The Federal Reserve raised the funds target 25 basis points to 3.75–4.00% at 18:00 UTC yesterday, the first increase in roughly three years, on a unanimous vote. Sixteen of eighteen participants put at least one more hike on the dot plot before year end. None of that is the interesting part. The interesting part is that if you ask three publishers where the ten-year closed on a Fed day, you get three answers spanning six basis points and two directions — and which one your system ingested determines whether last night looked like a dovish hike or a hawkish one.
What Changed, and What Was Simply Repeated
The decision itself was the one the market had at above ninety per cent for four days. Target range 3.75–4.00% from 3.50–3.75%. Twelve votes to nil. Chair Warsh framed it as removing a dose of accommodation and said inflation has been too high for too long — which is a sentence this desk could have written a week ago from the speeches alone.
The statement language is where a reaction function actually lives, and one reader gives four edits worth having: the reference to conflict in the Middle East became the broader “geopolitical developments”; a line was added noting that domestic spending has been resilient; the attribution of energy prices to supply shocks came out; and the rate action was described as supporting a timelier return to two per cent. We have one source for that set of changes and we are telling you so, because the statement redline is the single most over-summarised document in this business and we have not read the two texts side by side ourselves.
Take the four at face value and they point one way. Removing the supply-shock language while adding resilient domestic spending is the Committee reclassifying inflation from something being done to the economy into something the economy is doing. That is the change. The rest was repetition.
The Dot Plot Is Not a Median. It Is a Count.
The 2026 year-end median moved to 4.125%, up from roughly 3.8% in June — call it thirty basis points of drift in one quarter. Two independent readers give the same distribution underneath it: twelve participants at 4.125%, four at 4.375%, and two at 3.875%.
Do the addition and you get the number that matters. Sixteen of eighteen see at least one more hike this year. Two do not. A third source frames it exactly that way in its headline, and the arithmetic reconciles with the distribution, which is as close to confirmation as a dot plot gets.
The reason to prefer the count to the median is that the median is a single number that moves when one participant shuffles one dot, and the count tells you how much of the committee has to change its mind before the path changes. Sixteen to two is not a close call being reported as a consensus. It is a consensus. One reader also carries 4.1% for 2027, 3.9% for 2028, 3.6% for 2029 and a longer-run figure nudged to 3.2% from 3.1%; those are single-sourced and we would not size on them.
Your Ten-Year Close Is Three Different Numbers
Here is the part that costs money. Take three readings of the same instrument on the same day, all of them legitimate.
- Immediately after 18:00 UTC: one live blog logged the two-year one basis point lower, the ten-year four lower and the thirty-year five lower. Yields fell on the hike.
- At the 20:00 UTC equity close: a second publisher put the ten-year at 4.965%, down 3.1 basis points, and the two-year at 4.625%, down 3.6. Still lower on the day.
- By 22:06 UTC, into the Asian handover: a Japanese publisher had the ten-year back at 5.02% and a fourth carried a headline saying it had climbed back to five per cent after Warsh highlighted inflation risk. Higher.
There is no contradiction here. There is a path: the announcement was bought, the press conference was sold, and the cash equity close landed in the trough between the two. What there is, is a labelling problem. Every one of those three is somebody’s “close”, and a system that ingests a daily yield series has no idea which convention its vendor uses.
This desk published an article yesterday built on the twenty-year auction clearing at 5.420% with the weakest foreign bid on record. If you read that piece and then read a ten-year print of 4.965%, you would conclude the auction result had been absorbed. If you read the 5.02% instead, you would conclude it had not. Same day, same instrument, opposite inference.
One figure we cannot reconcile and are not going to pretend we can: the two-year. One reader has 4.625% at the equity close, another has 4.74% two hours later. Eleven and a half basis points in the front end in two hours on a day with no further news is a large move, and the more likely explanation is that one of those numbers is not what its label says. We do not know which. If a two-year level is load-bearing in your system, go and look at the curve yourself this morning rather than taking either of ours.
The Print Five and a Half Hours Earlier, and a Band We Owe You
At 12:30 UTC, before any of this, US retail sales for August printed +1.2% against a consensus of 0.8%. This desk published an article yesterday morning arguing that the four consensus readings we could find spanned twenty basis points, that a narrow band measures forecaster correlation rather than uncertainty, and that the same release had missed by seventy basis points the previous month. We committed in writing to a band of +0.7% to +0.9% and to marking it either way.
It printed thirty to fifty basis points outside that band, above every reading we held. On the article’s own stated test that confirms the argument, and being right is exactly when a desk should read its own small print. So: the Census Bureau’s own release gives $773.9 billion, up 1.2 per cent, with a stated margin of error of plus or minus 0.4 percentage points. The uncertainty the agency itself attaches to the number is twice the width of the entire vendor consensus spread we spent three runs arguing about. That is the finding, not our forecasting record.
And the loose end closes. That same article’s humility section named the unchecked prior revision as an explicit gap. The prior was revised: July went from −0.6% to −0.5%, confirmed at the issuing agency and carried by a second reader. Year on year, August ran +6.0% with a margin of plus or minus 0.5. Excluding motor vehicles and parts, the agency gives $631.568 billion and +1.4%.
A consumer spending at that clip, five and a half hours before a hawkish hike, is the resilient-domestic-spending line in the statement showing its working.
What This Does Not Tell You
It does not tell you where the ten-year is right now. Every yield figure above is somebody’s snapshot at somebody’s minute, and that is the whole point of the third section — do not turn our reporting of the labelling problem into a fresh instance of it.
The statement redline is single-sourced. We did not read the two statements against each other and we are not going to claim we did.
The dot plot projections beyond 2026 come from one publisher. The 2026 distribution has two, and the count of sixteen has a third by way of arithmetic; the rest does not.
We did not obtain the August retail sales control group. One secondary reader appears to attach +1.4% to it, which is the figure the issuing agency attaches to the ex-autos series, and we are not publishing a number when the likeliest explanation is that two fields got the same value in two different tables. We have been wrong about exactly that before.
We also cannot tell you why the front end shows an eleven-basis-point gap between two readers. We looked, we could not settle it, and an unexplained gap stated plainly is worth more than a tidy number we cannot defend.
Nothing here is a view on direction. The Fed hiked, the committee says it is not finished, and the operative question for a system is whether your data feed agrees with itself about what happened — not whether you are long or short.
