On Friday this desk published that three yen positioning figures drawn from one report for one Tuesday sat 92,838 contracts apart, and called that spread a measurement problem. Over a weekend in which the futures market was shut and not one contract changed hands, the spread grew to 152,700. The market did nothing. We read one more row of the same table. A fourth trader category — dealers, net short 84,905 contracts as of the 6 October close — sits on the same page as the two we quoted, and adding it widened our own published figure by 59,862 contracts without a single new fact entering the world. If a number you publish can move by sixty thousand contracts because you scrolled further down, it was never a measurement of positioning. It was a measurement of your reading.
Four numbers, one Tuesday, 152,700 contracts apart
Everything below is the Japanese yen as of the Tuesday 6 October close, from the report published at 19:30 UTC on Friday 9 October. The gross counts are published, so you can check every net yourself.
- Leveraged funds: 66,123 long against 91,166 short, net short 25,043 contracts, a widening of 10,882 on the week.
- Asset managers: 143,300 long against 75,505 short, net long 67,795.
- Dealers: 53,281 long against 138,186 short, net short 84,905.
- Large speculators, from a different publisher using the older reporting breakdown: net long 62,325, a change of plus 6,885 on the week.
The extremes are the asset managers at net long 67,795 and the dealers at net short 84,905. That is 152,700 contracts, which at the 12.5 million yen contract size is 1.909 trillion yen, or 12.06 billion dollars at Monday’s 158.26 indicative open. The dealer short alone is 1.06 trillion yen, 6.71 billion dollars, and is 3.39 times the size of the leveraged-fund net short that every wire headline reports.
So when a headline tells you speculators are short the yen, ask which of these four it means. Two of them are long. One of them is three and a half times larger than the one the headline almost certainly used.
Our own published spread was an artefact of how far we read
This is the uncomfortable part and it belongs at the top rather than buried. Friday’s 92,838 was arrived at honestly: we took the widest gap between the figures we had, which were the leveraged-fund net short and the asset-manager net long. We did not notice that the same page carries a third trader group whose net is further from both of them than they are from each other.
We also described the spread as having grown by 20,589 contracts in four days, as though positioning had changed. Some of it had. But the move from 92,838 to 152,700 contains no market information whatsoever, and a desk that reports a spread without reporting how many rows the spread was computed over has published a number with the same defect it is complaining about. The honest form of Friday’s sentence was: of the trader categories we read, the extremes are 92,838 apart.
Record the general rule, because it is the same one that cost us four days on a probability and nine on an uncategorised contract count. A maximum over a set is a property of the set, not of the thing being measured. Any spread, band, range or error bar you publish has to carry the population it was taken over, or your own future self will read it as a fact about the market.
Why none of the four is wrong, and why they do not sum
The temptation is to decide which publisher is sloppy. None of them is. The newer reporting breakdown partitions the same open interest into dealers, asset managers, leveraged funds and other reportable traders, plus the non-reportable remainder; the older one splits it into speculators, commercials and small traders. These are two different cuts of one pie, and figures from different cuts are not comparable at all.
Watch what happens if you try. The two categories we quoted on Friday net to 42,752 contracts long, which is 19,573 short of the 62,325 the older breakdown reports for speculators. Add the dealers and the three sum to 42,153 short — now 104,478 contracts away from the same 62,325. Adding a true figure moved the discrepancy by a factor of five. That is the proof that the sum was never meaningful: a partition you have only partly enumerated does not approach the whole as you add terms, because the terms you are adding belong to a different partition from the total.
And there is a reason the dealer figure is both the largest and the least interesting as a directional signal. Dealers are the sell side. A dealer net short of 84,905 contracts against an asset-manager net long of 67,795 is substantially one position seen from two ends — the bank is short because the real-money account is long. Reading the dealer row as bearish conviction is reading the other side of somebody else’s bullish conviction.
One category has a second reader for this Tuesday. None of the others does.
