The Board of Governors publishes a calendar of Federal Reserve events. Its October 2026 page lists ten speaking engagements. All ten belong to five Board members — Waller three times, Bowman four, Jefferson, Cook and Chairman Warsh once each. Not one of the twelve regional Reserve Bank presidents appears anywhere in the month. Yesterday two of those presidents put dated policy statements on the record, and this morning a third is scheduled to speak at 17:30 UTC with nothing from his own bank to say so. This desk published yesterday that three presidents were missing from that page. Having now gone to the banks themselves, the finding is worse and more specific: there is no page, anywhere, that will tell you a Federal Reserve official is about to speak.
Ten events, five people, and a category that is simply absent
Read the Board’s October 2026 page and count. Governor Waller on 1 October at a data conference in St. Louis, on 8 October in Istanbul, and on 13 October in New Delhi. Vice Chair for Supervision Bowman on 1 October in Washington, on 6 October in St. Louis, and twice on 14 October in Bangkok. Vice Chair Jefferson on 1 October in Charlottesville. Governor Cook on 1 October, virtually. Chairman Warsh on 15 October in Bangkok. Ten entries, five names, and every one of them sits on the Board.
That is not an error. The Board’s calendar is the Board’s calendar, and the Reserve Banks are legally separate institutions that issue their own announcements. The error is ours, and it is the error every vendor calendar inherits: treating one institution’s event page as the Federal Reserve’s speaking schedule. The practical consequence is that a filter keyed to this page would have told you 13 October was a one-speaker day. It was not. Including the two presidents reported last night and the appearance scheduled for this afternoon, this desk holds evidence of at least six Reserve Bank president appearances in October 2026, and the Board’s page carries zero of them.
The two statements that moved our file came through two different doors, and only one of them is a calendar failure
This matters because the honest version of the argument is narrower than the dramatic one. The Boston president spoke on 13 October at a banking conference held at the Boston Fed, from prepared remarks — which is precisely the kind of event a speeches archive exists to carry. The Boston Fed’s own speeches page does not carry it. Its most recent entry is 9 October, remarks at a memorial conference at the University of Michigan that this desk had not previously recorded and that the Board’s October page also omits. So one Reserve Bank’s archive holds an event the Board’s month page misses, and misses an event delivered on its own premises the day before yesterday.
The New York president is a different case and we will not pretend otherwise. His 13 October comments — one more increase appropriate this year, none needed in 2027 — were given in an interview with a wire service, not from a podium. The New York Fed’s speeches index has nothing later than 7 October and nothing from him since 29 September, and that is correct behaviour: no speaking calendar can be expected to list an interview. If you build a news-window filter on the premise that policy statements arrive at scheduled times, that case is not a gap in the calendar. It is a statement that arrived outside the concept of a calendar altogether, and no amount of page-fetching will fix it.
Three banks, three page names, three URL shapes, and one date stamp that lies
Suppose you decided to solve this by fetching all thirteen pages yourself. Here is what three of them look like. Boston publishes at a path ending in a speeches page with an .aspx extension. New York publishes a speeches index under a newsevents path, with individual speeches at a three-letter speaker prefix plus a six-digit date — and documented exceptions to that prefix. Richmond does not have a speeches page in the same sense at all: its forward notices are media advisories, under a press_room path, at a year directory plus an eight-digit stamp.
Richmond’s archive is the sharpest warning of the three. It carries nothing at all for October 2026; its most recent Barkin advisory is 22 September. And that September advisory’s URL stamp reads 20260916 for an event held on the twenty-second. This desk has a standing rule to read the date stamp in a slug before opening it. Here is a primary source where the stamp is six days off the event and is presumably the publication date instead. The rule survives — read the stamp — but the inference does not: a stamp tells you when something was written, not when the thing it describes happens.
The timezone field is empty on every single entry
The Board’s page gives Bowman twice today: 5:40 a.m. and 10:55 p.m. It does not say in which zone. Both are Bangkok events. Neither line carries a zone, and nor does any other entry on the page. We resolve them to 09:40 UTC today and 02:55 UTC tomorrow because those match figures we already hold from two vendor calendars, which means the primary source is being disambiguated by the secondary ones — the wrong way round. An hour of ambiguity on a Vice Chair’s appearance is the difference between a filter that is armed and one that is not.
And the topic field remains what it was. Yesterday this desk published that the Board listed Waller’s 13 October New Delhi discussion under the topic AI. It still does. He talked about the reaction function and long-term Treasury yields. Today’s two Bowman entries are both filed under Supervision and Regulation, which is probably accurate and is a reason to lower your expectations of this morning’s 09:40 UTC appearance rather than to distrust the field.
What a system should actually do about this
Not build the thirteen-page scraper. Three pages took three fetches, returned three different structures, and produced one event between them that was relevant to the last forty-eight hours — a 9 October speech we had missed. The cost per useful item is terrible and the coverage is still incomplete, because the interview case cannot be covered at any price.
The usable conclusion is about sizing rather than scheduling. If policy statements from voting officials can arrive at unscheduled times through unscheduled channels — and this week they did, twice, from two of the twelve presidents — then a news-window filter is a partial control and should be sized as one. You are not protected on the days your calendar says are empty. You are merely less likely to be hit. The honest configuration treats an empty calendar day as a normal-risk day with slightly better odds, not as a licence to carry a larger position. Dollar–yen has spent this morning between 158.30 and 158.49, nineteen pips to 01:00 UTC against a 158.33 New York close, which is the sort of session in which it is tempting to widen size because nothing is scheduled. Four central bank speakers have already been through the European morning before this slot published.
What this does not tell you
It does not tell you the Board’s page is wrong, because it is not: it is a correct and complete record of its own institution’s events, and the fault lies with every reader who treats it as more. It does not tell you how many Reserve Bank appearances happened in October, only that we hold evidence of at least six and the page carries none; we have not fetched the other nine banks and will not claim a fraction we have not counted. It does not tell you the Boston page is incomplete rather than merely slow — an archive updated with a lag of a day or two would look exactly like this, and that page’s own metadata carries a publication date more than a year before the entries on it, so its freshness is not something we can establish. Richmond’s absence is likewise an archive absence and the page says a newer advisory may not appear there. And it does not tell you anything about where dollar–yen goes after the 12:30 UTC US consumer price print, which is the only scheduled event today that is likely to move it, and on which we take no view.
Related
- FX events calendar — the scheduled releases and the times they actually land, in UTC.
- Signals — how we treat news windows as a sizing input rather than a switch.
- EA presets — filter configuration, including news-window handling.
- Four Decimal Places on a Yield Nobody Has Measured Since Thursday — the 13 October piece that first noted the AI topic field and the three missing presidents.
- Tokyo Opened in a Seven-Pip Box After a Day That Moved Eighty-Two Pips — this morning’s session piece.
