Before yesterday’s US inflation print, four feeds put an October Federal Reserve increase at 16, 19, 19.4 and 20 per cent. Four hours after it, one reading of the futures market had it at 43.7 per cent and another at about 45. The October meeting roughly doubled on one release. But the number worth your attention is not the October probability at all. It is what the same published bucket set says about December — and the arithmetic says December did not move with October, or against it, or in any relation to it whatsoever.
From about nineteen to about forty-four, and three publishers give three numbers
At 13:13 UTC a relay of the CME’s own futures-implied tool put an October increase at 43.7 per cent and a hold at 56.3. A research note later in the New York morning put it at about 45 per cent and said explicitly that it had been about 20 before the print. A third publisher’s headline put it at 49 per cent and called the meeting a coin flip; that page refused our reader, so we are not publishing 49 as a figure, only recording that a higher number was visible and that we could not read behind it.
So the readable spread is 43.7 to about 45, which is narrow by the standards of this file — a month ago three feeds were eight points apart on a binary event the day before it happened. The repricing itself is large and consistent: against our carried pre-print readings the October probability rose by 23.7 to 27.7 points, a factor of between 2.19 and 2.73 depending on which feed you started from.
What makes that uncomfortable rather than merely interesting is where it came from. Every one of the eight readers this desk assembled over five slots missed both core fields of yesterday’s print — core monthly came in at 0.3 against a unanimous 0.2, and core annual at 2.6 above the entire 2.39-to-2.5 range. A twenty-five-point move in a meeting probability built on the two fields nobody forecast correctly is not a market absorbing information. It is a market discovering it had the wrong prior.
The buckets sum to exactly 100.0 for a third date, so stop arguing about the method
The same relay publishes the cumulative distribution for December rather than a single meeting probability: 4.3 per cent for no change at all by December, 55.3 per cent for a cumulative 25 basis points, and 40.4 per cent for a cumulative 50. Those sum to 100.0 exactly.
This desk spent three slots in early October establishing that these buckets are a genuine distribution rather than a vendor’s presentation layer, and verified the sum for 24 September and 7 October. Yesterday is the third date. The method is settled: you can read a cumulative bucket set off this feed and do conditional arithmetic on it without worrying that the numbers are decorative.
The immediate outputs are worth having. The probability of at least one increase by December is 100 minus 4.3, or 95.7 per cent. The expected number of increases by December is 1.361, which is 34.0 basis points of tightening priced into a two-meeting window.
December is priced as independent of October, to two hundredths of a percentage point
Now the part that should change how you carry exposure through the next six weeks.
Work the conditionals off the published numbers. The 40.4 per cent two-increase bucket requires increases at both meetings, so the probability of a December increase given an October increase is 40.4 divided by 43.7, which is 92.45 per cent. The probability of an increase in December having not had one in October is the 95.7 at-least-one figure less the 43.7 October figure, divided by the 56.3 October-hold figure: 52.0 over 56.3, or 92.36 per cent.
Those two conditionals differ by 0.086 of a percentage point. Put the other way round: 43.7 per cent multiplied by the unconditional December probability of 92.4 per cent gives 40.379, against a published two-increase bucket of 40.4. The feed is pricing the December meeting as statistically independent of the October outcome, to within two hundredths of a point.
That is economically strange and operationally important. If October and December were substitutes — tighten now or tighten later — an October increase would pull the December probability down, and the two conditionals would be far apart. They are not apart at all. The market is saying that whatever happens on 28 October tells you essentially nothing about 9 December.
For a system carrying dollar exposure across both dates, that is a statement about your risk budget rather than about rates. You cannot hedge one meeting by taking a view on the other, because the market does not think they are related. Two near-independent events at roughly 44 and 92 per cent do not offset; they stack. And a desk that has been treating the December meeting as the resolution of the October one has been netting a position the price structure says is not nettable.
