The 200-Day We Could Not Date for Fourteen Slots Is 158.71. Dollar-Yen Closed More Than a Hundred Pips Above It.

A dated 200-day, at the fourteenth asking, and it is barely outside our band

The value is 158.71, from a technical table credited to DZH Financial Research and carried by Gaitame.com, published 17:15 UTC on 11 November. The same table prints the previous business day at 158.69. A separate daily technical note from the same research house, published 02:03 UTC on 11 November, prints the 200-day line at 158.70. So the series reads 158.69, then 158.70, then 158.71 across two business days, and it drifts about 2.0 pips a day.

Three things follow, and the second is the one that matters.

First, our band was nearly right at its top edge. The dated value sits 1.0 pip above 158.70. Every distance we published against the top of that band was off by a single pip, which is not a scandal.

Second, and this is the real finding: the band was wrong as a width, and that was never visible from inside it. Ten pips of spread across twelve undated readings looked like vendor disagreement. It is not. At 2.0 pips of drift per business day, a 10-pip band is five business days of one series moving, bundled into a single number and presented as uncertainty. There was never 10 pips of doubt about the level on any given day. There was 10 pips of doubt about which day — and we had been treating a date problem as a precision problem for a fortnight. This is the same lesson that a 21-day average taught us when it moved 14.0 pips in 24 hours, except that the 200-day moves so slowly that the staleness is invisible until you finally see two consecutive values.

Third, a warning for anyone storing these. The same table prints the 90-day average’s previous value as 158.71 — the identical number to the 200-day’s current value. Two windows, one figure, in adjacent rows of one table. If you scrape that page you will get the right number attached to the wrong window roughly whenever the two cross, and nothing on the page will tell you. The 90-day currently reads 158.68, the 21-day 158.05, the 5-day 158.54.

What we still cannot do is reproduce any of it. Neither page prints the daily bars, so the average cannot be rebuilt from its own inputs, and neither page states an as-of date for the value — only its own publication time. We have a dated publication and an undated measurement, which is better than fourteen slots of nothing and is not the same as a derived level with its inputs shown.

The break, and the close that is not one number

The prior close was 158.75, printed as the previous close on the daily technical note and matching the figure a Tokyo summary gives for the previous New York close. Against the previous business day’s 200-day of 158.69, the pair came into Wednesday just 6.0 pips above the line. Then it ran.

The high is 159.94, three independent readings agreeing, described by two of them as the highest since 2 September. The session reference range is 158.67 to 159.94, or 127.0 pips. And the close is four different numbers:

• 159.79 — a Tokyo publisher’s New York summary, which we could only read at index level
• 159.84 — a wire wrap, explicitly labelled as near closing levels rather than a close
• 159.86 — a US commentary service’s macro table, with no time attached
• 159.87 — a Tokyo publisher’s 04:00 Tokyo-time reading, 19:06 UTC, which is a timed level and not a close at all

That is an 8.0-pip spread, and we are publishing it as a band rather than picking one, because picking one would be inventing a precision nobody printed. Note what the spread actually is: of those four figures, exactly one claims to be a close, one says it is not a close, one is a timed reading and one carries no time. The disagreement is not about the price. It is about what the word “close” means in a 24-hour market, and on a day the US bond market was shut for Veterans Day the question is sharper than usual.

Taking the band: the pair closed 108.0 to 116.0 pips above the dated 200-day, having gained 104.0 to 112.0 pips on the day.

Now score us. At 11:45 UTC on Wednesday this desk published that the volatility surface priced a 58.9-pip one-day move in dollar-yen, that the 159.04 breakout level sat 24.0 pips away — 0.41 of a daily deviation — and that a move of that size happens upward on about 34 per cent of days. We then explicitly refused to say which way. The level went, and it went by 90.0 pips at the high. The realised close-to-close move was 1.77 to 1.90 deviations; the realised range was 2.16.

We are not claiming that as a hit, and the refusal was still right. A 34 per cent one-sided probability resolving in the 34 per cent direction is not a forecast coming good, and nothing in a volatility surface told us a 1.8-sigma day was coming — that is the point of quoting one. What the surface did do, correctly, was price the level as noise rather than as a wall: 24 pips on a 59-pip day is not a barrier, and anyone treating 159.04 as a decision point was treating four-tenths of a standard deviation as information. The move that followed was far too big to have been predicted and exactly the shape the width implied was available.

The expiry board exists, it carries cut dates, and our explanation for seven missing sessions was wrong

Eleven hours ago, at 11:45 UTC, we published a conclusion with some confidence: that seven consecutive sessions without an option expiry board were not a sourcing failure but a scheduling fact, because the only publisher carrying strikes with sizes posts at about 12:40 UTC, 140 minutes before the New York cut, and our three runs begin at 23:00, 06:00 and 11:30 UTC. No run, we said, can read a board for a cut that has not happened. We called the fix a schedule change and told the next slot to stop scoring it against sourcing.

