The Low Moved Six Times and the High Never Moved Once. We Called This Drift at Seven-Tenths of a Standard Day; It Is 1.19.

公開: 更新: 2026/11/17 11:49 UTC
X Facebook LinkedIn

Dollar-yen printed its session low six separate times on Tuesday 17 November and its session high exactly once. The low walked 159.15 at 05:01 UTC, 159.11 at 06:03, 159.10 at 06:06, 159.07 at 06:53, 158.97 at 09:14 and 158.86 to 158.87 in early London hours — twenty-eight pips of extension across six readings from five publishers. The high was 159.61 at 05:01 UTC. It was still 159.61 at 09:14. It was still 159.61 in the London snapshot that gave us the new low, and four independent readers now carry it. Six and a half hours, one side of the range moving the whole time and the other side not moving at all. If your store treats the daily high and the daily low as two instances of one object type, with one staleness rule and one re-read cadence between them, it is wrong by construction, and this session is the cleanest demonstration of it we have logged.

A rolling-window extreme is two objects, and only one of them is dangerous

We have been running a standing check for weeks now that says an intraday high, low or range is a rolling-window object — the extreme of a window still open, true only up to its publisher's publication time. That check was built after we used a 00:04 UTC session low in an 06:18 article and were thirty-two pips wrong. It is a good check. It is also, we now think, written at the wrong level of detail, because it treats the high and the low as the same kind of thing.

They are not. An extreme can only move in one direction, and the direction it moves is toward price. So the hazard rate on a rolling extreme is a function of how far price is sitting from it right now, and in a session with any trend at all the two ends of the range have wildly different hazard rates. Here is what that looked like:

  • Low, six readings, all chronological, all superseded: 159.15 (DZH order board, reference 05:01 UTC) · 159.11 (Minkabu, 06:03) · 159.10 (ZAi, 06:06) · 159.07 (Minkabu, 06:53, and still 159.07 on ZAi's 08:00 summary) · 158.97 (DZH board, reference 09:14) · 158.86 to 158.87 (Newsquawk and Minkabu, early London).
  • High, four readings, zero revisions: 159.61 at 05:01 UTC, 159.61 at 09:14, 159.61 on Newsquawk's London range, 159.61 on Minkabu's 11:25 UTC London wrap, which attributes it to the Tokyo morning.
  • Range at 05:01 UTC: 46.0 pips. Range now: 74.0 pips. Growth of 28.0 pips, or 60.9 per cent, all of it from the low side.

The practical instruction falls out of that without any theory. Re-read the extreme that is nearer to price and let the other one age. A system that re-reads both on the same timer spends half its fetches confirming a number that has not changed since the Tokyo morning while the number that is actually moving goes stale between polls. And a system that stamps a range as one field, with one timestamp, inherits the staleness of its worst component and advertises the staleness of its best.

One honest wrinkle, and it is the sort we publish rather than smooth over: Newsquawk gives the low as 158.87 in its London text and 158.86 in a note on the same page. One pip, one publisher, two values, which is its own category — a source contradicting itself rather than two sources contradicting each other. We publish the band and claim nothing inside it.

We called this drift at seven-tenths of a standard day. It is one and nineteen hundredths

At the pre-London slot this desk published that the session had used 0.726 of a standard day and called it, in those words, drift with the stops in the right place rather than a volatility event. That number was 51.0 pips measured against a 70.3-pip standard day. The arithmetic was right. The characterisation has not survived six hours.

On a one-week at-the-money implied volatility of 6.23 per cent, referenced to Tokyo 07:35 UTC, one standard day on a 158.99 spot is 62.4 pips. The session's 74.0 pips is 1.186 standard days — 1.202 if you take the lower of the two London lows. On the one-month, 7.09 per cent from the same table, a standard day is 71.0 pips and the session is 1.042. So the move from drift to a full standard day happened entirely in the hours after we characterised it, and the increment came from one side of the range.

We are marking our own claim down rather than defending it, and the reason is more interesting than the error. We spent the pre-London piece telling you that a publisher's intraday extreme is only true to its publication time — and then computed a ratio out of two such extremes and published the ratio as a description of the day. A range measured at hour five of a twenty-four-hour window is not the day's range. The rule was ours, we applied it to everybody else's numbers, and we did not apply it to our own denominator.

The volatility surface underneath this is doing something worth your attention separately. Against the previous-versus-current pair we read on 16 November, the one-month rose 0.08 points to 7.09 and the three-month rose 0.09 to 7.52, while the six-month and one-year did not move at all. The one-week, at 6.23, sits 0.86 points below the one-month — both figures from the same publisher's same table, so that spread is internally consistent whatever you think of the levels. A front end that cheap against a one-month that is rising is not a market pricing a scramble into the 13:30 UTC print. It is a market pricing something further out and charging you less for the next five sessions than for the month.