Two Fed officials spoke into a twenty-five-point repricing and neither named a number
This desk has a running count: officials who have discussed the rate path in public, against officials who have attached a probability to it. Yesterday added two more to the numerator and none to the denominator.
A Richmond-based official, speaking to a published account at 18:46 UTC, said the print ran hotter than he would have hoped, named today’s producer price release as the input that will inform his inflation outlook, and said in terms that he does not want to project the outcome of the October meeting. No probability, no threshold, one named input. The Chicago president, in an interview reported at 19:24 UTC, called the print discouraging, said more bad months of inflation would be unacceptable, said the focus belongs on the trend rather than a single report, and gave exactly one number: the 2 per cent target. He added that he is not a fan of signalling moves before meetings.
The tally is now eight officials and one cited probability, and that one — a governor’s roughly 85 per cent of 8 October, which he himself attributed to the futures market rather than to his own judgement — is nine days old and was superseded by yesterday’s print. A third official spoke before the release, billed under supervision and regulation; we predicted she would name no probability and we could not obtain a reading of her remarks, so that prediction stands untested rather than confirmed.
The practical reading is not that the Fed is being evasive. “More bad months would be unacceptable” is a reaction function, and so is naming the PPI release as the input. They are simply not quantified, and the only quantified thing in the entire repricing came from the futures curve. If you want a number for 28 October, the market is the only entity publishing one, and it has been wrong about this release’s inputs twice in two months.
And the long end is still unmeasured for a third trading day, past the issuer’s own release date
One loose end from yesterday, and it has not improved. The Federal Reserve’s own daily ten-year constant-maturity series, read at 23:10 UTC last night, still carries 9 October at 5.24 per cent as its most recent observation. There is nothing for 12, 13 or 14 October. The page was last updated at 20:17 UTC on 13 October, and it states its own next release date as 14 October — a date that has now passed.
So three trading days of the single most quoted yield in the world are unpublished at the issuer, through a session in which one research note put the ten-year up about four basis points to around 5.25 per cent on the inflation print. We cannot confirm that figure against the issuer, and we were wrong about this exact field twice last week by assuming the newest observation we could see was the newest that existed. We are not making that mistake a third time: the vendor’s 5.25 is plausible and single-sourced, the issuer is silent, and those are two separate facts.
What this does not tell you
The whole conditional argument rests on one bucket set from one feed, reached through a relay rather than from the CME directly. If the 4.3, 55.3 and 40.4 are rounded from different underlying precisions, the 0.086 gap between the two conditionals is inside the rounding, and the correct statement weakens from “independent to two hundredths of a point” to “indistinguishable from independent at the precision published”. That weaker statement still carries the operational conclusion, but it is the honest version and we would rather state it here than have it found.
The same note that gave about 45 per cent for October gave about 85 per cent for December. Off the bucket set, the unconditional December-meeting probability computes to 92.4, which is 7.4 points higher. We do not know whether those are two readings of one quantity or two different quantities — a December-meeting probability and a cumulative-by-December probability are not the same object, and this desk has already been caught once this week treating a model estimate and a survey as the same object. Neither publisher says which it means. That note’s separate claim of at least three increases priced over twelve months is a different horizon again and is not comparable with the 1.361 figure.
We have no primary source for any probability in this article. Every figure came from a publisher reporting a feed, and only one of them named the feed.
Nothing here is a forecast of the October decision, and the argument above is explicitly not one: a market pricing two meetings as independent is making a claim about structure, not about outcomes, and it can be wrong about both.
Related
- Economic calendar and event windows — 27–28 October and 8–9 December are both on it.
- Signals — for carrying exposure across two scheduled events rather than one.
- Expert advisor presets — where correlated exposure and a risk budget actually get set.
- Eight Readers Agreed on One Field. It Is the One They All Got Wrong. — the print this repricing came from.
- Waller Spoke in New Delhi and Named No Probability — where the bucket method was settled.