On 9 October this desk published two predictions about the leveraged-fund figure. The first — that the net short would widen — was scored correct last night: minus 14,161 to minus 25,043, a widening of 10,882 contracts. The second was that two publishers would be roughly 2,600 contracts apart on the level and within a few hundred on the change. That prediction is still unscored, and this is the third consecutive run on which we have gone looking for a second numeric reader of the 6 October yen leveraged funds and failed to find one.
We checked a report published on 10 October titled for the week of 6 October. It carries no yen figures at all — only the dollar index, where asset managers are net long 19,096 contracts after adding 1,089 and leveraged funds net long 863 after adding 504. That is a clean, dated report that simply does not cover this pair.
What we do have is a cross-check on the previous Tuesday, and it is a good one. A report published 5 October gives the 29 September leveraged-fund position as 77,429 long against 91,590 short for a net short of 14,161 — agreeing to the contract with the other publisher — and decomposes the weekly change into 15,494 longs cut and 6,090 shorts added. Those two add to 21,584, which is exactly the weekly change the other publisher reports for that week. Two independent publishers reconciling to the contract on both the level and the decomposition is the strongest evidence of source quality in this file. It says nothing about the figure we actually need.
So the honest state of play: the 6 October yen leveraged-fund figure has one numeric reader, the dealer and asset-manager figures have one each, and the older-breakdown speculator figure has one. Four numbers, four single sources, one report behind all of them.
What a system should do with a positioning number this week
Start with the staleness, because it dwarfs the taxonomy. These positions were measured at the Tuesday 6 October close and published at 19:30 UTC on Friday 9 October. It is now Monday 12 October. Any rule firing off them is acting on a six-day-old photograph of a market that has since had a US consumer sentiment print, a Canadian employment report that missed its sign by seventy thousand jobs, and a weekend. The next measurement is the Tuesday 13 October close, released at 19:30 UTC on Friday 16 October, and it will score our second prediction if a second publisher prints.
Then pick one category and write down why. There is no neutral choice. Leveraged funds is the fastest-moving and the most crowded with wire coverage; asset managers is the slowest and the closest to real money; dealers is the largest and is mostly the mirror of the other two; the older speculator series is the one with the longest history and therefore the only one with a usable percentile. A filter keyed to “yen net short exceeds X” without naming the category is not a signal, it is a configuration accident, and on this one Tuesday the four candidates span 152,700 contracts.
And keep the week’s actual structure in front of you rather than the positioning. Monday’s Asian session was a thirty-three pip range, 158.07 to 158.40, with Tokyo closed for a public holiday and the US bond market closed while equities traded. Today’s expiry list for the 14:00 UTC New York cut names exactly one strike as readable text — 1.1225 in euro–dollar, eight pips above the Asian high of 1.1217 — and puts every notional, and every dollar–yen strike, inside an image. A Federal Reserve governor speaks at 07:45 UTC on Tuesday and US September consumer prices land at 12:30 UTC on Wednesday. Those are the two things that will move the figure you are arguing about, and neither is in the report.
What this does not tell you
Every figure here is single-sourced for the 6 October Tuesday. The gross long and short counts reconcile to the nets to the contract, and the prior week reconciles across two publishers, which is strong evidence that the source is careful — but it is evidence about the source, not corroboration of the number. If the publisher has one bad row, we would not know.
We have not established the other reportable or non-reportable positions, which is why we cannot show you the full partition and can only show you that partial sums of it do not approach the older breakdown’s total. The claim that the dealer short is largely the mirror of the asset-manager long is a structural inference from how the sell side carries risk, not something either publisher states, and we have not measured the correlation between those two series over time.
The older-breakdown speculator figure of 62,325 comes from a page that never states its own reporting Tuesday. We attributed it to 6 October at our Tokyo slot because subtracting its published weekly change of 6,885 lands on 55,440, which is exactly the figure we had carried from the previous week. That is a strong vintage inference and it is still an inference. Nothing here is a view on which way the yen goes; the point of the article is that the number most people would use to form that view has four values.
Related
- Release times, measurement dates and what actually lands when: /en/fxevents
- What our systems are doing with this: /en/signals
- Preset configurations and filters: /en/presets
