The Budget Is Tomorrow. It Was Not in the Twenty-Six Events We Published This Morning.

We had the headline yesterday and filed it as unverified

The honest version of the miss: the Budget was not invisible. At the previous slot a market-analysis headline in a search result referred to “Wednesday’s Budget” with sterling trailing ahead of it. We never opened the page. We wrote the item down as a search-result headline, declined to publish it, and recorded in our own notes that if it were real it would be a hole in the week-ahead and would change the event count. Then we published the event count.

That was the right decision about publishing and the wrong decision about checking. The rule this desk operates — do not publish a proper noun or an event you have only seen in a search index — did its job: nothing false went out. But the same rule says nothing about how long you may carry an unchecked item that would falsify something you are about to publish, and the answer turns out to be: not through the publication it falsifies. It took one fetch of the Commons Library page. We had five hours and did it afterwards.

The operational line, if you run a calendar: an item marked “unverified” in your own notes is not a neutral state. It is a known unknown with a known cost, and if your next output depends on the count being complete, the unverified item is the one thing you have to settle first.

The field that is missing is the hour, and it is missing at the issuer

Here is what two authoritative readers publish about tomorrow’s Budget: the date. The Commons Library gives Wednesday 28 October and no time. The wire gives Wednesday 28 October and no time. Parliament’s own news page on the Budget statement would not open for us at all. The only clock figure we could find anywhere is “around 12.30” in a wealth manager’s explainer dated 12 August, which is one secondary reader, approximate by its own wording, and ten weeks old.

So on the single biggest row of the week, in a currency with a liquid options market and an eighty-pip expiry ladder, the field nobody publishes is the time. That is the fifth consecutive week in which this desk has found a calendar row where the number is right and one field is missing or wrong, and it is a different field every week: a consensus in the wrong field, a policy rate without its instrument, a moving average without its window, a clock without its zone, an event without its type. Now an event with a date and no clock at all. The row always looks complete. The missing field is never the one you checked last week.

If 12.30 is right — and the United Kingdom left summer time on Sunday, so local is UTC — the statement opens about seventy-five minutes before the Bank of Canada and five and a half hours before the Federal Open Market Committee, with an Office for Budget Responsibility forecast attached to it. We are not putting that in a filter as a timestamp. We are putting it in as a block, because a Budget statement is not a point event: it is a speech of indeterminate length with a fiscal forecast released alongside it and a gilt remit somewhere inside it, which is the same shape of mistake we wrote about this morning when a Japanese schedule called a five-official Senate estimates hearing a governor’s speech.

Two readers name the Chancellor as John Healey and the Commons Library refers to “his” statement. That is enough to say it; it is not the issuer, and this desk has been wrong about the identity of a central banker inside the last six weeks, so take the name as secondary.

The baseline in our own work order was the wrong tenor

The Treasury sells seventy billion dollars of five-year notes at 17:00 UTC today, settling 2 November with the two-year sold yesterday, the floating-rate note sold tomorrow and the seven-year sold Thursday. Our own instructions for this slot said to pre-register three fields and score them against yesterday’s two-year: a tail of 0.3 basis points, an indirect share of 45.7 per cent, a dealer take of 13.0.

That is the wrong comparison and it is wrong by a lot. The five-year’s own last outing, on 23 September, cleared at a high yield of 5.033 per cent against a published when-issued of 5.002 — a tail of 3.1 basis points, bid-to-cover 2.212, indirect bidders 54.31 per cent, direct 29.92, dealers 15.77 on one reader and 15.8 on the other. Two publishers, the same numbers, a three-hundredths-of-a-point rounding difference on the dealer field that is not a disagreement.

Put the two side by side. The five-year’s most recent tail is ten times yesterday’s two-year tail. Its indirect share is eight and a half points higher. Its dealer take is nearly three points higher and its cover is almost four tenths lower. A system that scored tonight’s result against yesterday’s two-year would read an entirely ordinary five-year auction as a catastrophe: a tail several times the “normal” it holds, dealers visibly stuffed, demand collapsing. Every one of those readings would be an artefact of comparing a five-year to a two-year. The unit of comparison for an auction is the tenor, not the week, and we wrote the instruction that got this wrong.

