For five runs this desk has quoted a single figure for speculative yen positioning: a net long of 55,400 contracts. It came from a vendor table, and this morning we went to the agency that produces it. The Commodity Futures Trading Commission’s own report gives non-commercial longs of 176,266 contracts against shorts of 120,826. That is a net long of 55,440, which rounds to 55,400. The number is right. What is wrong is treating it as a position.
The vendor series checks out at two consecutive points, which is more than we expected
We had flagged our own 55,400 as possibly a vendor’s “previous” field read as a print. It is not. The agency’s report is stamped as of 29 September 2026, and 176,266 minus 120,826 is 55,440. The week’s change is also published: longs down 16,008, shorts up 534, so the net fell 16,542 and the prior week stood at 71,982 contracts. The vendor table we have been reading carries 72.0 thousand for 22 September. Two consecutive weeks reconcile to the agency’s own arithmetic, to the rounding.
One thing we had recorded as a standing caution is also wrong, and we are correcting it here. We have been telling you that the vendors label this series by release date rather than by the date the positions were measured. The table we read labels its rows 1, 8, 15, 22 and 29 September. Every one of those is a Tuesday. It is labelling by measurement date, not release date, and our caution was pointing at a problem that this vendor does not have.
The same report decomposes the same open interest twice, and the two halves disagree
Open interest in the contract is 360,720. The legacy report splits that into commercial, non-commercial and non-reportable. A second report in the same release splits the identical 360,720 into five different buckets. We checked the arithmetic in code: the long positions of the five buckets plus the spreading positions sum to 360,720 exactly. The two reports are not two measurements. They are two cuts of one measurement.
And the second cut takes the net long apart. Asset managers and institutional accounts are long 117,526 against 65,892 short — a net long of 51,634. Leveraged funds are long 77,429 against 91,590 short — a net short of 14,161. Dealers are net short 86,702. The “speculative net long” that gets quoted as a crowded position is the residue of books pointing in opposite directions.
Last week they moved in opposite directions too. Asset managers added 6,794 longs and cut 3,211 shorts, a net change of plus 10,005. Leveraged funds cut 15,494 longs and added 6,090 shorts, a net change of minus 21,584. That single bucket’s change is 1.30 times the entire net change in the headline figure. If you read the headline as “specs trimmed a long”, you missed that one set of accounts was buying into it.
Nobody flipped. The long evaporated
Here is the part that matters for sizing, and it is in the legacy report where everybody can see it. Of the 16,542-contract fall in the net long, 16,008 came from longs being closed and 534 from shorts being opened. That is 96.8 per cent liquidation. Longs were cut thirty times as hard as shorts were added.
A net long that falls because shorts arrive is a market taking the other side. A net long that falls because longs leave is a market losing interest. Those two things behave differently when a headline hits: the first has somebody who needs to buy it back, the second does not. The headline number is identical in both cases, which is precisely why it is not enough to size on.
The five-week series makes the point at a larger scale. Net positioning ran minus 92.2 thousand on 1 September, plus 10.8 on the 8th, plus 120.4 on the 15th, plus 72.0 on the 22nd and plus 55.4 on the 29th. That is a 212,600-contract swing in two weeks and then a 65,000-contract give-back — 30.6 per cent of the swing handed straight back, leaving the position at 46 per cent of its peak. Any filter keyed to “positioning is stretched” fired four different ways inside a month.
It was measured seven days ago, and three days before the only print that mattered
The report is as of Tuesday 29 September. Today is 6 October. The 29,000 payrolls print — the one that missed a 27-analyst median by 61,000 and that this desk wrote about on Monday — landed on Friday 2 October, three days after the measurement. Nothing in the 55,440 knows about it.
The agency publishes each Friday at 3:30 p.m. Eastern time, which is 19:30 UTC while the United States is on daylight time, using data from the immediately preceding Tuesday. So the first reading that contains the payrolls week is released this Friday, 9 October, at 19:30 UTC, measured as of today. Worth noting where that rule is actually written down: the release-schedule page lists release dates and nothing else. The three-day lag is documented on a different page. Two separate reads of the schedule page this morning produced two different and both-wrong measurement dates, and the only reason we caught it is that neither of them landed on a Tuesday.
The 12:30 window, and whether it deserves to be in your schedule
At 12:30 UTC the Bureau of Economic Analysis publishes the August international trade figures, confirmed on the Bureau’s own release schedule at 8:30 a.m. Eastern. Consensus for the balance is minus 102.0 billion dollars against a minus 88.6 billion prior, with forecast exports of 314.6 billion and imports of 416.8 billion — which differences to minus 102.2, so the component forecasts and the headline consensus are consistent.
The reason to think about this one as a schedule question rather than a trade is that most of it is already out. The advance goods balance for August was published late last month at minus 132.6 billion against a minus 115.00 billion expectation, a 17.6 billion miss. If goods is minus 132.6 and the full balance is minus 102.0, the implied services surplus is 30.6 billion. The goods deficit alone is 1.30 times the total balance. A release whose largest component was published a week ago is a thin reason to be flat, and that is a decision you can make before 12:30 rather than after it.
What this does not tell you
It does not tell you which way the yen goes. A net long of 55,440 contracts is 46 per cent of what it was three weeks ago and it is still a net long; both halves of that sentence are true and neither is a signal.
The bucket labels are the agency’s, not ours, and they are self-reported classifications rather than descriptions of behaviour. “Leveraged funds” is not a synonym for fast money and “asset manager” is not a synonym for patient. We have published the decomposition because the sign disagreement between the two is a fact; the story about who those accounts are is an inference and we are not making it.
We have also not established that the legacy non-commercial bucket maps onto any combination of the five buckets in the second report. It does not, arithmetically, and we have deliberately not implied a correspondence. Both cuts add to 360,720 and that is the whole of what we have verified.
The trade-balance consensus and component forecasts are one calendar provider’s figures. The release date and time are the Bureau’s own. The advance goods figure is a second publisher’s reading of a release we did not open directly.
Related
- FX events calendar — the 12:30 and 19:30 UTC windows referred to above.
- Signals — how we treat positioning data that is a week old by the time you read it.
- EA presets — filter settings for scheduled releases whose main component is pre-announced.
