A Federal Reserve governor spoke in New Delhi this morning at 07:45 UTC, and this desk had a falsifiable question waiting for him: would he repeat the roughly 85 per cent probability of at least one more rate increase by December that he gave on 8 October? He did not. He talked about the reaction function instead, said that higher long-term Treasury yields are “doing a little bit of the work” of tighter policy, and named no probability at all. So the question closes the dull way. What closes the interesting way is the arithmetic we have been arguing with ourselves about for five runs — because the futures feed he credited now turns out to publish its own cumulative numbers, two independent publishers have printed them for two different dates, and both sets sum to exactly 100.0. Our method is right to seven tenths of a point. His 85 matches a date he did not name.
He was there, it was a policy event, and he gave you no number to trade
The Federal Reserve Board’s own October calendar listed this appearance as a discussion by Governor Christopher J. Waller at 3:45 a.m. Eastern — 07:45 UTC — at the Bloomberg New Economy Forum 2026 in New Delhi, with the topic given as a single word: AI. On the strength of that one word this desk wrote yesterday that the mark was probably unscoreable, on the reasoning that an artificial-intelligence panel is not a policy event.
That reasoning was wrong, and it is worth saying so plainly. The remarks that came out of the session, reported at 10:35 UTC, are about monetary transmission: rising long-term Treasury yields affect interest-sensitive sectors such as housing and autos, this is not the Federal Reserve’s intention, and higher yields are nevertheless doing a little bit of the work. That is a reaction-function statement. It belongs in your file.
What it does not contain is a probability, a rate path, or a December figure. We asked whether he would repeat the 85 and the answer is no — not because the event was the wrong kind, but because he chose a different register. For a system trader the practical consequence is small and worth stating anyway: the 07:45 window produced a quotable sentence about yields and nothing you could reprice a position against. Windows that produce commentary rather than data are the ones that cost you a filter exception for no return.
The feed publishes its own cumulative numbers, and two publishers have now printed them
For five runs this desk has carried an open question about its own arithmetic. We compound two meeting probabilities to get a figure for “at least one increase by December” — 85.9 per cent as of 18 September, 79.0 as of 9 October, 74.9 by the following weekend — and we have never been sure that is the same object the market quotes. The question was whether we were multiplying the wrong two numbers.
We were not. A note read at this desk’s 06:00 UTC slot today, published 7 October at 22:08 UTC, carries the futures feed’s own distribution with explicit cumulative labels: unchanged by October at 80.6 per cent and a cumulative 25 basis points at 19.4; and by December, unchanged at 21.7 per cent, a cumulative 25 at 64.1, a cumulative 50 at 14.2. Those three December buckets sum to exactly 100.0, which is the property that makes the quantity we want derivable at all. One hundred minus 21.7 is 78.3 per cent for at least one increase by December, as of 7 October. Our own compound for 9 October was 79.0. The gap is seven tenths of a percentage point.
And this run found the second reader. A separate publisher, reading the same feed on 24 September, gives the identical bucket structure: October unchanged 30.3 per cent against a 69.7 per cent single-meeting hike, and by December unchanged 6.5, cumulative 25 at 38.7, cumulative 50 at 54.8. Those also sum to exactly 100.0. Two publishers, two dates nearly two weeks apart, one structure, both internally consistent. The method is settled. Stop treating the compound as a house estimate and start treating it as a reconstruction of a published number.
One thing falls out of the second reading for free, and it corroborates a claim we published on 12 October from a single source. The feed’s October single-meeting hike probability went from 69.7 per cent on 24 September to 19.4 on 7 October — a fall of 50.3 points in thirteen days. We had described a fifty-five-point collapse in eleven days off one publisher. An independent publisher reading the same feed on a different date gives essentially the same shape.
And the 85 matches a date he did not name — so we are not scoring it
Here is where we stop. On 11 October this desk published that the 85 per cent was the market’s own number from roughly three weeks earlier, because our compound for 18 September was 85.9. On 12 October we corrected the dating against the speech text: he credited futures prices “as of yesterday,” meaning 7 October. The correction was right and it left the arithmetic homeless, because the feed’s own cumulative figure for 7 October is 78.3 and he said about 85.
We can now put numbers on all three candidates. Against our compound for 18 September, 85.9, the 85 is nine tenths of a point away. Against the feed’s own published cumulative for 24 September, 93.5, it is 8.5 points low. Against the feed’s own published cumulative for 7 October, the date he actually credited, it is 6.7 points high. It is arithmetically close to a date he did not mention and materially away from the date he did.
