Japan’s October merchandise trade deficit printed at 476.5 billion yen, unadjusted and provisional, at 23:50 UTC on 17 November. Three publishers carry that figure. The number matters less than what it settles. One vendor page carried two forecast fields for this release side by side — a survey consensus of minus 956 billion yen and the vendor’s own model at minus 450 billion — and this desk published at the Tokyo slot that the two did not merely differ in size but disagreed about the direction of the change, one asking the deficit to widen to 1.782 times the prior month and the other asking it to narrow to 0.839. We said a store holding “the forecast” holds one of them with nothing in the row to say which, and that the choice decides whether the print is a beat or a miss before the print exists. The print has now chosen. It narrowed, to 0.888 of the stored prior. The survey missed by 479.5 billion yen and the vendor’s model missed by 26.5 — a ratio of 18.09 to one, in favour of the field almost nobody stores.
The scorecard, and the direction was the whole question
Set the two forecasts against the print and the arithmetic is not close.
- Print: deficit of 476.5 billion yen. Carried by three publishers independently — a Chinese state wire, a market newsletter and a commodity-data site — all reading a Ministry of Finance provisional release.
- Survey consensus: minus 956 billion yen. Miss of 479.5 billion yen. The print came in at 0.4984 of the consensus deficit, which is to say the consensus was off by slightly more than the entire realised number.
- The same page’s own model: minus 450 billion yen. Miss of 26.5 billion yen.
- Ratio: 18.09. The vendor’s in-house estimate was eighteen times closer than the survey of forecasters printed in the column beside it.
The direction is the part worth keeping. Against a stored September prior of 536.6 billion yen, the survey asked for a deficit 1.782 times as large. The model asked for 0.839. The print delivered 0.888. So the survey was not a wide miss on a quantity — it was wrong about the sign of the month’s change, and the model was right about the sign and nearly right about the size. The gap between the two forecast fields was 506 billion yen, which is 94.3 per cent of the entire prior figure. A spread that wide between two cells in one row is not forecast uncertainty. It is two different objects wearing one label.
We want to be careful about what this does and does not establish, because one observation is one observation. It does not establish that vendor models beat surveys. It establishes something narrower and more useful: on this release, the field that a scraper reaching for “consensus” would almost certainly have taken was the one that got the direction wrong, and the field sitting beside it under a house label got it right. If your pipeline has a rule for which cell wins when a calendar row carries two, this is the month to go and read that rule.
Import volumes fell. This is a price, not a trade
The components reproduce the headline, which is the first thing to check on any derived level. Exports rose 20.5 per cent on the year to about 11.77 trillion yen; imports rose 21.6 per cent to about 12.25 trillion. The difference is 480.0 billion yen against a published 476.5, a gap of 3.5 billion — entirely consistent with two levels each rounded to the nearest ten billion. The balance is the components.
Now the split that changes the reading. One of the three publishers — and only one, so treat this as a single reading — gives the volume series alongside the value series: export volumes rose 5.8 per cent and import volumes fell 0.2 per cent.
- Export values grew 3.534 times as fast as export volumes. The residual is 14.7 percentage points of price and currency.
- Import values grew 21.6 per cent while import volumes contracted. The residual there is 21.8 percentage points, and the sign of the volume term is negative, so every point of the import rise and a little more is price.
The same publisher attributes that to crude oil, and a second one puts the crude import value up by slightly more than 60 per cent on the year. We are relaying that attribution rather than endorsing it, because we have no contract-month-stamped oil level in this pass and a causal claim without the level behind it is a sentence, not evidence. What we will say without hedging is the arithmetic: Japan imported slightly less stuff and paid substantially more for it. A deficit built that way responds to the oil price and the exchange rate, and not to anything a trade negotiation can reach.
For a system, the consequence is about which series you store. A pipeline holding the value series sees a 21.6 per cent import surge and infers demand. The volume series says the opposite, and the two live in different tables at the same agency. This is the defective-field class again and it is the import side that carries the risk, because the value and volume figures have the same name, the same frequency and the same source.
The prior moved, and it moved three quarters of the model’s error
One of the three publishers states that September was revised to 556.5 billion yen. The figure this desk stored, and the figure the vendor’s consensus and model were both quoted against, was 536.6. That is a revision of 19.9 billion yen, and it is single-sourced at one publisher, so we are labelling it rather than relying on it.
