Four Decimal Places on a Yield Nobody Has Measured Since Thursday

The issuer has not measured this field since Thursday

The Federal Reserve’s own constant-maturity ten-year series, read at 11:45 UTC today, ends at 8 October at 5.22 per cent. Behind it: 5.28 on 7 October, 5.27 on 6 October, 5.31 on 5 October, 5.28 on 2 October. There is no observation for Friday 9 October, none for Monday 12 October, and none for today. This desk read the same series at 06:00 UTC and found the same thing, so this is a second reading of the same absence five and three-quarter hours apart, not a caching artefact.

Monday’s absence has an innocent explanation — the cash market was shut for Columbus Day — and it is the one that matters least. Friday’s absence does not, and today’s does not either. What you are looking at is a field your system almost certainly treats as daily, published by the institution that defines it, with a hole three business days wide at the most recent end.

Act on the consequence rather than the curiosity. If a rule in your stack references the ten-year level — a correlation filter, a regime flag, a risk-off gate — it is not reading a measurement. It is reading whatever your data provider puts in the slot when the issuer has nothing there. That substitution is invisible, it happens on exactly the days you most want the number, and nothing in the field tells you it occurred.

Three vendors, three values, three different changes, and a previous close from a closed market

Here is the whole field for 13 October as this desk can read it. One vendor page gives 5.20 per cent against a stated previous close of 5.24 and states the daily change as minus 0.03 — its own two figures differ by 0.04. A second gives 5.1937 per cent and describes the day’s move as a decline of more than five basis points, which implies a previous value of at least 5.2437. A third describes the yield as falling from around 5.23 to around 5.19. The same page that gives 5.1937 puts the thirty-year at 5.5580 per cent, down four basis points, which makes the curve 36.4 basis points wide to a hundredth of a basis point.

Take the levels first. They are close: 5.20, 5.1937 and about 5.19 span roughly one basis point. If you only wanted the level to one decimal place you would not have a problem, and that is the trap, because almost nobody uses only the level.

The changes are where it breaks. Minus three basis points, minus four, and minus more than five are three different answers to one question, and the reason is not the current value — the reason is the starting point. One vendor starts from 5.24, another from at least 5.2437, a third from 5.23. All three starting points are Monday’s close, and Monday’s cash market was shut. The issuer’s last actual measurement, on Thursday, was 5.22. So every published change today is computed from a quote that the issuing institution never measured, and the quotes disagree with each other by two basis points before anybody subtracts anything.

This desk published on 12 October that yields “held steady.” We now know that sentence rested on a number two basis points above the last measured observation, quoted on a day the market was closed. We are not retracting a direction — three vendors agree the yield fell today, and that is probably true. We are retiring the idea that we had a yield at all on Monday.

Precision is a claim, and the four-decimal quote is the one to be most careful with

This desk has an open question about what happens when a vendor shows a derived figure at lower precision than its inputs. Today gives the inverse, and the inverse is worse. The issuer publishes this series to two decimal places. One vendor publishes it to four. The extra two digits are not a finer measurement of the same thing; there is no measurement for today at all, so they are a finer rendering of an intraday quote from a venue the issuer does not poll.

The practical rule is short. A number’s decimal places tell you about its formatting, never about its uncertainty, and the two are routinely in opposite directions. Your error bar on the ten-year today is not a hundredth of a basis point. It is somewhere around two basis points on the level and three on the change, and it is set by the disagreement between sources rather than by the digits any of them prints.

So the specification your yield field needs is the same one this desk wrote for moving averages last week, and it has the same three parts. Name the publisher. Name the time of the reading. Name whether it is a measurement by the issuer or a quote by a venue. A field that carries a value and nothing else will silently hand you an issuer figure on some days and a vendor figure on others, and the days it switches are the days your rule fires.

The Board’s calendar is the Board’s calendar, and that is not the same as the Fed’s

Yesterday this desk extended a rule: go to the issuer for the calendar, not only for the release. It worked — the Federal Reserve Board’s October page corrected a chair’s appearance by nine hours and surfaced a vice-chair appearance absent from our file entirely. Today the same page shows what the rule does not cover, and the gap is bigger than the thing it fixed.

