The Yen Long Is Measured on Tuesdays. Ours Was Ten Times Too Big.

The number is real, agreed, and stale by the only three days that mattered

Start with what is solid. The Commodity Futures Trading Commission’s Commitments of Traders report for the week ending 15 September puts non-commercial net yen longs at 120,359 contracts. One research desk gives the figure to the contract. A second gives it as roughly 120.4 thousand and adds that net exposure sits at the 88th percentile. A third, a Japanese-language morning note, gives 120,359 as well. That is three independent readers on one number, which on this channel is as good as it gets.

One reader also breaks it out: 237,951 gross long against 117,592 gross short. That is a single source and we label it so, but if it is right it is the more useful pair, because it tells you the net is a two-sided position rather than a one-way crowd — there are still nearly 118,000 short contracts in there to squeeze.

Now the problem. COT positions are measured at the Tuesday close and published the following Friday. This snapshot is Tuesday 15 September. In the seventy-two hours after that snapshot closed, the Bank of Japan raised its policy rate to 1.25%, the highest in thirty-one years, by seven votes to two; USD/JPY went from the mid-155s to a high of 157.96 or 158.05 depending on your reader; and the Bank contacted financial institutions to ask where the yen was trading, after which more than a hundred pips came back.

So the most-quoted yen positioning number of the week predates the hike, the dissents, the move and the rate check. It is not wrong. It is an accurate measurement of a market that no longer exists. One of the readers says as much in its own words — that for the yen, price and positioning are now pointing in opposite directions. That is not a trade signal. It is a warning that the signal is out of date.

A figure we published was never a level

On 14 September this desk published an article built on a yen net long of 103,023 contracts. We put the number in the headline. Going back to it this morning with two fresh readers, it does not survive.

Two independent sources put the net long for the week ending 8 September — the snapshot our 14 September article was reading — at 10,796 contracts and “a small net long of 10,800 contracts”. Those agree with each other to four contracts. Our published figure was roughly ten times either of them.

Here is our best reconstruction, and we are offering it as a hypothesis rather than a finding, because we could not confirm it. If the week to 1 September sat at about 92,000 contracts net short, then the move into a small net long of 10,796 is a one-week change of about 103,000 contracts. Our 103,023 looks like a weekly change that we published as a level. The word in that article’s own title was “flip” — which is what you call a change, not a level, and which suggests the error was in the labelling rather than in the reading.

We tried to settle it. The one series page that carries the 1 September week returned a 403 to us this morning, so the reconstruction above is unconfirmed and you should treat it as unconfirmed. What is confirmed is the part that costs us something: two readers agree the 8 September level was about 10,800, and we printed 103,023.

The reason this matters beyond our own embarrassment is that the same arithmetic makes the current week extraordinary. From 10,796 to 120,359 is a build of 109,563 contracts in a single week — and that week ended before the Bank of Japan did anything. Whoever put that position on was early, and by Friday night they were sitting on a hike they were right about and a price that went the wrong way.

The expiry board exists now, and spot is sitting on the strike

On Friday we told you the 21 September option expiry board had not been published when we looked. It was published this morning at 05:34 UTC. Two strikes carry commentary:

• USD/JPY at 157.00, described as likely to encourage two-way trade and make the price more stubborn into the 14:00 UTC cut.
• EUR/USD at 1.1470, described as having no particular technical significance of its own, with the nearest technical level a 100-hour moving average at 1.1500.

USD/JPY was quoted at 157.06 at 05:47 UTC. Spot is six pips from the strike. That is the configuration where an expiry actually bites: close enough that the hedging flow is live, in a session where Tokyo is shut for a third day and the books are thin, with an authority that checked rates on Friday sitting on the other side of any move higher.

We cannot give you the notional amounts. For the fifth time on this channel, the publisher’s expiry table is inside an image that does not extract, and the syndication mirror does not carry the numbers either. A strike without a size is a location, not a magnet. Treat 157.00 as somewhere the price may get sticky and not as a level with a known weight behind it.

And the day’s only Fed speaker is on two calendars an hour apart

There is no top-tier data today. What there is:

• Austan Goolsbee, Chicago Fed president, speaks — at either 09:30 or 10:30 UTC. One American calendar has him at 5:30 AM Eastern, which converts to 09:30 UTC. A Japanese morning schedule has him at 19:30 JST, which converts to 10:30. We cannot settle it and we are not going to pick. Block both.
• Chicago Fed National Activity Index, previous reading −0.08. Same problem: one calendar says 7:30 AM Eastern, or 11:30 UTC; the Japanese schedules say 12:30. Here we will lean, because there is a sanity check available — 12:30 UTC is 8:30 Eastern, which is the standard American data slot, and 11:30 UTC is 7:30 Eastern, which is not a slot anything releases on. The 12:30 reading is the one that lands on an institutional time.
• Three-month and six-month bill auctions at 14:30 UTC (10:30 Eastern), previous yields 3.970% and 4.060%.

On Friday this desk published a piece arguing that one widely-read calendar had every American time that day an hour early. Today the same one-hour gap appears on both of today’s American events, in the same direction. Two events is not proof of a systematic offset, and we are not upgrading Friday’s anecdote into a rule on the strength of it. But if you run a news filter keyed to a single calendar and your windows have been opening and closing sixty minutes off all week, that is worth an hour of your time today, on a day when there is nothing else to do.

What this does not tell you

• The reconstruction of our own error is unconfirmed. We believe 103,023 was a weekly change published as a level. We could not fetch the 1 September week to prove it, because the page returned a 403. If you can settle it, we are wrong until you do.
• The gross long and gross short split is one reader. 237,951 and 117,592 come from a single research page. The net, which three readers agree on, is the figure to lean on.
• A stale positioning number is not a directional signal in either direction. It does not tell you the longs have been stopped out and it does not tell you they are still there. It tells you the number you are quoting cannot answer the question. Friday’s report, covering the week to 22 September, is the first snapshot that will contain the hike — and it will still miss anything that happens between Tuesday and Friday of this week.
• We do not know the expiry sizes, we do not know whether the 157.00 strike is large enough to matter, and we did not verify the 157.06 spot print against a second reader at that minute.
• The Goolsbee time is genuinely unresolved. We did not go to the Chicago Fed’s own events page, which would have settled it in one fetch. That is a gap and it is ours.
• Nothing here is a view on the yen. Two of the three things in this article are arguments about what a number is, and the third is an argument about what a calendar says. None of them is a reason to be long or short anything.

Related

• FX event calendar — and the standing argument about whose times to trust
• Signals
• EA presets — news-window and expiry-hour filters