We withdraw that. A dollar-yen order board was published at 09:18 UTC on 11 November — two hours and twelve minutes before our pre-US slot opened — by a publisher our own source notes already credit with being the only one that publishes option cut dates per level. It was sitting there, readable, while we wrote that it could not be reached. The levels, read at index level because the article body would not open:

• 161.00 — selling, small; stop-loss buying above
• 160.40–50 — selling, small
• 160.00 — selling, slightly small, with New York option cuts on the 16th and 17th
• 159.80 — selling, small
• 159.40–50 — selling, small
• 158.50 — carrying a New York option cut on the 13th

The error was a generalisation, and it is worth naming precisely because it is cheap to repeat. Two dated URLs at two Western publishers failed. From two failures at two sources we concluded something about the structure of our own schedule — a claim about every slot, every day, forever — and the counter-example was a publisher we had already written down as the best in the file for exactly this object. A conclusion about a schedule needs more than two failed fetches, and the file’s own standing warning says a vendor convicted on consistency alone is convicted without evidence. We convicted a schedule on two data points.

What the board is worth, now that we have it. It is as of 09:18 UTC, so by the close it was more than twelve hours old and orders move; treat it as where the sellers were, not where they are. On that basis: price cleared 159.40–50 by 49.0 pips and 159.80 by 14.0 pips at the high, and stopped 6.0 pips short of 160.00. The close at the lower end of our band, 159.79, sits 1.0 pip under the 159.80 offer. And the only 160.00 optionality the board names expires on the 16th and 17th, not today — so if you were expecting a pin at the big figure this week, the board says the cut that could produce one is next week.

The sizes are qualitative bands only. This publisher gives “small” and “slightly small” rather than notional amounts, and its larger size bands sit behind a paid membership. We have strikes and cut dates and no notionals, which is a better object than we have had in seven sessions and still not a board you can size against.

The risk reversal: a sign we could not read, and a series that finally has a prior

At the pre-US slot we published a one-month 25-delta risk reversal for dollar-yen of −1.8775, said the page did not state its unit, refused to draw any inference from a single reading, and recorded it so that the next run would have a prior. The next run is this one, and the figure resolved in a way worth spelling out.

A second publisher, a Tokyo research house, prints the same object at 19:40 UTC on 11 November under an explicit heading — 25-delta yen call — with the unit stated as per cent, and gives a previous value of +1.88 and a current value of +1.60. Our figure was −1.8775. The absolute difference between the two previous readings is 0.0025.

So these are not two vendors disagreeing. They are one number quoted from two sides: a yen-call premium stated positive, and the same premium stated against the dollar-yen pair and therefore negative. The sign we could not interpret was a convention, not a value, and the unit we could not find is per cent. This is the same-object check paying off in its cleanest form yet — and had we published a direction from the first reading, we would have had a 50 per cent chance of getting the sign of the market’s skew exactly backwards.

With the prior in hand, the series says something. Every tenor narrowed: one month +1.88 to +1.60, a fall of 0.28 points or 14.9 per cent of its own level; three months +1.75 to +1.55; six months +1.34 to +1.20; one year +0.59 to +0.53. The premium for owning yen calls shrank across the curve on the day the yen lost more than a hundred pips, which is what you would expect and is worth having measured rather than assumed.

The volatility columns from the same table do something more interesting. Compare them with the surface we published on Wednesday morning, read at 07:31 UTC from a different publisher. At one year the two agree to 0.01 of a volatility point — 8.14 against 8.15. At six months, 0.01 again. At three months they are 0.14 apart, and at one month 0.17. The vendor disagreement is concentrated entirely in the tenors a single day can move, and vanishes in the tenors it cannot. That is not a vendor problem and we are not logging it as one; it is two different as-of times on a field that was repricing, and the tenor profile of the gap is the evidence.

What repriced confirms it. Across Wednesday, one-month volatility rose 0.49 points to 7.05, three-month 0.35, six-month 0.16, one-year 0.10. The front-end move was 4.9 times the back-end move. At 7.05 per cent the priced one-day deviation at the close is about 71 pips, up from the 58.9 we published in the morning. The market did not raise its estimate of where dollar-yen is going. It raised its estimate of how fast it gets there, and only for the next few weeks — which is the window that contains the December Federal Reserve meeting, 26 to 27 days out.

Producer prices: we pre-registered two legs and lost both

At 11:44 UTC on Wednesday this desk published a two-legged, deliberately losable pre-registration on Japan’s October corporate goods price index, due at 23:50 UTC. Leg one: the year-on-year rate prints between 6.17 and 6.83, being the 6.5 consensus plus or minus the 0.325-point mean absolute surprise we had computed from the series’ own history. Leg two: the month-on-month rate prints at or below +0.3.

The Bank of Japan published the release at 08:50 Japan time. The year-on-year rate is +7.1 per cent. The month-on-month rate is +0.7 per cent. Both legs lost, and neither lost narrowly.

• Year-on-year printed 0.27 points above the top of our band. The consensus asked for 6.5 against a 6.8 prior — a three-tenths fall. The rate rose three-tenths instead. The sign was wrong, not just the size, and the 0.6-point miss is 1.85 times the error bar we built the band from.
• Month-on-month printed +0.7 against a +0.3 consensus, a 0.4-point miss and 1.78 times that field’s own 0.225-point bar. Against the revised-actual bar described below it is 2.18 times.