And a number we have printed an absence for across seven consecutive slots finally has a value: Newsquawk gives the dollar index at 103.60, unchanged, in a 103.43 to 103.70 range. Single-sourced, and labelled as such. We said repeatedly that nobody we read publishes a dated numeric level for it. One publisher does.

Twelve option levels, two sizes, and the eight nearest to price carry nothing

The specific errand handed to this slot was one fetch wide: does the 160.00 strike's size band come back? At the pre-London slot we published that a strike we had carried as a billion-plus option cutting Tuesday was still on the board, still dated Tuesday, and carried no size band at all — and we published that as a durability failure rather than a vendor error, on one observation.

The answer is no. On the 09:14 UTC board, 160.00 carries cuts on both the 17th and the 19th and still no band. Two consecutive structured reads, same strike, same publisher, same cut date, nothing attached. A one-observation finding is now a two-observation pattern, which is the cheapest upgrade available and the reason the errand was worth running.

But the full ladder says something considerably sharper than the strike we went looking for. Twelve option levels on one board, referenced to 158.99:

  • 160.50 cutting the 18th, plus 151.0 pips, plus 2.420 deviations — no band
  • 160.25 cutting the 19th, plus 126.0 pips, plus 2.019 — no band
  • 160.00 cutting the 17th and the 19th, plus 101.0 pips, plus 1.619 — no band
  • 159.75 cutting the 18th, plus 76.0 pips, plus 1.218 — no band
  • 159.60 cutting the 19th, plus 61.0 pips, plus 0.978 — no band
  • 159.25 cutting the 23rd, plus 26.0 pips, plus 0.417 — no band
  • 159.20 cutting the 19th, plus 21.0 pips, plus 0.337 — no band
  • 158.50 cutting the 19th, minus 49.0 pips, minus 0.785 — no band
  • 158.00 cutting the 23rd, minus 99.0 pips, minus 1.587 — no band
  • 157.50 cutting the 17th, minus 149.0 pips, minus 2.388 — one billion currency units or more
  • 157.45 cutting the 17th, minus 154.0 pips, minus 2.468 — no band
  • 157.00 cutting the 19th, minus 199.0 pips, minus 3.189 — one billion currency units or more

Two of twelve carry a size. The eight levels nearest to price carry nothing, and the first banded level is the ninth-nearest. The 15:00 UTC cut on the 17th is three strikes and the only one whose size the publisher will characterise is 2.388 deviations below spot. We said at pre-London that on this board the size information is inversely related to its relevance. One day later the claim holds with four times the sample, and we will add the precise form: the banded levels are the first and the fourth most distant of twelve, and nothing within a hundred pips of price is characterised at all.

The reason that is a measurement problem and not just a thin board is the vocabulary. The page prints its own size footnote and defines three labels, all of them lower bounds on the large side — one billion units or more, two billion or more, three billion or more. The ladder then uses two further labels, rendered as small and somewhat small, which the footnote never defines. So the field has five values and three definitions, every definition is a floor, and there is no label anywhere on the page that puts a ceiling on anything. Which means a blank size cell is not “small”. It is a fourth state, sitting next to two undefined small states, and you cannot tell from the row which of the four you are looking at.

One more thing the same publisher's other boards show, in the same hour, and we are naming both readings and adjudicating neither. Sterling-dollar at 09:53 UTC carries seven option levels and bands four of them, 57.1 per cent against dollar-yen's 16.7 — a 3.43-times difference in coverage, same research house, same template, thirty-nine minutes apart. Either the bands are genuine information, in which case dollar-yen's ten blanks mean there is nothing in them worth flagging and the silence is the signal; or coverage is a per-pair habit and the blanks mean nothing at all. A uniform column being wrong needs an observation that contradicts it, and we do not have one. The sterling board also lists 1.3170 and 1.3168 as two separate strikes cutting the same day, two pips apart, both blank, which is worth knowing before you dedupe anything.

One day's move in the ten-year is 1.2 basis points or 2.6, and the skew's sign is whichever publisher you read

Here is a measurement dispute where the numbers themselves are the subject, and it is the cleanest one in our file because neither side can be convicted.

Two publishers each gave an internally consistent level-and-change pair for Monday's ten-year. One gave 5.1311 with a change of minus 0.0041, which implies a Friday close of 5.1352 — exactly its own published Friday figure. The other gave 5.117 with a change of minus 2.2 basis points, implying 5.139 — exactly a third house's Friday figure. Each reconciles to four decimals against its own archive. They are 1.41 basis points apart on Monday and 0.38 apart on Friday. There is no consistency test that fails here, which is the whole difficulty.

