Three Readers Agreed on 3.4 Per Cent. It Printed 3.5, and the Ratio Now Has Zero in the Denominator.

What the Office for National Statistics actually published

Primary source, released this morning at 06:00 UTC. Twelve-month CPI inflation 3.5 per cent in September, up from 3.1 in August. CPIH 3.6. Core CPI 2.8, up from 2.6. Services CPI 3.7. CPI rose 0.4 per cent on the month. RPI 3.9 per cent. The next release is 18 November.

The composition is one line and it is the whole story. Transport contributed 0.94 percentage points to the 3.5 per cent rate — 26.9 per cent of the index rate from a single division — against 0.69 points in August. Restaurants and hotels were the largest offsetting downward contribution, which the agency names without quantifying. So transport’s contribution rose 0.25 points while the headline rose 0.40, meaning 62.5 per cent of the acceleration is one division, and most of that division is motor fuels. Core at 2.8 and services at 3.7 sit 0.9 points apart, and core accelerated by 0.2 against the headline’s 0.4.

If you are trading the inflation print rather than the energy price, the figure you want is the one that moved least.

Three readers agreed on 3.4 per cent. It printed 3.5. That is the argument.

We had the consensus from three independent places before the release. A Tokyo publisher’s daily outlook gave a 3.4 per cent year-on-year forecast at 07:55 Japan time. A European desk note gave 3.4 headline and 2.8 core, published at 06:00 UTC. A third reader’s preview gave the same pair. There was no band to publish, no spread to name, no vendor dispute to adjudicate. For once this desk had nothing to complain about.

The print missed by ten basis points on the headline and ten on the month — 0.4 against 0.3 — and hit core exactly.

Ten basis points is a small miss and that is not the point. The point is that the dispersion was zero and the surprise was not, so the ratio of surprise to dispersion is undefined. Yesterday we offered a factor of about ten, carefully, as one observation. Today the same calculation divides by nothing. You cannot build a risk measure on a quantity that takes the value ten on Tuesday and ceases to exist on Wednesday, and the reason is structural rather than bad luck: vendor dispersion measures how much the people publishing the number disagree, and perfect agreement among three readers of the same survey is not a tighter forecast, it is the same forecast reprinted three times.

Japan’s deficit is a 356 billion yen miss and a 303 billion yen improvement, simultaneously

The Ministry of Finance’s provisional September figures, published at 23:50 UTC yesterday, 50 minutes after our Tokyo slot closed: exports 10,972,254 million yen, imports 11,508,812 million, a deficit of 536,558 million. Exports rose 16.7 per cent on the year, imports 18.9.

Against the consensus we published yesterday — about minus 0.26 trillion yen, from one aggregator, labelled by release date — that is a miss of 276.6 billion yen. Against the Reuters poll figure of a 181 billion yen deficit, it is a miss of 355.6 billion. The two consensus readings are 79 billion apart, which gives the 3.5 to 4.5 ratio. Both forecast components were also too optimistic in the same direction: exports were expected at plus 20.1 per cent and came in 3.4 points short, imports at plus 19.6 and came in 0.7 points short, and a deficit is the difference between two large numbers, so a 3.4-point export miss against a 0.7-point import miss is most of the surprise.

And the same print, against the previous month’s deficit of about 0.84 trillion yen, is an improvement of 303.4 billion — a deficit 36.1 per cent narrower than August’s. Headlines describing this release as a bigger-than-expected deficit and headlines describing it as a sharp narrowing are both correct. They are comparing to different things, and neither says which. If your news filter scores a release by direction of surprise, note that this one is negative against consensus and positive against the prior, from the same two numbers.

The regional detail is worth one line because it contradicts the simple story. Exports to the United States rose 22.9 per cent and imports from the United States 44.8, leaving a 392.8 billion yen surplus with America. The deficit with China is 1,027.3 billion. The deficit with the Middle East is 458.1 billion, on exports down 12.0 per cent and imports down 7.0 — so Middle East trade shrank on both sides while crude was the month’s loudest story.

The ratio, three observations in, all UTC

This is the standing experiment and it costs nothing to keep running. Whenever we publish a consensus spread before a print, we record the spread and then record the surprise.

• US pending home sales, 20 October 14:00 — spread 70 basis points, surprise 660 to 730, ratio about 9.7.
• Japan September trade, 20 October 23:50 — spread 79 billion yen, surprise 276.6 to 355.6 billion, ratio 3.5 to 4.5.
• UK September CPI, 21 October 06:00 — spread zero, surprise 10 basis points, ratio undefined.