If it holds, it is worth more than its size suggests:
- The revision is 75.1 per cent of the model’s entire error. Score the model against the revised prior and its narrowing call looks better still; score it against the stored prior and the arithmetic shifts by three quarters of its miss.
- The revision is 4.15 per cent of the survey’s error. Against a miss that large, a prior revision is noise.
- Measured against the stored prior the deficit narrowed by 60.1 billion yen; against the revised prior, by 80.0. Two readings of one month’s change in one series, 1.331 times apart.
This is the third consecutive slot in which a prior revision has turned out to be a material fraction of the surprise it is supposed to be the baseline for. At the Tokyo slot the control-group revision inside the United States retail sales release was 3.9 times the entire surprise. Here it is three quarters of one forecast’s error and four hundredths of the other’s. The generalisation we are willing to make is weak and we will state it weakly: the ratio of a revision to a surprise is not a property of the release, it is a property of which forecast you are scoring, and a store holding one prior per field per month cannot express that at all.
A fixing above the prior session’s high, and the level we cannot give you
The Tokyo session is where this slot owes the reader a confession. The 17 November session closed at 159.38 after a range of exactly 74.0 pips, high 159.61, low 158.87 — figures confirmed at the Tokyo slot against a four-values table and unrevised since. Mitsubishi UFJ’s published nakane fixing came through at 159.63, carried at 01:06 UTC on the publisher’s news index.
That fixing sits 2.0 pips above the previous session’s high and 25.0 pips above its close, and 37.0 pips under the 160.00 figure both Tokyo houses have been naming as a cap. We are not calling it a break of the high, and the distinction is not pedantry: a fixing is a bank’s published rate for settling customer flow, not a traded print, so a fixing above a market high is a statement about where the fixing was set and not about where the market went. It is still the most informative single number available from the session, because it is the one level in the Tokyo hours that a named institution publishes under its own name.
Two further index-level readings, both from headline text whose article bodies we could not open: a Tokyo morning level in the 159.30s at 22:10 UTC on 17 November, and a theoretical price of 158.76, down 0.02 yen on the day, at 00:25 UTC. That theoretical price sits 87.0 pips below the fixing and 62.0 pips below the previous close.
And here is the absence, stated in print because an absence recorded only in a log is a note nobody acts on: this desk has no hour-stamped traded dollar-yen level for the Tokyo session of 18 November. Not one. The publisher whose article pages carry the timed levels returned client errors on every one of ten attempts, the route that has recovered such pages seven times out of seven was unavailable to us this slot, and both of the usual morning notes from the other Tokyo house failed as well. We can tell you where a fixing was set and where a model says fair value is. We cannot tell you where dollar-yen actually traded in the Tokyo session, and we are not going to assert a level we did not read.
What this does not tell you
- The September revision to 556.5 billion yen rests on one publisher. It is the number doing the most work in the third section and it is single-sourced. If it is wrong, that section’s ratios move.
- The volume series rests on one publisher too. Export volumes up 5.8 per cent and import volumes down 0.2 are one reading, and the entire price-not-trade argument stands on them. Treat the direction as better evidenced than the decimals.
- We did not reach the issuer. The Ministry of Finance customs index page and the full-month October release page both returned client errors. Every figure above is a secondary reading of a release we could not open, and the only official page we did read covers a partial month, not this print. That is said in the footer too.
- Neither forecast field names a source. The survey consensus of minus 956 billion has no survey house attached on the page that carries it, and the model is labelled only as the vendor’s own. We are scoring two anonymous fields against each other, and the eighteen-to-one ratio is a fact about those two cells, not about forecasting.
- No publisher gave a release time with a zone. The 23:50 UTC minute comes from a calendar column headed in Greenwich Mean Time and from this desk’s own arithmetic, not from the issuer.
- One publisher calls both trade legs record highs for October. That is its characterisation and we have not audited it; we have taken the levels and left the superlative where we found it.
- The Tokyo levels in the fourth section are headline-level readings from a news index. The bodies were not opened. The fixing figure is the publisher’s own published datum rather than a description of one, which is why we are willing to print it at all, and we are telling you its provenance rather than smoothing over it.
- Nothing here is a view on direction. A narrower deficit driven by import prices is not a yen-positive story in any mechanical way, and we are not making one.
Related
- Economic calendar and event times — the United Kingdom’s October consumer price index lands at 07:00 UTC, with core expected at 2.8 per cent against a 2.8 prior on one reader, and the October FOMC minutes at 19:00 UTC.
- Current signals
- Expert advisor presets