Seven hours ago this desk published four live consensus splits resolving today and said the euro area manufacturing PMI was forecast at 52.6 on one feed and 52.7 on another. At 05:48 UTC a third number appeared for the same release: 53.3. Two hours before the print, that looks like the spread widening from a tenth to seven tenths — wide enough to flip the sign of the surprise for any print between 52.8 and 53.2. It is not. The 53.3 is one house’s own forecast, and the only thing in the text that tells you so is the verb.
One note, two kinds of number, and nothing but the verb separating them
The note in question ran on FXStreet at 05:48 UTC this morning. On the euro area it says the author forecasts the manufacturing index to rise to 53.3 from 52.7, and that services activity is likely to edge lower to 51.4 from 51.6. Further down, on the United States, it says September flash PMIs are expected to remain solidly above 50, with manufacturing at 53.6 and services at 56.0.
Read those two passages as a data feed and you get five forecasts. Read them as English and you get two of one kind and three of another. “We forecast” is the author’s model. “Are expected” is the market’s consensus. The numbers sit in the same note, in the same format, a few paragraphs apart, and nothing but that distinction marks which column each belongs in.
This is not a criticism of the note, which is doing exactly what a house note is for. It is a warning about what happens when that note is scraped, summarised, or read quickly at six in the morning by somebody assembling a board. This desk has published a house model sitting in a consensus column before. This is the tighter version: both kinds of number in one document, correctly written, and still trivially confusable.
The identification check comes before the containment check
We have a standing rule for two vendors showing different consensus figures: check whether they are actually disagreeing before publishing it as a conflict. That rule assumes both numbers are consensus figures. The check that has to happen first is cruder, and we have not written it down until now: is this number a forecast of the market, or a forecast by the author?
Applied here, it costs us an article and improves the board. Had we published 52.6 against 53.3 as a fifth split, we would have printed a seven-tenths disagreement that does not exist, on the most-watched European release of the day, two hours before it lands. The correct entry is that FXStreet’s own release page for the euro area manufacturing PMI carries a consensus of 52.7 against a previous of 52.7 — which is what this desk read there yesterday and what it still says this morning.
That also settles which number wins under our other standing rule: trust a vendor’s release page over its preview article. The release page says 52.7. The preview says 53.3 and means something else. Both are FXStreet and neither is wrong.
Where the four splits actually stand, two hours out
We published four this morning. Here is the mark, and one of them has quietly closed.
- Euro area manufacturing, 08:00 UTC. Still split, and still only by a tenth: 52.6 on Forex Factory’s calendar, 52.7 on FXStreet’s release page. Both carry the previous as 52.7.
- Euro area services, 08:00 UTC. Closed. We published 51.4 against 51.7. This morning Forex Factory shows 51.4 with a previous of 51.6, and the house note also has 51.4 from 51.6. The 51.7 is gone. Nobody announced a revision; the number simply stopped being there.
- UK manufacturing, 08:30 UTC. Forex Factory has 51.5 against a previous of 51.7. We did not re-read the second feed on this one today, so we are not claiming the split survived — only that we did not check it.
- US services, 13:45 UTC. Still split: 55.8 on Forex Factory against a previous of 56.5; 56.0 in the house note, where it sits in the “expected” column and is therefore a genuine second reading of the consensus rather than a third model.
Three of four still live, one closed silently, and the one that looked like it had exploded overnight never moved at all. That is a better outcome than this desk expected when it set the claim up, and the part worth keeping is that the closure was invisible: a feed changed a forecast between two readings a day apart and nothing marked the change.
The prior is still not in dispute, which is this morning’s claim holding
The earlier piece argued that a feed disagreement leaves the level and the month-on-month change intact and corrupts only the surprise. Today’s board is a second test of that and it holds cleanly. Every reader we have on euro area manufacturing — Forex Factory, FXStreet’s release page, the house note — carries the previous as 52.7. All three agree on where the series was. They disagree, by a tenth, about where it is going.
So whatever prints at 08:00, the month-on-month change will be the same number on every desk in Europe, and the beat-or-miss will differ by a tenth. If you size by surprise, that tenth is your error bar and it is small. If you size by the change, you have no error bar at all. That is the second consecutive session on which the cheaper quantity has been the more reliable one, and we would rather say it twice than have it read as a one-day coincidence.
The board, in UTC, and where spot sits inside it
07:15 French flash PMIs, manufacturing 50.9 against a previous of 51.1 and services 48.3 against 48.0. 07:30 German flash PMIs, manufacturing 54.1 against 54.3 and services 49.9 against 49.7. 08:00 euro area, the two above. 08:30 UK, manufacturing 51.5 against 51.7 and services 52.0 against 52.5. 13:45 US, manufacturing 53.6 against 53.9 and services 55.8 against 56.5. 14:30 crude inventories, forecast minus 0.7 million barrels against a previous of minus 0.6 million — a forecast that has itself moved a tenth since our scan yesterday afternoon, which we note and are not building anything on.
The euro arrives at this with a fresh low since 29 July around 1.1425 and a dollar index near 100.70. Today’s 10:00 New York cut carries EUR/USD strikes at 1.1400 and 1.1450 with spot between them, less than thirty pips apart, and a 100-hour moving average at 1.1468 sitting above. The notionals are inside an image on the source page and did not extract, for the fourth time this month, so we cannot tell you how large any of that is. What we can tell you is that the euro area print lands with spot inside a fifty-pip expiry bracket and several hours of London liquidity to work through before that bracket expires.
What this does not tell you
It does not tell you what the PMIs will print. It does not tell you that 52.7 is the right consensus — only that it is the one on the release page, which is where our rule says to look, and that Forex Factory’s 52.6 is a tenth away and equally readable.
It does not tell you whether the euro area services 51.7 we published yesterday was an error, a stale render, or a forecast that genuinely moved. We can see that it is no longer there; we cannot see why, and the feed does not say.
It does not tell you whether the UK manufacturing split survived, because we did not go back to the second feed this morning and we would rather record the gap than fill it. And it does not tell you the size of this morning’s option expiries, because that number is in a picture.
Related
- FX events calendar — every release above, in UTC
- Signals
- EA presets