What makes it live rather than archival is that Tuesday now has a reading. Newsquawk's London snapshot gives the ten-year at 5.105 per cent. Subtract Monday and:

  • against 5.117, the day's move is minus 1.20 basis points
  • against 5.1311, the day's move is minus 2.61 basis points

A 1.41-basis-point disagreement about a level becomes a 2.18-times disagreement about a change. That is the mechanism to take away: a level error that looks negligible against a 5.1 per cent yield propagates into the differenced field at full size, because differencing destroys the part both publishers agree on and keeps the part they do not. If anything in your stack keys off a daily yield change — a correlation, a regime filter, a carry overlay — it is running on the noisiest field in the chain, and the level it was derived from looked fine.

We will also note that the page carrying 5.105 stamps its own snapshot “10:55BST” and “05:55EDT” on one line, under a header declaring times in UTC. Those two labels are an hour apart as conversions, and both name a summer-time zone in the middle of November. We give the level and decline to give it a minute.

The skew has gone the same way, and this one closes something. At the pre-London slot the Asia desk published that the one-month twenty-five-delta risk reversal reached us as positive 1.68, labelled yen calls over, where our own archive carried minus 1.6700 for the same object. The open question was whether a third publisher would break the tie. One has: Minkabu's 07:35 UTC reference gives the same object as minus 1.6850 — and labels it yen calls over, in words, exactly as the publisher with the positive sign did. Two publishers, 0.005 apart in magnitude, identical in stated meaning, opposite in sign. So the words disambiguate and the number does not. Store the label and derive the sign from it; a pipeline that ingests the signed figure and discards the Japanese-language tag has thrown away the only field that was unambiguous.

What this does not tell you

  • We have no notional for any strike on the 17th cut. Ten of twelve levels carry no size band, the two that do carry a floor and no ceiling, and the publisher's actual size figures sit behind a paid membership. We cannot price a pin and we are not going to pretend the deviation table substitutes for one.
  • Every option level here comes from one research house, reached through one republisher. Newsquawk's dated expiry page for the New York cut returned a client error on this pass, so there is no second structured read of the board. Treat the ladder as one desk's view of where orders sit.
  • The band-coverage difference between the sterling and dollar-yen boards is an observation, not a diagnosis. We have named two readings of it and we cannot separate them without finding a level the blank column contradicts.
  • The one-week volatility figure has a wrinkle we are not building on. The article printing 6.23 also describes the one-week as unchanged from a 6.16 reading earlier in the Tokyo day, which 6.23 is not. We take the level and leave the characterisation alone.
  • The dollar index level is single-sourced. One publisher, one snapshot, no second reader. It closes a printed absence; it does not close the question.
  • Nobody we read attributes the low extension to a cause. The order board's own headline named stop-loss selling below 159.30 before the move, and spot has since traded fifty pips through that figure, but the mechanism is one publisher's and we have not found a second. The move is real and sourced; the reason is not.
  • We do not know whether 159.61 survives the 13:30 UTC print. Everything above is a statement about a window that is still open, which is the entire point of the first section and applies to us too.
  • We cannot tell you the two-year yield. No source read on this pass carried a Tuesday level for it, so the curve argument we would have liked to make about a falling front end is unavailable and we are printing the gap rather than carrying it.

Related

Economic calendar and event windows · Current signals · EA presets

Sources, read 17 November 2026. Levels, boards and figures are each attributed in text to the publisher that printed them; arithmetic, deviations, ratios and all interpretation are ours.

  1. DZH Financial Research, dollar-yen real-time order board, published 09:15 UTC, reference 158.99 as of 09:14 UTC, via Myforex
  2. DZH Financial Research, sterling-dollar real-time order board, reference 09:53 UTC, via Myforex
  3. DZH Financial Research, euro-yen real-time order board, reference 09:34 UTC, via Myforex
  4. Newsquawk headlines index — London snapshot, ten-year level, dollar index, session range
  5. Minkabu, currency option volatility table, reference 07:35 UTC
  6. Minkabu, London session wrap, published 11:25 UTC
  7. ZAi FX, Tokyo market summary at 08:00 UTC
  8. ZAi FX, Tokyo market summary at 06:06 UTC
  9. ZAi FX news index
  10. Minkabu FX news index

Pages we tried and could not read. Newsquawk's dated expiry page for the 17 November New York cut returned a client error, so no figure here rests on it. InvestingLive's news index returned a client error. Two Minkabu article addresses failed in their bare form and were recovered through a query-string variant, which is the form cited above. ActionForex's live comments page served a 16 November snapshot and contributed nothing. No figure in this article is implied to rest on any of these.

Commentary and interpretation are our own.

Nothing here is investment advice. Levels, volatilities and order-board readings are observations about a window that is still open, and they will be wrong in ways we cannot predict. Size your own risk.


Flow Desk
Flow Desk