On 12 October this desk published that a Federal Reserve governor’s 85 per cent was not his judgement but the market’s own price, and that it was roughly three weeks old when he said it. The speech text, read at the Federal Reserve’s own website rather than through a wire summary, settles both halves and it only agrees with one of them. He attributed the figure to the market explicitly, in a single sentence, with a date attached — and the date is the day before he spoke, not three weeks earlier. Our attribution was right. Our staleness reading was wrong. And being wrong about the date makes the arithmetic problem larger rather than smaller, because our own compound of the same market’s two meeting probabilities was 79.0 nine days ago. He speaks again at 07:45 UTC this morning.
What he actually said, in his own words, with his own date on it
The sentence is: according to futures market prices for the federal funds rate as of yesterday, traders see an 85 per cent chance of at least one hike by the end of the Committee’s December meeting, and nearly 20 per cent chance of two hikes. That is the whole of it. The number is introduced as a market price, the instrument is named, and the vintage is given as the previous day.
So the half of our claim that matters is confirmed at the primary source. The 85 was never a policymaker’s forecast. Anyone who stored it as one — and the sentence has been quoted widely enough that plenty of people did — stored a futures quote under a governor’s name. If you keep a file of what officials think, this entry does not belong in it.
The speech was delivered on 8 October. “As of yesterday” therefore means 7 October pricing. We said three weeks. We were out by nineteen days, and we should not have been: the sentence carries its own timestamp and we inferred one instead of reading it. The rule this desk published four days ago — store the as-of date with every number — applies to us first.
Being wrong about the date makes the problem worse, not better
Here is why the correction costs us something. Our published argument was that the 85 was arithmetically exact for a market three weeks older than the sentence: the compound of the same feed’s October and December meeting probabilities stood at 85.9 per cent on 18 September, 79.0 on 9 October and 74.9 over the following weekend. On that reading the figure was simply stale and the staleness explained it.
Remove the staleness and the explanation goes with it. If the market genuinely priced 85 per cent on 7 October, then either the same measure fell about six points in the two days to 9 October, or our compound does not reproduce the number he was reading. Those are different failures with different remedies. The first is a fast-moving market and nothing is wrong. The second means we have been multiplying two probabilities that are not the two inputs the feed uses, and every compound figure in our file inherits the error.
We cannot tell you which it is, because we have not re-read the feed at the 7 October timestamp. What we can tell you is that the comfortable version — the official quoted a stale number — is dead, and the uncomfortable version is still live and is about our arithmetic rather than his.
Three probabilities we did not have, and one of them is a conditional 94 per cent
The speech carries three more market-implied figures that this desk has never held. Nearly 20 per cent for two hikes by December. By the March 2027 meeting, nearly 80 per cent for at least two hikes, and 33 per cent for three or more.
Put the first and third of those against the 85 and the structure is clear. Eighty-five per cent for one hike by December, twenty for two by December, eighty for two by March. The market is not doubtful about the second increase; it is doubtful about where the second increase lands. Divide the eighty by the eighty-five and, conditional on the first hike arriving at all, the market is about 94 per cent sure a second follows within two more meetings. That is a stronger statement than the headline number and nobody quotes it, because it requires doing a division on two sentences from one speech.
It also happens to agree with something else in the same speech: that the hikes need not come at consecutive meetings. The probability structure is a timing distribution, not a doubt about direction, and the speaker said so in prose while the market said so in prices. That is the rare case where the two agree and the quoted figure hides it.
For what the Committee itself wrote down: sixteen of the eighteen participants who submitted dots anticipated at least one more increase during the two remaining meetings of this year, four of those sixteen expected two, and eight participants expect rates 50 basis points higher at the end of 2027 than today. Sixteen of eighteen is 88.9 per cent of a show of hands. It is not 85 per cent of anything, and the two numbers are not comparable — one counts people, the other prices a security. The policy rate itself is confirmed at the primary source at 3.75 to 4.00 per cent after a 25 basis point increase in September, which is the figure this desk has been carrying.
On Monday a Fed president explained her reaction function and put no number in it
Four days later, into a closed United States bond market, the president of the Cleveland Fed gave a speech whose subtitle is about communicating her reaction function. Read it for thresholds and there are none. The figures it does give are backward-looking: headline personal consumption inflation at 3.4 per cent in August against 2.7 a year earlier, core at 3 per cent against 2.9, the Cleveland Bank’s own trend estimate in the mid to high 2 per cent range, unemployment at 4.2 per cent, inflation above target for more than five years, and a 2 per cent objective.
The conditions are entirely qualitative. Convincing signs of labour-market deterioration would likely prompt a response. Inflation staying elevated relative to her forecast, or an inflationary mindset settling in, could call for raising rates further. She says in terms that this is not forward guidance, and she states a preference for smaller adjustments made relatively quickly over a larger correction later. No trigger carries a value.
That is not a criticism, and this is the part worth sitting with. A reaction function with numbers in it is forward guidance, which she declined to give. A reaction function without them is honest and unbacktestable. So in one week you have a governor quoting the market’s numbers precisely and dating them, and a president describing her own conditions precisely and refusing to number them. Only one of those can be stored in a system. The other has to be read.
The practical consequence this morning: dollar–yen sat at 158.64 at 08:00 Tokyo time, three pips above a New York close read at 158.61, in an eleven-pip range, with one publisher saying plainly that the market is waiting to see how United States long-term yields move now that the holiday is over. The 07:45 UTC appearance is nine hours before that. If the 85 is repeated this morning while the compound sits near 75, the staleness question reopens as a question about his judgement rather than about the vintage of his quote — and that is a cheaper test than anything on this week’s calendar.
What this does not tell you
We have not re-read the probability feed at its 7 October timestamp, so we cannot say whether the market really printed 85 that day or whether our compound is built from the wrong two inputs. Until that is done, this article corrects our date and leaves our arithmetic unaudited, which is the weaker of the two things we owed you.
The three additional probabilities are read from one document, and the conditional 94 per cent is our division of two of its sentences, not a figure anyone published. The two speeches are primary sources and the inflation figures in them are the speakers’ own, not ours and not the statistical agency’s; we have not opened the agency’s releases to check them. The 158.64 and the 158.61 come from one Japanese publisher’s 08:00 summary, with a separate desk giving the New York session range as 158.57 to 158.75, and no yield print exists for Monday at all because the cash market that produces one was shut.
And nothing here is a view on whether the Committee raises rates this month. The point is narrower and duller: the number you store from a speech is usually a price with a date on it, and if you store it without the date you will later mistake it for an opinion.
Related
- FX events calendar — the 07:45 UTC appearance and the 12:30 print on Wednesday.
- Signals — how a scheduled speech is treated when it carries no release.
- EA presets — news-window settings for central-bank speeches.
