On 7 October this desk published three readings of the dollar–yen 200-day average three pips apart, and on 8 October we published the correction that mattered: the simple and exponential conventions sat about 52 pips apart, so the three-pip spread was a bad estimate of the uncertainty by a factor of seventeen. We ended that piece with a specification — a rule referencing “the 200-day” is unspecified until it names the convention, the publisher and the time of reading. Two publishers we had not read then give the figure now. One satisfies the specification and one does not, and the difference decides whether Monday’s break was a regime change or the eleventh day of an existing one.
Two new readings, and only one of them says which average it is
A broker’s technical note dated 7 October gives the 200-day at 158.53 and calls it a simple moving average in those words, alongside a 50-day at 157.70 and a 100-day at 159.53. A second house’s weekly outlook dated 10 October gives 158.54, a 50-day at 157.70, and a horizontal level at 159.50 it describes as converging with the 100-day.
Both land inside or one pip above the band we published on 7 October from three different readers — 158.50 relayed from a bank note, 158.52 from a Japanese technical page, 158.53 from a Tokyo broker’s morning outlook. So the first reading is a fourth independent reader on the same day, arriving at the top of our band to the pip. That is a good result and it is not the interesting one.
The interesting one is that only the 7 October note states its convention. The 10 October figure is given as “the 200-day moving average at 158.54” and nothing else. On our own published arithmetic the two conventions are about 52 pips apart, which means an unlabelled 200-day is a figure whose instrument is unnamed. We can infer it is simple, because 158.54 sits in the simple band and nowhere near the exponential one. An inference is not a statement, and the whole point of the specification we published five days ago is that you should not have to make it.
So the score on the test we set ourselves: one of two new readers passes, and the one that passes is the older of the two. Nine runs of recording this omission and the answer is not that the number is unavailable — it is that the label is.
What this does to Monday’s break, which we published without saying which convention it settled
This desk carried 158.51 as a mark since 7 October. On Monday at 09:54 UTC it went, and we published that it went: 158.61 in London on two independent readers, with Tokyo shut for Sports Day and the United States Treasury cash market shut for Columbus Day.
Read it as a simple average and Monday was the real thing. The simple band tops out at 158.53 or 158.54, and 158.61 cleared it by seven or eight pips before the overseas high extended to 158.75, twenty-two pips above. Read it as an exponential average — which we published on 8 October at just under 158.00, below every daily close since 1 October on that reading — and nothing happened on Monday at all. The pair had been above its 200-day for a week and a half and spent Monday getting further away from it, by something over sixty pips.
One indicator, one price, two opposite statements about whether a regime changed. We published the break and did not say which of the two it settled, and that is the omission in our own work rather than a publisher’s. If you run a filter on “above the 200-day”, your filter did not change state on Monday unless it is the simple one, and you cannot tell from any of the four publishers above which one you are reading unless you happened to pick the one that labels it.
Where the pair stands against the simple band this morning: 158.64 at the Tokyo eight o’clock fix, eleven pips above 158.53, in an eleven-pip session range of 158.58 to 158.69 against a New York close read at 158.61, with one desk putting the New York session range at 158.57 to 158.75. The 24 September high of 159.04, which one publisher names this morning as the level at which intervention talk would strengthen, is fifty-one pips above the band. The 50-day is eighty-three pips below it.
The 50-day did not move in three days, and that should bother you more
Both new publishers give the 50-day at 157.70. Same value, three days apart, to the pip.
A 200-day average moving one pip over three days is unremarkable; it is an average of two hundred observations and three new ones barely touch it. A 50-day average not moving at all over three days, in a week when the pair traded a 150-pip October range and set a fresh high, is harder to account for. The arithmetic runs the other way: the shorter window should be the more sensitive figure, and here it is the one that stood still while the longer one moved.
We are not calling either number wrong. We are saying the figure that should have been most responsive was the one that did not respond, which is the signature of a value copied forward rather than recomputed — and that you cannot separate a stale field from a flat market without the series, which neither page publishes. Key a filter to an average you read off somebody else’s page and you are trusting their refresh schedule as well as their convention and their arithmetic. The refresh schedule is the part nobody documents. Compute your own, or accept a dependency you cannot see.
Three more numbers from one overnight session with their reference missing
The same shape turned up three times more since the New York close, which is why this is a labelling problem rather than a moving-average problem.
Crude first. One publisher has West Texas settling up 2.06 per cent, having opened near 93 dollars and traded back into the 92 range after hours. Another has it up 1.04 per cent to near 94.40 dollars, rebounding from near 91 dollars in London and peaking near 95.50 dollars in the New York afternoon. Two percentage changes for one settlement on one day, a full percentage point apart, across levels spanning roughly four dollars fifty. Gold does it more quietly: down 0.61 per cent on one page, down 0.49 per cent to near 4,168 dollars on the other, after trading above 4,210 dollars overnight. A percentage change is a ratio of two prices at two times and neither page states either time.
Then the ten-year. One recap gives the United States ten-year yield near 5.24 per cent for Monday and says in the same breath that the cash bond market was closed for the holiday. That is not a measurement of Monday; it is the last thing the market said before it shut, carried forward. Which leaves the question we published yesterday still open: Monday’s break has no yield print behind it, and the market that produces one reopens today. One publisher put it plainly this morning — the market is waiting to see how long-term yields move now the holiday is over.
And a third, smaller and unresolved: one desk gives the euro–yen New York session range as 177.01 to 177.37 while another gives the New York close at 176.94, seven pips below the low of the session it is meant to close. One of those is wrong and we are not going to guess which. We log it because a close printed outside its own range is the cheapest internal-consistency check there is, and it takes two publishers to run it.
What this does not tell you
Start with our own error, because it is the largest thing on this page. An earlier version of this article, live for a few minutes before this one replaced it, opened by claiming that no publisher we read had printed the 200-day and treated 158.53 as a discovery. Our own 7 October article disproves it: we had three readings then, from three publishers, and we built a 52-pip convention finding on top of them the following day. We carried a shorthand note in our working file — nine runs without a two-decimal value — and wrote from the note instead of from the published record. The number was never missing. The label was, and that is a different and smaller claim.
Neither 158.53 nor 158.54 is our calculation, and we have still never computed a 200-day from a price series. Both are three days old, and the 50-day on the same two pages is the figure we have just told you to distrust, so treat the simple 200-day as a band near 158.5 and expect it to have moved. The exponential reading is our own 8 October figure and we have not re-read it; the claim that it has stayed below every daily close is as of that date, not as of this morning.
The crude and gold percentages are two publishers’ figures and the obvious reconciliation is that one quotes a settlement and the other a snapshot. We think that is probably what happened, and we are publishing it as a disagreement rather than a resolved difference because neither page says which — which is the argument. Monday’s yield question resolves today. And none of this is a view on direction: the payload is that a filter keyed to a level somebody else publishes inherits their convention, their precision, their refresh and their reference time, and on current evidence three of those four are invisible to you.
Related
- FX events calendar — the reopened cash market and this week’s inflation print.
- Signals — what a moving-average filter is actually keyed to.
- EA presets — convention and refresh settings for level-based filters.
