Four PMIs in Seventy-Five Minutes. The One Your Filter Carries Was the Least Informative of the Four.

The cluster, with every surprise measured

Ten fields printed in seventy-five minutes. Actual against consensus, with the September prior in brackets:

• 07:15 France — manufacturing 49.3 vs 50.5 (50.6), surprise −1.2; services 48.4 vs 50.7 (51.2), −2.3; composite 48.4 vs 50.5 (51.1), −2.1.
• 07:30 Germany — manufacturing 54.2 vs 53.5 (53.9), plus 0.7; services 51.7 vs 52.5 (52.9), −0.8; composite 52.8 (53.8).
• 08:00 euro area — manufacturing 52.4 vs 52.7 (52.9), −0.3; services 51.7 vs 52.5 (53.0), −0.8; composite 51.9 vs 52.7 (53.1), −0.8.
• 08:30 Britain — manufacturing 52.8 vs 51.7 (51.9), plus 1.1; composite 53.1 vs 51.8 (52.0), plus 1.3.

Seven negative surprises, three positive. Every one of the ten fell relative to its own prior except the two British ones and German manufacturing. The largest miss of the morning is French services at −2.3 and the largest beat is the British composite at plus 1.3, and they are 75 minutes apart on opposite sides of the Channel.

Now put the two manufacturing prints that matter side by side. Germany 54.2, France 49.3 — a 4.9-point level gap between the euro area’s two largest economies, and a 1.9-point gap in their surprises with opposite signs. The 08:00 aggregate that followed them is one number, 52.4, which is 0.3 below what was expected of it. If the only entry in your schedule is the euro-area release, the morning registered as a small miss. What actually happened is that France contracted and Germany accelerated, and you learned both of those facts before the entry you were waiting for.

An aggregate is a lagging indicator of its own components

This is not a complaint about the statistical agency. The sequencing is a publication convention: national flashes go out first, the area aggregate is assembled from them and goes out afterwards. So the 08:00 release is, mechanically, a summary of information that has been public for between thirty and forty-five minutes. It contains almost nothing a reader of the 07:15 and 07:30 releases did not have.

What makes it worse than merely redundant is the arithmetic above. Averaging an accelerating Germany against a contracting France produces a number whose surprise is smaller than either input — 0.3 against 1.2 and 0.7 — and whose sign matches the smaller economy. A filter keyed to the aggregate therefore sees the mildest version of a divergent morning, 30 to 45 minutes late, and sizes accordingly.

The practical form of this, for anyone whose system has one euro-area PMI entry in it: the entry is in the wrong place. The information arrives at 07:15 and 07:30. If the point of a news window is to avoid being positioned while a repricing happens, the windows to keep are the national ones, and the aggregate is the one that can be dropped without losing anything. That inverts the usual instinct, which is to keep the headline and ignore the detail.

And sizing by surprise magnitude fails on the same morning

The obvious response is to size by the size of the surprise. The morning refuses that too. The biggest beat of the ten fields was the British composite at plus 1.3, with manufacturing alongside it at plus 1.1 — an unambiguous upside surprise in both fields that were published with a consensus. Sterling fell. One reader has cable at about 1.3380 in European trade, down 0.1 per cent on the day, and describes the pound as weakest against the yen; the same note attributes the move to a dollar rebound ending a four-day losing streak and to the morning’s retail sales miss rather than to the survey, and adds that a Bank of England increase in November now looks close to certain on one house’s view.

So the currency with the largest positive surprise in the cluster underperformed, on a day when its own central bank is being priced for a hike. We are not going to build a rule out of one morning. The narrow point is enough: the magnitude of a surprise does not tell you the magnitude of the move, because the surprise competes with whatever else is in the window, and this window contained a dollar reversal and a retail sales print from six hours earlier that was still being digested.

