The One Issuer That States Its Own Time Zone Has the Wrong Event Fifteen Minutes Before It.

Three issuers, four days, three different versions of one field

Run them side by side, because the pattern is the useful object and no single one of these is.

• The Ifo Institute, Saturday: a minute, and two zone labels against it — “10:30 CET/CEST” — in the one week of the year when those differ by an hour. We corrected our own table on the strength of it and the index then printed thirty minutes before the issuer’s stated time. The issuer was wrong about its own release.
• The Bureau of Economic Analysis, yesterday: a minute, 8:30 AM for Thursday’s advance GDP, and no time zone anywhere on the page. Not an ambiguous label; no label. The page cannot be converted to UTC at all without a second source, and on 1 November the same unchanged “8:30 AM” will mean 13:30 UTC instead of 12:30.
• The Australian Bureau of Statistics, today: a minute, a correct zone label, and an explicit sentence giving the convention for the whole page. Complete. Converts cleanly. Nothing to adjudicate.

The lesson is not that issuers are unreliable, which is too broad and we published a retraction on Monday for exactly that overreach. It is narrower and more usable: the completeness of an issuer’s time field is a property of that issuer, it varies enormously between agencies that are otherwise equally authoritative, and it is cheap to check once and store. The ABS page is worth a standing entry in your calendar sources. The BEA page is worth an entry too, with a flag saying it cannot carry an hour.

The complete one has a different defect, and it is in the field beside the clock

Fifteen minutes before that CPI, at 00:15 UTC, there is a Reserve Bank of Australia item. One Japanese schedule we read this morning lists it as a Governor’s speech. The Reserve Bank’s own calendar says it is nothing of the kind: it is an appearance before the Senate Economics Legislation Committee for supplementary budget estimates, by videoconference, with the Governor one of five Bank representatives attending, starting 11.15am AEDT.

For a human reader that is a footnote. For a news filter it is the whole thing, because the two objects have different shapes. A speech is a point event with a text and a finite duration; you pad it by thirty or sixty minutes and you are done. A parliamentary estimates hearing is open-ended, runs for hours, has five officials who can each say something, and is interleaved — here — with a quarterly inflation print landing fifteen minutes into it. A filter holding “speech, 00:15” unblocks at 00:45 and puts you back in the market while the Governor is still being questioned and fifteen minutes after the CPI.

This is the failure mode this desk keeps finding in a new field each week. The clock was right and the event type was wrong. Last week it was a consensus figure sitting in the wrong field; yesterday it was a policy rate published without saying which instrument it was. Today it is a calendar row whose time survives scrutiny and whose noun does not. A system stores the row. Nothing in the row tells it which part to doubt.

Twenty-six events, eighty-seven hours, and widening your filter buys you almost nothing

Count the week properly. From 06:00 UTC today to 21:00 UTC on Friday is eighty-seven hours, and this desk’s calendar carries twenty-six scheduled items in it across twenty-five distinct timestamps — four central bank decisions, an advance GDP print, two Treasury auctions, two quarterly inflation releases and the rest. We computed what a symmetric news filter does to that week.

• Padding every event by thirty minutes either side blocks 21.00 hours, 24.1 per cent of the week, in fourteen separate windows.
• Padding by sixty minutes blocks 34.17 hours, 39.3 per cent — and leaves thirteen windows.
• Padding by ninety minutes blocks 44.67 hours, 51.3 per cent, in eight windows.

Look at the second line again. Doubling the pad costs you thirteen more hours of the week and removes exactly one window. That is the structural point and it is not obvious in advance: these events are clustered tightly enough that a wider filter mostly fuses windows you already had rather than catching anything new. Between thirty and ninety minutes you lose another 23.67 hours of market to eliminate six re-entries. Whether that trade is worth it depends entirely on whether your losses come from being in the market at the wrong minute or from the churn of stopping and restarting — and most people have never measured which.

If you only ever pad one thing this week, pad Wednesday’s Asia hour. The Senate appearance at 00:15, the Australian CPI at 00:30 and the Reserve Bank of New Zealand at 01:00 are three events in forty-five minutes, in the thinnest liquidity of the day, and no symmetric pad separates them into distinct windows — at thirty minutes they merge into one 105-minute block running from 23:45 Tuesday.

The quietest stretch of the week starts the moment the Fed stops talking

The same computation gives the clear gaps, and the longest one is counter-intuitive enough to be worth stating on its own. At a thirty-minute pad the biggest uninterrupted stretch in the week is seventeen hours, from 19:00 UTC on Wednesday — the end of the Warsh press conference window — to 12:00 UTC on Thursday, when the pad around the advance GDP opens. The second longest is 11.75 hours, from 01:30 Wednesday to 13:15, covering the whole European Tuesday night and the London morning.

So the week is not evenly difficult. It is a quiet Tuesday morning, a dense American Tuesday afternoon, a brutal forty-five minutes in Asia on Wednesday, a European and American Wednesday afternoon that includes the Fed, and then seventeen hours of nothing at all before Thursday’s GDP. A schedule that treats “Fed week” as uniformly dangerous is wrong about more than half of it.

One correction to our own prior work while we are here. The two remaining coupon auctions now have sizes: the five-year today at 17:00 UTC is seventy billion dollars and Thursday’s seven-year is forty-four billion, with a twenty-eight billion dollar two-year floating-rate note on Fed day. This desk published “amount unsourced” against all three. The five-year figure is two-sourced — a Japanese schedule gives it as 700 oku dollars at 26:00 Japan time, which is 17:00 UTC, and an English wire carries the full Treasury announcement. The seven-year and the floating-rate note rest on that one announcement.

What this does not tell you

• The flat-hours figures are arithmetic on our event list. A different calendar with fewer speakers or more second-tier data gives different numbers, and the twenty-six items include several this desk judged worth a window that you may not.
• We did not test whether padding improves results. We computed what it costs. Those are different questions and only one of them is answerable from a calendar.
• The Bank of Japan time on Friday is still approximate. Its own schedule says around noon Japan time and gives no minute, so our roughly 03:00 UTC is a window, not a timestamp, and it is the one event this week we cannot pad precisely.
• The Reserve Bank of Australia item is a scheduled committee appearance, not a market event with a release. It may produce nothing at all. We are arguing about how a filter should classify it, not predicting that it moves anything.
• The event-type error is one observation at one publisher. It is not evidence that Japanese schedules are generally wrong about event types, and that publisher was right about every time we checked, including four American times where an English-language aggregator was an hour off.
• The seven-year and floating-rate note sizes are single-sourced to one wire’s reading of the Treasury announcement. We did not reach the announcement itself.

Related

• FX events calendar — this week’s decisions and prints in UTC.
• Signals
• EA presets
• The BEA Publishes a Minute and No Zone — the second of the three issuers above.
• We Corrected 09:00 to 09:30 for the Ifo. It Printed at 09:00 — the first, and our own retraction.