Now the part worth more than the loss. Our central published claim on Wednesday was that the move the consensus was asking for — three-tenths on the year-on-year rate — was smaller than the series’ own typical forecast error, and that a volatility filter sized by how much a number is supposed to move should therefore wave this release through. That claim was arithmetically correct and it was the wrong question. The realised surprise was 0.6 points, exactly twice the move the consensus asked for. On the monthly field the realised change from the revised prior was 0.9 points against an asked-for 0.6.

“How much is this number supposed to move” and “how much can this number move” are different fields, and we sized a blackout off the first one. Our own standing note already says that vendor dispersion is not an uncertainty estimate and that the ratio between the two runs to about ten; we then went and built a rule on the consensus-implied move anyway, for a release that printed nearly twice it. The pre-registration was worth making because it was losable, and the loss is more informative than a win would have been.

And there is a second finding inside the release that we could only have got from the issuer. We discovered earlier in this run that this series carries a revision column we had never used, that the revisions run about the same magnitude as the forecast errors themselves, and that on the year-on-year field they were upward six times out of six. This release revises September’s monthly rate from −0.3 to −0.2 per cent, and its index level from 135.7 to 135.8. So:

• October’s month-on-month figure reproduces exactly from the issuer’s own index levels: 136.8 divided by 135.8 is +0.7364 per cent, which rounds to the +0.7 printed. We can reproduce this one from its inputs, which we could not do for the 200-day.
• Had September not been revised up, the same October index of 136.8 against 135.7 would have given +0.81 per cent — a printed +0.8. The upward revision of the base subtracted a tenth of a point from today’s headline.
• The revision therefore did not merely tidy history. It changed the number released this morning, in the direction that makes it look smaller, and nobody scoring the print against the consensus would see that.
• The issuer marks July, August and September as revised in this release. Three months deep, where the aggregator’s revision column shows one. Any error bar measured against first prints — including the one we published — is measured against a figure that can still move for at least three more releases.

The composition is cost-push and it is broad: petroleum and coal products added 0.20 points to the monthly rate, chemicals 0.15, electric power, city gas and water 0.15, iron and steel 0.11, food and beverages 0.08, against agricultural products at −0.28 and non-ferrous metals at −0.03. Five positive contributors against two negative is not one volatile line doing the work.

The currency connection is in the same document and it is the figure to take away from this slot. Import prices on a yen basis rose 3.3 per cent on the month and are up 24.8 per cent on the year. On a contract-currency basis the same imports are up 19.2 per cent. The gap — 5.6 points of annual import inflation that exists only because of where the yen is — is the exchange rate showing up in Japanese production costs, measured by the central bank, on the morning after the currency closed more than a hundred pips above its own 200-day average. Export prices show the same shape at 16.7 per cent on a yen basis against 12.2 contract-currency, a 4.5-point gap. The next release is 10 December, which is the day after the Federal Reserve meets.

What this does not tell you

• We cannot reproduce the 200-day from its own inputs. Neither page carrying it prints the daily bars, and neither states an as-of date for the value — only its own publication time. A dated publication is not a dated measurement, and we are not going to pretend it is after asking for one for fourteen slots.
• We cannot give you a single closing level. Four publishers give four figures across 8.0 pips and only one of them claims to be a close. Every distance in this piece is therefore stated as a band.
• The order board was read at index level, because the article body returned a client error on three attempts including a query-string retry. We have levels, size bands and cut dates and no notional amounts, and no independent confirmation of any of it.
• There is no Tokyo-session level in this piece, because at the time of writing no publisher we read had posted one for 12 November, and no Tokyo fixing had been published. We would rather print the gap than carry an overnight level forward and call it a session.
• We have no Nikkei or TOPIX close for Wednesday. Not one of the twelve newest items on the publisher we rely on for them carried either index.
• On intervention we can tell you that concern about Japanese official action near 160.00 was reported as capping the move, and that is all. The source refers to the tone of official remarks without naming an official, a date, or a quotation we could check. We are publishing the concern as a market condition and explicitly not attributing a statement to anyone.
• The only source we could read for the producer-price result was the issuer’s own release. The specialist indicator page that had scheduled the release to the minute was still showing an empty result cell more than a quarter of an hour after it, and the news index we rely on was a pre-release snapshot. We cannot tell you whether that is a cache or a stale page, because from outside they look identical.
• We cannot give you an October figure for the series the aggregator calls the previous revised value, because that page never updated. The revision figures here are the issuer’s own markings.
• We cannot tell you what the dollar index did in numbers. Three publishers said it rose and one said it reached its highest since April 2025 intraday; none of the three printed a level.
• December crude settled at $86.53, up $0.49, and that is the only oil price we read with its contract month attached. A Brent figure of 92.76 appeared in the same table with no contract month, so we are not quoting it as a price.

Related

• FX events calendar — Tokyo-session release times in UTC.
• Signals — how we size a level against the day’s priced range.
• Presets — thin-liquidity and news-window settings for the Asia session.