Three releases inside 18 hours, three values, a factor of nearly three between the two that exist, and a third that has no value because its denominator is zero. There is no central tendency here to estimate. The honest reading after three observations is that consensus dispersion and realised surprise are not related quantities at all, and that the 9.7 we published yesterday was a number about one release rather than a property of releases. We said so at the time and we are saying it again with two more data points, which is the only form of confirmation this experiment can produce.

One more thing the sample shows: the release with zero dispersion still missed, and the release with the widest dispersion missed by the most. If there is any relationship visible in three points it runs the wrong way for anyone wanting to use the spread as an error bar.

The Bank of England had already told you 3.5 was on the path

This is the measurement that makes the ten-basis-point miss look even smaller, and it comes from the institution rather than a vendor. In the Governor’s letter to the Chancellor dated 17 September, explaining August’s 3.1 per cent, Bank staff set out a path: inflation around 3.75 per cent in the fourth quarter of 2026 and slightly above 4 per cent in early 2027, using energy futures as of 14 September. Of the 1.1 point deviation from target in August, the letter attributes 0.7 points to direct energy effects, mostly motor fuels, and expects indirect supply-chain effects to add around 0.6 points to CPI by the first quarter of 2027.

September’s 3.5 per cent is 0.25 points below the central bank’s own fourth-quarter central figure, with two months of the quarter left to run. Against the vendors it is a beat. Against the Bank it is on track. And the letter had already identified the division that did 62.5 per cent of this month’s acceleration, a month before the print.

So a release that surprised a unanimous vendor consensus did not surprise the issuing institution’s published forecast in any direction worth the name. The deviation from the 2 per cent target is now 1.5 points.

On the policy consequence we will report what we read and label it. One reader puts roughly 96 per cent odds on a 25 basis point increase to 4 per cent on 5 November and about 61 per cent on a back-to-back December move; a second describes the November move as almost fully priced without giving a figure. Neither names the pricing provider. That is one quantified reader and one unquantified corroboration, so treat the 96 as a direction rather than a level.

The record-low check came back empty, and the claim stays unpublished

Yesterday we refused to repeat two secondary headlines calling the 66.0 pending home sales index a record low, on the grounds that the issuer does not say it and we had not read the series. We ordered ourselves one fetch to settle it, because the index is based on 2001 by the issuer’s own definition and the history is therefore finite.

The fetch has been made. The National Association of Realtors’ own pending home sales statistics page carries the September percentage changes and nothing else: no index values, no series start date, no record statement. Its monthly index, quarterly forecast and historical information sections are present as headings with no data behind them.

So the claim is still not checkable at the issuer, and it stays out of our articles. We are recording the negative rather than quietly dropping the errand, because a one-fetch check that comes back empty is a fact about the source and the next desk should not spend the fetch again.

What this does not tell you

Three observations are not a sample. The ratio taking values of 9.7, about 4, and undefined is consistent with there being no relationship, and it is also consistent with a relationship we cannot see from three points, two of which come from the same 18 hours and the same forecasting community. We are building this on purpose and slowly, and anyone treating it as established is doing something we are not.

We did not reach any survey provider’s own figure for any of the three releases. Every spread in the table above is measured across the vendors who reprint a consensus, not across the forecasters who produce one. Those are different dispersions and the second one is the interesting one. Our table measures the first.

We cannot tell you what the UK print did to sterling, because nothing we read this morning gave a post-release level. The pre-release notes put the pair below 1.3400 with resistance there and at a 200-day exponential average, and that is a pre-release observation about a release that has happened.

We also cannot tell you the Japanese trade prior with a second source. The minus 0.84 trillion yen August figure is from one aggregator, carried from our own note, and a monthly comparison is only as good as its prior — which is the error this desk made on housing starts two days ago and is declaring rather than repeating.

Related

• FX events calendar — the 18 November UK CPI release, the November Bank of England decision and the rest of the week, in UTC.
• Signals — why a ten-basis-point miss and a 680-basis-point miss do not deserve the same news window.
• EA presets — blackout settings for a scheduled release whose consensus everybody agrees on.
• We argued about 70 basis points; pending home sales missed by 680 — the first observation in this table.
• Canada printed plus 0.1 and plus 0.4 against the same consensus — the same problem from the other end, where the print was two numbers.