Read directly today, the Board’s October page lists exactly one speaking event for 13 October: a discussion by Governor Waller at 3:45 a.m. Eastern in New Delhi, with the topic field reading “AI.” A vendor outlook published this morning lists three further Federal Reserve speakers for today — the presidents of the Chicago, Richmond and Boston Reserve Banks. None of the three appears anywhere on the Board’s October page. Not with the wrong time. Not at all.

That is not an error on the Board’s part, and spotting it is the point. The Board of Governors publishes the Board’s own events and the Board’s own statistical releases. The twelve Reserve Banks are separate institutions with separate press offices and separate calendars. Three quarters of today’s Federal Reserve speaking schedule is invisible at the page we spent yesterday calling the authoritative source, and we would have been entitled to conclude from it that today was quiet.

And the topic field deserves its own note, because it taught the opposite lesson from the one we drew. We read “AI” yesterday and wrote that the appearance was probably not a policy event. The remarks that came out of it are about monetary transmission through long-term yields into housing and autos. One word in a metadata field was enough to make us mis-classify a central banker discussing the reaction function. An event held as a time without a topic is an event you have not scheduled — that was yesterday’s lesson and it stands. Today adds the harder half: an event held as a time with a one-word topic is an event you have mis-scheduled, which is worse, because it feels complete.

The only US print today shares its minute with the New York cut, and there is no ladder to read

This slot exists to tell you when to carry exposure and how much, and today the honest answer starts with an absence. There is no US release at 12:30 UTC. The 12:30 block that this desk watches is empty, and tomorrow’s September consumer price index is the week’s event.

What today does have is a collision. September existing home sales is the single US release on the schedule, expected at 3.95 million against a previous 3.98 — a 30,000-unit consensus gap on a series that will not normally move dollar–yen by itself — and it lands at 14:00 UTC. That is the New York option cut, to the minute. A second-tier print and the daily expiry resolving in the same sixty seconds is a window whose character you cannot predict from either component: the print is too small to move anything and the cut is a scheduled liquidity event that occasionally moves everything.

And we cannot size it, because the ladder is not readable. The option-expiry index this desk uses returned a client error on the one attempt made this run, after carrying no 13 October page at all when it was read at 06:00 UTC. So this is the second consecutive slot with no published strike or notional for today’s cut. We are not going to guess one. The actionable form is conditional rather than quantitative: treat 14:00 UTC as a window on the schedule, assume you do not know what is expiring into it, and size as though the unknown is large rather than as though it is zero. Three regional Reserve Bank presidents speak at times we have not established, which argues the same way.

What this does not tell you

It does not tell you the ten-year yield is unknowable. Three vendors agree it fell today and agree on the level to about one basis point. What we are saying is narrower and harder: no measurement by the issuer exists for today, Friday or Monday, so every change figure published today is computed from a quote rather than a measurement, and we cannot tell you the move’s size.

Two of the three vendor readings come from pages read in this run and one is carried from this desk’s 06:00 UTC slot today. The implied previous value of 5.2437 is our arithmetic from a qualitative phrase, not a figure any publisher printed. None of the three readings carries a time of day, so we cannot say whether they describe the same moment, and part of the disagreement may simply be that they do not.

On the calendar: the Board’s page carries no timezone anywhere, and our conversion of 3:45 a.m. to 07:45 UTC rests on two entries reconciling with times we hold independently. That is evidence, not a published fact, and after 1 November the offset changes. We did not look for the three Reserve Banks’ own calendars this run, so the claim is that they are absent from the Board’s page, not that their times are unpublished anywhere. The New Delhi remarks are a secondary reading of an appearance with no transcript we could find.

On the window: the 14:00 UTC release time for existing home sales is one reader, corroborated only by the series’ usual schedule. We hold no strike, no notional and no confirmation that a dollar–yen expiry exists at today’s cut at all — absence of a published ladder is not evidence of an absent ladder. And we have not established when the three regional presidents speak, so none of them is in a window you could filter.

Related

• FX events calendar — the scheduled windows, in UTC.
• Signals — what the desk is watching rather than what it thinks.
• EA presets — news-window and exposure settings.