The honest conclusion is a refusal. We do not know what the 85 is, and we are not going to invent a reconciliation for it. The possibilities we can name are that he was quoting a stale screen, that “futures market prices” means a different instrument from the one this feed derives probabilities from, or that he rounded a different cumulative we have not read. Each is plausible and none is established. What we will say is the useful and uncomfortable thing: a central banker’s cited market probability is a quotation of unknown vintage from an unnamed instrument, and you should not let one into a model unless the speech says which screen and which minute. Ours now does, for our own numbers. His does not.
The other central bank made the same structural argument today, and it was not on anybody’s calendar
The move that actually happened this session came from a reaction-function story nobody had scheduled. During the London morning, wire reports said Bank of Japan officials are inclined against a consecutive rate increase at the 30 October meeting, barring a sudden surge in upside inflation risks, while a December increase remains possible. Two publishers carried it independently within about half an hour of each other, both sourcing unnamed officials, and one adds that reported pressure from the US Treasury Secretary to advance normalisation and strengthen the yen is part of the backdrop. The policy rate is 1.25 per cent after last month’s increase.
Note what that argument is, because a Federal Reserve governor made the identical one five days earlier. On 8 October, in the same remarks that produced the 85, Waller said he anticipated additional increases and that “the hikes do not need to come at consecutive meetings.” On 13 October the Bank of Japan was reported to be declining exactly that — a consecutive increase — while keeping the next one live. Two central banks, two different inflation problems, one structural argument about spacing. If you are carrying a yen position through 30 October, that is the sentence to have in front of you, and it is not a directional call: it says the near meeting is quieter and the far meeting is not.
The price did something worth your attention. Dollar–yen had already been pulled toward 158.00 by broad dollar selling in London. On the report the yen sold off and the pair printed about 158.37 — 37 pips up, the intuitive direction for a central bank declining to tighten. Then the selling intensified and took it through, to 157.83, a 54-pip reversal, with the pair at 157.91 shortly after. Euro–yen made 176.49 and sterling–yen 208.90 on the same leg. A dovish report produced 37 pips of yen weakness and then 54 pips of yen strength through the starting point. That is what a reaction function looks like when the market was already positioned for the news.
Both conventions finally agree, and that is the only reason this level is worth anything
For two consecutive runs this desk has published that the dollar–yen 200-day average decides nothing, because the answer depends on which average you mean. On 12 October, 158.61 was a clean break against the simple band near 158.53 and an eleventh day of an existing regime against our own exponential reading just under 158.00. On 13 October, the Tokyo give-back to 158.19 was a clean failure against the simple band and a non-event against the exponential one — the same indicator, opposite verdicts, two days running.
At 157.83 that stops. The simple readings in our file run 158.53, 158.54 and — new this morning from a fourth publisher, and again unlabelled — 158.56. The low is 70 to 73 pips below all of them. Against our exponential reading just under 158.00 it is 17 pips below. For the first time in the sequence, every convention we hold says the same thing. It is also 78 pips below Monday’s close, 92 below the New York high, and 121 below the 159.04 high of 24 September that this desk uses as its intervention reference.
The lesson is not that the level finally worked. It is that a level is only usable when the spread of its definitions is smaller than the move you are testing against it. Our definitions span about 56 pips. A 36-pip excursion past one of them told you nothing on either of the last two days. A 73-pip one tells you something today. Size the test to the measurement error, not the other way around.
What this does not tell you
It does not tell you that the 85 per cent is wrong. It tells you we cannot source it, which is a statement about our reading and his sourcing rather than about his judgement.
The feed figures are not ours and not simultaneous. The 7 October breakdown comes from a publisher that is new to this desk and is a single reader on that date; the derivation of 78.3 from 100 minus 21.7 is ours, not the publisher’s. The 24 September breakdown is a second publisher on a different date, so it corroborates the structure and not the level. We hold no reading of the feed’s own cumulative buckets for today, and nothing here should be treated as current pricing.
The Bank of Japan story is wire reporting from unnamed officials, carried by two publishers that both appear to trace to the same originating report. That is two readers of one source, not two sources. Nothing has been said by the Bank of Japan.
The Federal Reserve Board’s calendar page carries no timezone anywhere on it. Our conversion to 07:45 UTC rests on two entries on that page reconciling exactly with times we hold from independent sources, which is good evidence and not a published fact. Note also that the remarks reported from New Delhi are a secondary reading: no transcript or prepared text for this appearance has been published, and the quotation we use is one fragment.
On the price: the 157.83 low and the 158.37 high are confirmed by two publishers, but we have no settled London close and no figure for where the pair sits at the New York open. Our exponential 200-day reading is five days old and was never recomputed; this desk has still never calculated a 200-day average from a price series of its own. And the three regional Federal Reserve presidents a vendor lists as speaking today appear nowhere on the Board’s own calendar — we have not established when any of them speaks.
Related
- FX events calendar — the scheduled windows, in UTC.
- Signals — what the desk is watching rather than what it thinks.
- EA presets — news-window and exposure settings.