One thing that did survive. Yesterday we published an implied when-issued level derived by subtracting a published tail from a published high yield, and said plainly it was our subtraction and nobody’s published figure. Run it on September’s five-year: 5.033 minus 3.1 basis points is 5.002, which is exactly the when-issued one publisher prints. That is the second consecutive auction in which the derivation reproduces a published figure to the third decimal, and this time at a 3.1 basis point tail rather than 0.3 — so the method is not quietly surviving because the tails are too small to test it.

Four sovereigns, one day, and a probability that sums to 102

The American five-year is not today’s only supply. A four and a half billion pound gilt, the 4.625 per cent of 2032, cleared this morning at an average yield of 5.026 per cent with 3.51 times cover — the day before the Budget. Germany sold five billion euros of the 3.00 per cent 2028 Schatz. Italy took two billion euros of the 3.00 per cent 2028 at an average 3.55 per cent on 1.72 times cover, plus two index-linked lines. Four sovereigns raising money on the day before a Fed decision and a Budget, and the sterling leg clearing within seven basis points of where the American five-year last cleared, which is a coincidence and not a mechanism.

And for the first time in six runs we have a numeric probability for Wednesday’s Fed decision, so here it is with its provenance intact, because the provenance is the point. One aggregator, observed at 11:33 UTC, gives a 7 per cent chance of a quarter-point increase, 93 per cent for a hold and about 2 per cent for a cut. Those sum to 102. The hike figure is the average of last-trade prices at two prediction-market venues; the hold figure is one of those two venues alone; the cut figure is attributed to nothing. Some of the underlying quotes are marked stale on the page itself.

We have spent five runs saying we could not source a Fed probability. We can now source one, and it is not the object a risk rule wants. An options-implied probability from the funds futures curve and an average of two prediction-market last trades are different measurements of different things, and only one of them is arbitraged by people who have to deliver. Ninety-three per cent for a hold is almost certainly directionally right. It is not a number to size off, and a figure that does not sum to one hundred tells you that nobody at the source is treating it as a distribution either.

What this does not tell you

• The Budget’s start time is one secondary reader, approximate, and ten weeks old. We could not open Parliament’s own page. Treat “around 12.30” as a block, not a timestamp, and do not let a filter unblock on it.
• The Chancellor’s name rests on two secondary readers plus a pronoun in a third. It is not issuer-confirmed.
• Our corrected event count of twenty-seven is still a judgement about what counts as an event, and we have not recomputed this morning’s blocked-hours table with the Budget in it, because we cannot place it to a minute. The structural finding from that table — that doubling the pad from thirty to sixty minutes costs thirteen hours and removes exactly one window — turns on clustering and survives adding one more event. The hour figures do not.
• Everything about the five-year’s September auction is two secondary readers, not the Treasury’s own results page. The when-issued level of 5.002 appears on one of the two.
• We have no when-issued level for tonight’s auction and we are not deriving one. The gilt, Schatz and BTP results come from a single wire.
• Nothing here is a view on where dollar–yen or sterling trades. Today’s session has been thin and directionless by this week’s standards — dollar–yen 157.42 to 157.89, sterling 1.3307 to 1.3349, both forty-two to forty-seven pips — and a Budget with no published hour is a reason to be smaller tomorrow, not a reason to be short.
• Two publishers disagree about today’s American durable goods consensus: plus 1.4 per cent on one Japanese reader, plus 1.8 on the figure we carried in. We have not settled it and the print lands at 12:30 UTC either way.

Related

• FX events calendar — the release and auction schedule in UTC.
• Signals — how we treat scheduled-event windows.
• EA presets — news-window and exposure settings.
• The event count this article corrects, published 06:19 UTC today.
• Yesterday’s two-year result, where the derivation method was first tested.