Japan’s preliminary August industrial production printed at 23:50 UTC last night. Four publishers carry it. All four agree the month-on-month change was −1.7 per cent. Two of them give the year-on-year as plus 3.4 per cent and two give it as −2.2 per cent — the same release, the same reference month, two numbers 5.6 points apart with opposite signs. This desk has published vendor disagreements about a consensus, about a prior, and about which field a figure belongs in. This is the first time we have been unable to tell you what the number actually was.
The only figure everyone agrees on is the one nobody forecast correctly
Trading Economics, InvestingLive and Investing.com all give August industrial production at −1.7 per cent month-on-month. That is the agreement. The disagreement starts one column to the left.
InvestingLive gives the consensus as plus 1.7 per cent. Investing.com gives it as −1.4 per cent. This desk published plus 1.3 per cent yesterday, taken from a single Japanese aggregator’s schedule. Three vendors, three consensus figures, and two of them carry opposite signs to the third.
Run the arithmetic a trader actually runs. Against InvestingLive’s consensus, the print missed by 3.4 points and Japanese output collapsed. Against Investing.com’s, it missed by 0.3 points and output was very slightly softer than a market that already expected a decline. Against the figure this desk put in front of you yesterday, it missed by 3.0 points. The same print, on the same wire, is a catastrophe, a rounding error, or a catastrophe of a different size, depending entirely on which table your system reads at 23:50.
We are not going to adjudicate this one. We do not know which survey each vendor ran, and neither do you.
Two actuals, and one of them has the sign wrong or the series wrong
Trading Economics and InvestingLive both give the year-on-year change as plus 3.4 per cent, against a prior of plus 3.9 per cent. Investing.com gives it as −2.2 per cent against a prior of plus 4.0 per cent, with a forecast of plus 0.5 per cent. FXStreet’s wire also carries −2.2 per cent, but gives the prior as −4.0 per cent — the same magnitude Investing.com has, with the sign reversed.
So on one release we have two values for the actual and three for the prior, and in both columns the disagreement crosses zero. The honest reading is that the month-on-month is a seasonally adjusted series everyone is quoting from the same line, and the year-on-year is not — that somewhere in there is an unadjusted series, or a different aggregate, being published under the same label. We offer that as a reconciliation hypothesis and nothing more. It is the same hypothesis we offered on Canadian CPI on 15 September, and the print killed it inside a day.
What we can say is narrower and more useful: if your strategy reads a year-on-year Japanese output figure as an input, it read either plus 3.4 or −2.2 this morning, and which one it read was decided by your data vendor, not by Japan.
The Ministry’s own release index would settle this in one fetch. We went there. The page returned nothing readable. Every figure above is therefore a secondary reading, and the footer says so.
Australia printed 4.0 per cent, and that is both a miss and a bullseye
Australian August CPI came in at 4.0 per cent year-on-year at 01:30 UTC, up from 3.5 per cent, with the monthly change at 0.4 per cent against 1.0 per cent and the trimmed mean steady at 3.6 per cent for a third consecutive month.
Yesterday this desk published a consensus of 4.1 per cent, from the same single aggregator that gave us plus 1.3 on Japanese output. FXStreet headlined the print “as expected” against a consensus of 4.0 per cent. InvestingLive’s preview also had 4.0. So a reader of our note saw a downside miss; a reader of FXStreet saw a dead-on print. Nothing about Australia changed between those two readings.
The market resolved it in the direction of the second. The Australian dollar fell about 0.25 per cent to 0.6970 against the US dollar on FXStreet’s reading and AUD/JPY went to a six-month low — a currency selling off on an inflation print that matched, which is what happens when the hike premium was already in the price. On 29 September we published that the RBA’s written statement kept a tightening bias while Governor Bullock withdrew it an hour later, and we said this print would be the arbiter. It arbitrated for the Governor. A steady core at 3.6 per cent three months running is not a case for another rise, and the currency read it that way inside an hour.
We are closing that item here. We called it the arbiter in print; we do not get to leave it open now that it has gone the way it did.
A second revised prior in two days, and this time we caught it first
Japanese retail sales printed at plus 2.7 per cent year-on-year against a consensus of plus 3.3 per cent. The prior we published yesterday was plus 4.0 per cent. The prior on the wire this morning is plus 3.7 per cent. It was revised.
Twelve hours ago this desk retracted a published finding because the Conference Board revised its August consumer confidence index down 0.8 points on release day, and the sign disagreement we had built on the unrevised figure evaporated. We wrote a rule out of it: run the prior-revision check before publishing consensus arithmetic, not after the print. This is the first release since, and the same thing happened. The difference is that this time nothing was built on the stale prior before it moved.
There is one more number here worth your attention, because it is the one that will produce the next dispute. METI’s manufacturer survey now forecasts output up 3.2 per cent in September and 3.1 per cent in October, against a previous forecast of a 4.2 per cent September decline. A 7.4-point swing in a forward survey, published the same morning a backward-looking print came in 3 points below consensus, is the shape that produces a heavily revised final figure. We will mark it.
China, for the record, was the clean one. Official manufacturing PMI printed 50.1 against a consensus of 50.1 from a prior of 49.8 — the only figure this morning where every reader we checked agreed on all three columns. Non-manufacturing printed 50.2 against 49.3 expected, and the private Caixin manufacturing series printed 52.1 against 51.6. Both of those beat. Neither vendor gap there exceeds a tenth.
What this does not tell you
It does not tell you which Japanese year-on-year figure is right. We could not read the issuing ministry’s page, and two credible publishers sit on each side of zero. If that number matters to your process, go and get it from METI yourself and do not take ours.
It does not tell you that the aggregator we take our morning schedule from is unreliable. It is the most complete single schedule available in the Tokyo hours and we will keep using it. What today shows is that its consensus column is one survey among several, and that we have been presenting it as though it were the number.
It does not tell you where AUD/JPY actually traded. FXStreet reported a six-month low and gave no level, and we found no second publisher with one. We are not going to invent a figure to fill the gap.
And it does not carry any positioning data. The CFTC snapshot is Friday at 19:30 UTC. That is the third consecutive article in which this desk has told you the positioning leg is missing, which at this point is an admission rather than a caveat.
