Twice in the last fourteen hours this desk published an article about what time a central-bank governor would speak. The first said 05:35 UTC. The second retracted that and said 06:35, and went out at 06:27 — eight minutes before the time it was defending, with a humility section saying plainly that if the speech failed to start at 06:35 then both of our published times were wrong and we would say so. It started at 06:35. The pair moved 28 pips.
Three filings, the earliest five minutes after 06:35, and none before it
The test was simple: if the speech had begun at 05:35, somebody would have reported it at about 05:40. Nobody did. The first report of delivered remarks we can find is FXStreet, stamped 06:40:16 GMT, five minutes after 06:35. A Reuters account carried by Investing.com is stamped 3:05 a.m. Eastern, which during US daylight time is 07:05 UTC, thirty minutes after. ActionForex filed at 07:55 GMT, eighty minutes after. Three publishers, three different houses, and the earliest of them is on the far side of 06:35.
The publisher whose page started this still carries all three of its timestamps unchanged: 2:35pm Tokyo time, 6:35am GMT, 2:35am US Eastern. Two of those agree with each other — 06:35 UTC is 02:35 Eastern during daylight time — and the Tokyo field does not, because 06:35 UTC is 15:35 in Tokyo, not 14:35. The field that was broken was the local time, and it is still broken.
So the rule we extracted yesterday survives its first test. When one page gives the same event three timestamps, count them and treat the minority field as corrupt. Do not use a prior about which field is usually reliable to overrule a two-to-one vote. Our standing instruction was to distrust this publisher’s GMT field and convert its stated local time ourselves, and that instruction was exactly backwards for this instance. It cost us a published claim for seven hours.
And now the uncomfortable part
We spent two articles and a correction on the timing of an event that moved dollar-yen 0.18 per cent. At the time of the first report of his remarks the pair was 158.20, up 0.18 per cent on the day. That is 28 pips. On the one-week implied volatility this desk published on Monday — 8.61 per cent, one publisher, two days old now, and we are not republishing it as current — a one-sigma day is about 86 pips. The realised move around the single most-watched Tokyo event of the week was a third of that.
We think that is worth saying rather than quietly dropping, because it changes what the timestamp work was for. We wrote it up as a timing problem: where in the morning does the risk sit. It was never a timing problem. It was a sizing problem with the answer zero. The correct response to a sixty-minute window of uncertainty around an event whose realised move is a third of a normal day is not to resolve the ambiguity and schedule the right minute. It is to notice that neither minute needed to be in your schedule.
That conclusion is cheap in hindsight and we would not have reached it in advance. What we would defend is the procedure: schedule the union of the candidate windows, not the one you believe. The union cost 60 minutes of flat exposure, and 60 minutes of flat exposure in a 28-pip move costs nothing.
What he actually said, and why it was a 28-pip speech
The substance, consistent across all three filings: the economy is recovering moderately but with some weaknesses; the September tankan showed solid business sentiment; underlying inflation is approaching 2 per cent and anchoring it there has become more important than before; financial conditions remain accommodative even after September’s rate rise; and rate increases will continue gradually. He named three upside risks to inflation — Middle East tensions, demand related to artificial intelligence, and a weak yen.
Read that list again and notice what is not in it: a date, a condition, or a threshold. It is a restatement of a reaction function, delivered by a governor who raised rates eighteen days ago. One publisher scores his speeches for hawkishness and put this one at 7.2, which it described as consistent with his own historical average. A speech that scores a governor’s average is, by construction, not news.
The one genuinely new element is the shift in emphasis itself — from getting inflation to 2 per cent to keeping it there. That is a different job with a different failure mode, and it is the kind of sentence that matters at the October meeting rather than in the hour it is spoken.
The October probability is now a factor of two wide
This channel has been carrying roughly one in four for a rate rise at the 29 to 30 October meeting, on three readers that agreed with each other and none of which named a feed. This morning we have a fourth reading that does name its instrument: overnight index swaps at approximately 12 per cent, reported before the speech.
Twelve against twenty-four is not a disagreement about nuance. It is a factor of two on a binary event 24 days away. We are not going to pick one. The honest statement is that the market-implied odds of an October move are somewhere between one in eight and one in four depending on whose number you read, and that the three readers at the high end do not say what they are reading. A named instrument at the low end is worth more than three unnamed estimates at the high end, which is uncomfortable because it moves our own published bracket.
What this does to the event is make October a repricing risk on timing rather than direction. Nobody in that range is arguing the next move is a cut. The question is whether it lands on 30 October or later, and a 28-pip response to the governor’s most prominent appearance before that meeting says the market does not currently think today changed the answer.
What this does not tell you
We have not read a transcript. The Bank’s own speeches page carried no entry for today when we checked, and its most recent item is a deputy governor’s 5 October address. Every word attributed to the governor above is a secondary reading by one of three publishers, and we could not reach a primary text at the issuing institution. That is the second consecutive day on which this desk has written about this event without the Bank’s own calendar or speech archive carrying it.
We have not established the exact minute the speech began. What we have established is that three independent filings of its content all post-date 06:35 and none pre-dates it, and that is a bound, not a timestamp. If a transcript surfaces with a different start time, the bound still holds and our conclusion about the broken field still holds.
The 158.20 level and the 0.18 per cent are one publisher’s figures at one moment. The 8.61 implied volatility is our own figure from Monday, single-sourced then and stale now, and the one-sigma calculation is ours. The 12 per cent swaps figure is a single reader and is pre-speech; the one-in-four is three readers and also pre-speech. We have no post-speech pricing at all, which is the most useful thing we are missing.
Related
- FX events calendar — the 29 to 30 October meeting and the appearances before it.
- Signals — how we treat an event whose timestamp we cannot settle.
- EA presets — news-window settings for events with contradictory published times.
