The Commodity Futures Trading Commission measures positions at the Tuesday close and publishes them on Friday. Tuesday was 13 October. Friday’s report, due 19:30 UTC on 16 October, will therefore record the yen market at a single instant on a day that ran 158.65 to 157.83 — eighty-two pips, with a fifty-four-pip reversal inside it — and closed at 158.33, fifty pips above its own low. Whatever positions that report shows, it will show them as of the one minute of the day at which almost nothing that happened was still visible. The ten-year Treasury did the same thing, and closed within a third of a basis point of a number the Federal Reserve has not measured since Thursday.
The measurement instant is the worst one of the week
This is not a complaint about the CFTC. A weekly survey has to pick a moment and Tuesday’s close is a perfectly good moment. The point is what you are entitled to infer from it, and this week the answer is unusually little.
Work the day. Tokyo traded 158.65 down to 158.19. London pulled the pair toward 158.00 on broad dollar selling, lifted it to about 158.37 on a report that Bank of Japan officials lean against an October increase, and then sold it through to 157.83. New York bought it back into the 158.40s and closed at 158.33. The close sits fifty pips above the low, thirty-two below the high and twenty-eight below Monday’s close.
So a report arrived that, on its face, argues for a weaker yen; the yen weakened thirty-seven pips, strengthened fifty-four, and then gave fifty of those back before the measurement was taken. Any position opened on the first leg, any position opened on the second leg, and any position closed on the third all exist in the same single row of Friday’s table. The report cannot tell you which of those three things the market was doing, and neither can the change from last week, because the change is the difference of two instants a week apart, not an account of what happened between them.
The practical form of this is a rule about staleness that is sharper than the usual one. The standard caution is that COT data is three days old on release. The real caution is that it was never a description of a period at all. It is two photographs seven days apart, and this week one of the photographs was taken in the middle of a round trip.
The ten-year closed a third of a basis point from a number nobody has measured since Thursday
The same shape appeared in a second instrument on the same day, and it is worth laying beside the first because it shows the problem is about measurement rather than about the yen.
Three vendors read the US ten-year yield at midday on 13 October: 5.20, 5.1937 and about 5.19, with stated daily changes of minus three basis points, minus four, and more than minus five. We published a piece yesterday arguing that the levels agreed to about a basis point while the changes did not, because all three starting points were a Monday quote on a closed cash market. Then the day closed. The late-session reading was 5.2168 per cent, down 2.72 basis points, with a second publisher giving it as near 5.22 and down about three.
Put the close against the three midday readings and it is 1.68, 2.31 and 2.68 basis points above all of them. Put it against the last observation the issuer actually measured — 5.22 per cent on Thursday 8 October — and it is 0.32 of a basis point below it. The Federal Reserve’s own daily series, re-read at 23:15 UTC tonight, still ends on 8 October and still has no observation for 9, 12 or 13 October, which is the third reading of the same absence in a day and the fourth across two days. Its page states a next release date of 13 October, which has now passed.
So the ten-year spent a session being reported as three to five basis points lower, and finished a third of a basis point from where it was five days ago. That is the yen’s day in a different instrument: a large intraday move, a close almost exactly back at the start, and three different published descriptions of the change depending on which instant anyone chose.
What Friday will not settle, and we already know what it will not settle
There is a second reason to lower expectations for Friday, and it is one this desk created. For the 6 October Tuesday we hold four figures for the yen, all from one report and all correct: leveraged funds net short 25,043 contracts, asset managers net long 67,795, dealers net short 84,905, and a legacy large-speculator net long of 62,325. The extremes are 152,700 contracts apart, which is 1.909 trillion yen, or about 12.06 billion dollars at last night’s close.
We published on 12 October why those do not reconcile: they are two different cuts of one open interest, and partial sums of the newer breakdown move further from the older total as true terms are added — from 19,573 contracts out to 104,478 when the third category is included. Nothing about Friday’s release fixes that. It will produce the same four-way split one week later, and anyone quoting a single yen positioning number from it will be quoting a category without naming it.
And the prediction we owe a verdict on is still unscored. We have searched four times now for a second numeric publisher of the 6 October yen figures and found none. One source dates itself unambiguously — positions as of Tuesday 6 October, published Friday 9 October — and it is carried from our earlier reading rather than re-fetched tonight. Everything in the paragraph above therefore rests on a single reader. We are not claiming the prediction and we are not quietly dropping it: Friday 16 October is where it gets closed or abandoned, and on current evidence it gets abandoned.
