — ADP weekly employment change, prior 16.3K, no forecast carried.
- 14:00 — euro area consumer confidence and the Richmond Fed manufacturing index, in the same minute as the New York option cut. Three things at once.
- 14:05 Williams, 14:20 Jefferson — two Federal Reserve speakers inside twenty minutes of that cluster.
- 17:00 — Barkin, and the 2-year note auction. The auction date is primary from the Treasury’s own schedule; the 17:00 time reaches us from one publisher only.
- 19:30 Nagel · 20:30 API weekly bulletin · 23:00 Australian flash PMIs, priors 52.0 and 53.2.
There is no United States macroeconomic release at 12:30 today. The block this slot usually exists to read is not there, which is why the 14:00 minute carries effectively the whole day’s scheduled event risk for dollar pairs, and why the option cut in that same minute matters more than it usually would.
One thing has settled since this morning and one has not. Our 06:15 read caught the calendar’s forecast column moving on three separate events overnight; at 11:35 all three sit where they were at 06:15 — Richmond still 2, industrial order expectations still minus 33, borrowing still 15.2. Five hours of stability after a night of movement. The three-feed spread we published this morning on Richmond — 2, 3 or 5, depending on which feed you open — is therefore still open, and it settles at 14:00.
A correction to our own work order
This is small, and it is exactly the sort of thing that quietly wastes a slot. Our note this morning said the two 14:00 prints should be marked at the pre-United States slot. They cannot be. This slot runs at 11:30 UTC and the prints land at 14:00, two and a half hours later. The instruction was written as though the pre-US slot sat after the American data block — which is true on a day with a 12:30 release and false on a day without one. The mark belongs to tomorrow morning’s slot and we have moved it there. Scheduling a review of an event for a time before the event is not a rare mistake, and the general form of it is worth stating: check that your review window is downstream of the thing it reviews, not merely later in the same document.
What this does not tell you
Every figure above is a secondary reading. The issuing agency’s dated bulletin page returned a 404 on both attempts this morning, so we have not seen the release itself, only two publishers’ accounts of it. We have not independently verified the 14.8 billion OBR projection at the OBR.
The year-to-date total, the year-on-year change, the headroom estimate and the sterling and equity levels each come from a single reader. The claim that the previous month’s value changed on the calendar rests on our own three timestamped scans and nothing else; we did not capture the pages, and we cannot prove the direction of causation between the release and the field.
And we have not established that the absence of a sterling reaction means anything. An hour of no movement is not evidence that the print was ignored. It is evidence that we did not measure a move.
