We Said the Issuer Was Silent. It Had Published Three Hours Earlier.

The retraction, and what the issuer actually did

Here is the state we recorded at 23:10 UTC on 14 October: most recent observation 9 October at 5.24 per cent, nothing for 12, 13 or 14 October, last updated 13 October at 20:17 UTC, stated next release date 14 October. We wrote that the issuer had let its own date pass. Here is the state this morning: most recent observation 13 October at 5.21 per cent, a dot rather than a number against 12 October, nothing for 14 or 15, last updated 14 October at 3:16 p.m. CDT, stated next release date 15 October.

Convert the update stamp and the problem is plain. 3:16 p.m. CDT is 20:16 UTC. Our read was at 23:10 UTC. The issuer published its 13 October observation two hours and fifty-four minutes before we opened the page and told you it had not published. The gap between the two update stamps we have now recorded — 13 October 20:17 UTC and 14 October 20:16 UTC — is twenty-three hours and fifty-nine minutes, which is about as close to a fixed daily cadence as a public series gets. The issuer was running on time. We were reading something else.

One thing survives from yesterday, and it is worth keeping. Against the issuer’s 13 October value of 5.21 per cent, the vendor close this desk carried for the same date was 5.2168 — 0.68 of a basis point apart. The vendor snapshots have been good. What has been bad is our inference from the absence of an issuer observation, twice in two weeks, in opposite directions: last week we treated the newest visible observation as the newest existing one, and yesterday we treated a stale serve as the issuer’s silence.

And the 14 October observation still does not exist on that page, against a vendor research note putting the ten-year at around 5.25 per cent that afternoon and a Tokyo desk note this morning putting it near 5.10 — eleven basis points below the issuer’s last printed value. We are not going to tell you today whether that gap is real. The whole point of the paragraphs above is that we cannot tell from one read.

A session low that rose seven pips

The same publisher’s Tokyo desk put out running range notes through this morning’s session. Take two of them, 195 minutes apart, and line up the six high and low fields.

• Dollar–yen high: 158.33 at 10:34 JST, 158.33 at 13:49. Unchanged.
• Dollar–yen low: 157.65, then 157.58. Fell seven pips. Legal.
• Euro–dollar high: 1.1219, then 1.1241. Rose twenty-two pips. Legal.
• Euro–yen high: 176.89, then 177.23. Rose thirty-four pips. Legal.
• Euro–dollar low: 1.1149, then 1.1153. Rose four pips.
• Euro–yen low: 176.45, then 176.52. Rose seven pips.

A running session low is monotone. It can fall and it can stay where it is; it cannot rise, because the session that produced the earlier figure is inside the session that produced the later one. Four of the six fields behaved. Two did not, both of them lows, both on the euro, both on the same publisher’s page, within one session.

This is the fourth time this desk has logged one publisher running two versions of the same number, and it is a new shape. The previous three were two surfaces disagreeing with each other. This is one surface disagreeing with itself across time, in the direction that is arithmetically impossible, on a field that readers quote as though it were an observation rather than an accumulation.

We are not going to guess the cause and we are not going to name a correct euro low. What we will say is operational: a running high or low is not a fact until the session it describes has closed, and a system that pins a stop or a bias to an intraday extreme read off a vendor page is pinning it to a figure that is still being written.

Every Fed speaking time in today’s schedule came from a broker

Three Federal Reserve officials are on today’s schedule as a Japanese broker publishes it: a governor at 02:55 UTC, the Chicago president at 14:00, the Cleveland president at 20:30. Two of those marks have been sitting in this desk’s open file for a week. We took them this morning.

The Chicago Federal Reserve Bank publishes no calendar of upcoming events on either of the two pages we read. Its media-appearances list runs backwards from 4 October; its newsroom page’s most recent dated item is 6 July. Neither page carries a single clock time, and neither carries a timezone. Its president speaks today, and there is nothing on the institution’s own pages that says so.

The Boston Federal Reserve Bank does list its president’s speeches, including two in October — 13 October in Boston and 9 October at the University of Michigan. It gives dates and locations and no times. It gives no timezone. It does not label any entry as upcoming or past, so you cannot tell from the page which of them has happened. The page’s own metadata publication date is in May 2025 while it lists events through October 2026, which is the same class of problem as the one in the first section: the surface’s stamp and the surface’s contents are describing different moments.

So the count for today stands at three Federal Reserve speaking times in your calendar and zero of them obtainable from the institution holding the event. That is now seven published readings this desk has filed with no issuer-sourced as-of time behind them. The uncomfortable version, which is the useful one: the timestamp column in your economic calendar is not data from the issuer. It is a vendor’s editorial estimate, and for speeches it is frequently the only estimate in existence.

What to actually change

None of this is a market call, and we are not going to convert it into one. It is three instances of the same mechanical fault, and the fault is that your system treats a read as a state.

Three rules fall out of it, and all three are about timing and sizing rather than direction. First: a filter that fires on the absence of a release — no print, therefore no event, therefore trade — needs two reads separated by more than the issuer’s update cycle before it is allowed to conclude anything. Here that cycle is just under twenty-four hours, which means a single read can be almost a full day behind and look completely normal. Second: an intraday extreme pulled from a vendor page during the session is a provisional field; if your sizing depends on it, the sizing is provisional too. Third: a news window built around a central banker’s speech is built on a vendor’s guess at the time, so the window should be wider than the vendor’s precision suggests — and when the vendor’s calendar and your own disagree on a speaker, neither of you has a primary source to appeal to.

What this does not tell you

It does not tell you the issuer’s page was served stale rather than updated late by its own clock. Both explanations fit what we can see, and we cannot distinguish them from outside. What we can say with certainty is narrower and is the only thing we have published: the page now carries an update stamp that precedes our read by 174 minutes, and our read did not show that update. We published a claim about the issuer and we should have published a claim about our reader.

It does not tell you the euro lows are wrong rather than the euro highs. We have two readings of one running field and arithmetic saying they cannot both be complete. Which of them is the defective one is not visible to us, and we have deliberately not published a corrected level.

It does not tell you the Chicago or Boston reserve banks fail to publish times anywhere. We read two pages at one and one page at the other. A press advisory we did not find may carry a time; the claim is about what is reachable from the obvious entry points, which is what a calendar builder would actually use.

It does not tell you the ten-year is at 5.10 or at 5.25. One reading of each exists, from vendors, and the issuer is silent on both dates. And nothing here is a view on dollar–yen, on Gilts or on anything else. It is a note about the quality of the inputs.

Related

• Economic calendar and event windows
• Signals
• EA presets
• The Market Doubled October on One Print. December Did Not Notice. — the article whose closing section this one retracts.
• The Option Market Priced a Sixty-One-Pip Day — the previous correction on this same series.