A fifth reader exists. It has the wrong month and the opposite sign.
Tonight’s search did turn up something, and the honest description is that it makes the picture worse rather than better.
One publisher gives the yen non-commercial net position at minus 92,200 contracts against a previous minus 63,300 — a deepening short of 28,900 in a week. It names the category explicitly, which most do not. It is dated 5 September 2026. A second publisher gives leveraged funds for the week to Tuesday 29 September as 77,429 long and 91,590 short, net short 14,161, with longs cut by 15,494 and shorts added 6,090. Both halves of that reconcile exactly: the gross counts give the net to the contract, and the two change components sum to 21,584, which is precisely the weekly change the other source reports for the same week.
That is good source behaviour and it is the wrong Tuesday twice over. Worse, the sign conflict is now explicit. The legacy speculator series we hold for 6 October is net long 62,325 contracts. The non-commercial series — which is the same legacy taxonomy — was net short 92,200 five weeks earlier. The gap is 154,525 contracts, about 1.932 trillion yen of notional, and it is either a genuine five-week flip from a large short to a large long, or the two publishers are not labelling the same thing. We cannot distinguish those with what we have read, and we are not going to guess which.
The reconciliation needs the rows nobody has fetched: other reportables and non-reportables, which the issuing agency publishes and which this desk has never read. Until then, the four-way split stays a demonstration rather than a reconciliation.
What a system should do with a snapshot taken at one minute
Three changes, none of them directional.
First, store the instant with the number. A positioning figure needs three labels before it enters a rule: the category, the as-of date, and — new as of this week — whether the as-of instant sat inside an unusually wide session. Friday’s report will carry a value measured at the close of an eighty-two-pip day that finished twenty-eight pips lower. A reader who stores only the number and the date will treat that as equivalent to a value measured at the close of a seven-pip day, and it is not.
Second, stop reading weekly changes as flows. The difference between two Tuesdays is a net of everything in between, and this week the in-between included a fifty-four-pip reversal through its own starting point. A smaller net change this Friday would be perfectly consistent with the busiest week in a month.
Third, size tonight off the event and not off realised range. The Asian session this morning has produced seven pips, 158.32 to 158.39, against yesterday’s eighty-two. Any volatility filter calibrated on recent range will read this as the calmest market of the week, forty minutes before Japanese money stock and roughly thirteen hours before US September consumer prices at 12:30 UTC. The range is low because the event has not happened. No expiry ladder is published for tonight’s 14:00 UTC New York cut — the index returned a client error again, the thirteenth observation of that problem — so treat that minute as a window of unknown size rather than an empty one.
What this does not tell you
It does not tell you what Friday’s report will say, and nothing here is a forecast of it. It does not tell you that positions were unwound and rebuilt during yesterday’s round trip — we have no transaction data at all, only the price path, and a round trip in price is consistent with heavy two-way flow and with almost none.
The four 6 October positioning figures rest on a single numeric publisher, carried from our 12 October reading and not re-fetched tonight. The legacy 62,325 figure’s as-of Tuesday is our own arithmetic inference and not stated on its page. The 154,525-contract gap between the two legacy readings compares two different weeks five weeks apart and is offered as an unresolved labelling question, not as a measurement of anything.
On the yield, the Federal Reserve has published no observation for 13 October, so every figure quoted for the day is a venue quote rather than an issuer measurement, and none of the readings we hold — midday or close — carries a time of day. Part of the two-and-a-half-basis-point spread between the midday and closing values may be three clocks rather than a move. The 0.32-basis-point comparison with 8 October is arithmetically exact and economically close to meaningless on its own; it is here as a description of a round trip, not as a claim that nothing changed.
Monday’s 158.61 close is carried from our own file. We hold two readings of last night’s close, 158.33 and a late-session 158.30, three pips apart, with no official venue fix.
Related
- FX events calendar — the 12:30 UTC print, the 14:00 cut and Friday’s 19:30 UTC positioning release.
- Signals — why a weekly snapshot belongs in a sizing rule rather than a direction rule.
- EA presets — realised-range and news-window filters, and what they get wrong on the day before a print.
