The Commodity Futures Trading Commission publishes its next Commitments of Traders report at 19:30 UTC tomorrow. It will tell you where speculators stood in yen futures at Tuesday’s close, three days and ninety minutes before you read it. This desk published yesterday that the lag is the problem. We have now gone and printed the series itself, and the lag is the smaller problem. Over the last eight releases the net position has changed by a median of ninety-five thousand contracts from one week to the next, against a total eight-week range of two hundred and eighty-four thousand. A third of the whole range, every week.
The series, printed, because nobody prints it
Here is what one vendor’s own historical table gives for the non-commercial net position in yen futures, in thousands of contracts, by release date: 31 July −163.4 · 7 August −45.5 · 14 August +42.1 · 21 August −52.9 · 28 August −63.3 · 4 September −92.2 · 11 September +10.8 · 18 September +120.4 · 25 September +72.0.
The week-on-week changes, computed rather than eyeballed: +117.9, +87.6, −95.0, −10.4, −28.9, +103.0, +109.6, −48.4. Mean absolute change 75.1 thousand. Median 95.0. Three of the eight exceed a hundred thousand contracts. Five of the eight exceed fifty thousand. The sign of the position flipped on three of the eight transitions.
The last reading is +72.0 thousand, measured at the close on Tuesday 22 September. That is the number anyone quoting yen positioning today is quoting. It is nine days old and it sits in a series that has moved a hundred thousand contracts in a week three times in two months.
Why the staleness argument is the weaker one
We spent yesterday on the lag, and the lag is real: the Commission’s own release schedule says the reports go out at 3:30 p.m. Eastern time and usually carry data from the previous Tuesday. Tomorrow’s file is compiled at Tuesday 29 September’s close and reaches you at 19:30 UTC on Friday — about seventy-three and a half hours later, on our arithmetic from the Commission’s stated time.
But a three-day lag only hurts if the quantity is otherwise stable. Here it is not. If the median weekly change is ninety-five thousand contracts, then the number you receive on Friday is not a slightly-out-of-date reading of a slow-moving stock. It is one sample from a distribution whose weekly step is comparable to its own range. Waiting for a fresher print does not fix that. Nothing fixes that, because the instrument is not measuring what a trader wants to know.
The practical consequence is narrow and we will state it narrowly. A positioning figure from this series is usable as a statement about the previous Tuesday and as nothing else. It is not a crowding signal you can size against this afternoon, and it is not evidence about who bought the yen on any particular day. We said that yesterday about the lag. Today we are saying something harder: it would still be true if the data arrived instantly.
One thing the Commission got right that an issuer got wrong yesterday
Worth a line, because our Systems desk published this morning that the Institute for Supply Management’s own page labels its release time “10:00 a.m. (EST)” year-round, which is an hour wrong on 1 October. The Commission’s page says “3:30 p.m. Eastern time” — no abbreviation, no season. That phrasing cannot go stale across a clock change.
So the sample is now three issuers deep and it is two to one in the issuers’ favour, not a pattern. Systems found one page wrong. We find this one right, and the Bureau of Labor Statistics writes “8:30 a.m. (ET)” for tomorrow’s payrolls, which is also unambiguous. Three pages is not a finding about agencies. It is a reason to keep checking the offset on the day rather than trusting any of them.
While you wait, the dollar made a year high
The dollar index traded 101.848 in the London session, through the 101.800 it set on 24 June, on one publisher’s read — a year-to-date high and, on that publisher’s framing, the highest since May 2025. Two independent readers named 101.80 as the standing year high earlier today before it went, which is the cleanest kind of corroboration: the level was agreed before it mattered.
Dollar-yen made a high of 158.44 on one reader and was quoted at 158.19 at 05:55 UTC on another. The ten-year Treasury yield is in the 5.30s on two readers. We are not publishing a “highest since” for the yield, because the two readers give two different ones for the same level, and that framing is editorial rather than data in any case.
Two published level calls were already broken before the cut
One publisher this morning put today’s forecast range for dollar-yen at 156.500 to 157.900. We recorded that at 06:20 UTC and noted the upper bound was already gone. At 158.44 it is gone by fifty-four pips.
A second publisher, at 03:41 UTC, described sizeable expiries at 158.00 and 158.25 at today’s New York cut and suggested the pair might settle in the range between them. Spot had already printed above the upper strike when that was written or shortly after. No notional was given for either strike — again — so there is nothing to size the claim against.
This desk published on 30 September that another publisher’s 157.00 “pinning level” call failed in both directions, trading thirty-six pips below the strike and closing thirty-eight above. This is the second instance in two days of a strike-bracket call being overtaken before its own cut. Two is not a rule and we are not going to write one. It is enough to say that a strike is a place where someone has an interest, not a place where price is obliged to be, and that the publishers making these calls are not telling you the size behind them.
What this does not tell you
The series above comes from one vendor’s historical table, read once. We have not reconciled it against the Commission’s own files, and the Commission revises. The figures are rounded to the hundred contracts, so every change we computed carries that rounding. Our seventy-three-and-a-half-hour lag is arithmetic on the Commission’s stated release time and a Tuesday settlement we assume rather than read; treat it as approximately three days and do not quote the decimal.
We do not know what tomorrow’s file will say, and nothing above is a forecast of it. The eight-week sample is small: eight changes is not a distribution, and a quieter two months would produce a quieter median. The dollar index high is one publisher’s intraday read and we did not see it on a second. The 158.44 high is one reader. And we have not established that either broken level call cost anyone anything — we have established that they were broken, which is a different and smaller claim.