One field-labelling caution, since it bears on anything automated. Two publishers carry the same pair of British figures, 53.1 against 51.8 expected; one labels it the composite with a 52.0 prior, the other labels it services. Only one of them can be right, and the prior that comes with the composite label is the one consistent with the separately published manufacturing figure. We have used the composite label and we are telling you it is a judgement, not a confirmation. A system that keys on a field name rather than a release will take whichever version its feed happens to carry.

Two clock changes, eight days apart, and one week when the gap is wrong

The forward-looking part, and it is the kind of defect that costs nothing to fix now and is expensive to discover live. Europe leaves summer time on Sunday 25 October; the United States leaves daylight time on Sunday 1 November. For the eight days in between, European releases have shifted an hour later in UTC while American releases have not moved.

Take this morning as the measuring stick. The British survey at 08:30 UTC and the American one at 13:45 UTC are 315 minutes apart. The same two local release times next week are 09:30 and 13:45 — 255 minutes, because only one of them moved. From 2 November, with the American clock changed too, the gap is back to 315 minutes. So for one week the distance between the European and American event blocks is 60 minutes shorter than it is on either side of it, and any schedule whose entries are stored as UTC offsets rather than as local times with a zone attached will be wrong about every European entry for eight days and right again afterwards, which is the worst possible failure pattern because it heals before you investigate it.

The entries to re-derive before Monday are the ones in that eight-day window, and there are several that matter: a Federal Reserve decision on 28 October, a European Central Bank decision and a Bank of Japan decision on 29 and 30 October, and American third-quarter output on 29 October. We are deliberately not publishing new UTC times for those, because we have not sourced them individually and a confidently wrong time is worse than a flagged gap. The instruction is to re-derive them from each issuer’s own local-time announcement rather than from last month’s UTC stamp.

What is still to come today, and what it is clustered with

The American block is three events inside fifteen minutes. The October flash surveys are at 13:45 UTC — two-sourced, from one publisher giving 13:45 GMT and another giving 9:45 Eastern — with manufacturing consensus at 55.6 against a 55.9 prior and services at 57.9 against 58.8, so both are expected to soften without turning. Note that a second reader gives the manufacturing consensus as 55.5; a 0.1-point disagreement is not material to a trade but it is material to any rule that compares a print to a stored number. The final October consumer sentiment reading and its one- and five-year inflation expectations follow at 14:00, forecast at 46.3 on a single reader, and the New York option cut is at the same minute.

That is the same structural problem as the European morning in a tighter space: a survey and a sentiment index fifteen minutes apart, with a mechanical flow event on top of the second one. Fifteen minutes is inside most news pads, so in practice this is one window, not three, and treating it as three invites a system to re-enter between them.

Also outstanding and unscheduled: Moody’s review of France, which one reader expects after the European close and which no source in our sweep times. The current rating is Aa3 on a negative outlook; a one-notch cut to A1 would still leave it above two other agencies at A plus. The French ten-year was 5.03 or 5.04 per cent this morning with the spread to Germany testing 175 basis points, and the French survey above is the first French data since that yield went through 5 per cent. An unscheduled rating action landing after the cash close, on a day the country’s own survey went into contraction, is the clearest case there is for a position limit rather than a news filter — a filter needs a time, and this event does not have one.

What this does not tell you

The release times are taken from the morning’s publication stamps and a Japanese schedule, not from the issuer’s own calendar, which we did not open. Composite consensus figures are the weakest numbers here: no source in the sweep gave one for the German or euro-area composite, and the French composite consensus appears as 50.5 on one reader and 50.3 on another, which makes that miss either 2.1 or 1.9 points. We have used 50.5 and the difference does not change the sign or the argument. The British field-label conflict is unresolved, as stated. The sterling observation is one publisher’s level and one publisher’s attribution, and the claim that the pound fell on a day of positive surprises is an observation about two facts in the same window, not a causal finding. The clock-change arithmetic is definitional and we are confident in it; the specific future release times it implies are exactly what we are declining to publish. And nothing here is a view on the euro, the pound or the dollar.

Related

• FX events calendar — the 13:45 and 14:00 UTC American block and the New York cut.
• Signals
• EA presets