Yesterday evening this desk published a two-way, losable pre-registration on the Bank of Japan’s Summary of Opinions for the 29–30 October meeting. We named the field, the shape and the losing condition: does at least one opinion argue for a near-term hike, and are wage developments the stated ground? We added that a hike argument resting on the currency or import prices instead of wages would be a loss, and the more interesting one. The document published at 23:50 UTC. We have now read it at the issuer, in full, in Japanese. The first leg won: six of the fifteen monetary-policy opinions argue for moving sooner. The second leg lost. And the loss we had pre-named — the currency instead of wages — also lost, because in thirty-four opinions the yen is not mentioned once. The test offered two answers and the correct one was a third.
Six of fifteen want to go faster, and not one of them says wages
The issuer document carries thirty-four opinions across three sections: thirteen on the economic and financial situation, fifteen on monetary policy, six from government representatives. One Tokyo publisher independently counts thirty-four and independently counts six pro-hike opinions in the policy section, which matches our own reading of the issuer bullet by bullet. That is as close to corroboration as a classification gets, and it is worth noting that both counts are classifications: the document does not label its own bullets hawkish or cautious.
Here is what the six actually rest on. One: keep raising to adjust accommodation, because underlying inflation is roughly at the 2 per cent target and financial conditions are accommodative. Two: do the next hike when you can, without being overly cautious, having confirmed economic resilience. Three: corporate tolerance for higher rates is rising, past hikes have not shown visible effects, so this is a phase of faster progress toward neutral. Four: underlying inflation has already reached 2 per cent, so the rate should reach neutral sooner, checking conditions every few months. Five: a regime shift is needed to reach neutral early, so the bank can move in both directions. Six: hikes continue as the premise but there is no need to rush.
Tally the grounds and you get underlying inflation three times, the neutral rate three times, financial conditions twice, corporate rate tolerance once, economic resilience once. Wages: zero. That is not an argument about emphasis. Wages appear five times in this document and every one of them sits in the Prices sub-section — the wage-price linkage under labour shortage, next spring’s negotiations as the anchoring watch point, companies saying they will raise pay next year at roughly this year’s pace, one view that next year will not rise much above this year with little sign of second-round effects, and one saying the wage trend must be checked before judging underlying inflation. Two of those five lean dovish. None of the fifteen policy opinions reaches for a wage number to justify moving.
Thirty-four opinions, zero mentions of the currency
This is the finding we did not pre-register and should have. We searched the issuer’s own Japanese text for the exchange rate, for foreign exchange, for the yen as a currency and for import prices. There are none. The character for yen occurs exactly once in the document, inside the word meaning “smoothly”, in a bullet about exchange-traded fund disposals. Energy and oil costs do appear, twice — as a drag on this fiscal year’s growth that fades from the next, and as the reason current inflation is elevated with limited secondary effects so far. Both sit in the assessment of the economy and prices. Neither is attached to a policy argument.
Hold that against the market you traded for the last six weeks. This desk has published a record single-day yen-buying intervention, a 7,975.6 billion yen operation disclosed two mornings ago, and a Tokyo publisher’s view that 159 is the level above which intervention becomes a concern. If you built a news filter that keys BOJ documents on currency language — and plenty do, because the currency is what moves on them — this document returns nothing, and a filter returning nothing looks identical to a filter that failed. The practical instruction is narrow and we will state it plainly: on this document class, score the policy section’s grounds, not its vocabulary, because the vocabulary you are watching for is absent even in the month the record intervention was disclosed.
One more structural point, because it is the sort of thing that breaks stored expectations. The Summary of Opinions does not state the policy decision and does not state the vote. It is opinions only. The decision — hold at 1.25 per cent, with one board member dissenting in favour of 1.5 per cent — reaches you from secondary readers, not from this page. So a board that voted to hold produced six opinions arguing to move faster, and the two facts arrive in different documents on different days. Key on the vote and you saw a hold. Key on the opinions and you see forty per cent of the policy section pressing the other way.
The consumption print the board never saw, and our own consensus was wrong
Several of the cautious opinions cite soft private consumption; one of the pro-hike opinions says consumption should be read in light of demographic factors. The meeting ended on Friday 30 October. September household spending printed at 23:30 UTC yesterday — eleven days later. It fell 6.8 per cent year on year against a 3.1 per cent expected decline, 3.7 points below consensus, a tenth consecutive monthly fall, and on a seasonally adjusted monthly basis −2.8 per cent against an expected rise of 0.3, a 3.1-point swing in the wrong direction. Three readers carry 6.8 in body text, not merely in a headline, and all three carry the 3.1 consensus.
We had that consensus at 3.0 per cent. It is 3.1. The error is a tenth of a point and the mechanism is worth more than the correction: the consensus for this release and the prior month’s actual are the same number. When the expected value equals the previous value, nothing in the row tells you which field you copied. That is the sixth consecutive week in which the defective field on our own table was not the one we were checking, and this instance is the cleanest yet, because the two fields were indistinguishable by inspection.
Now the part that matters for what you do with the print. The internal affairs ministry official attached a cause to it: poor weather kept people from going out, with Tokyo recording precipitation on every day of the month, reportedly a first. One aggregator instead narrates the fall as the impact of persistent inflation and higher borrowing costs, attributing that to nobody. Those are not two readings of one cause; one is the issuer’s attribution and the other is a house narrative, and the standing instruction is to keep the level and cut the because. The components are not evenly weather-shaped either: education −23.5 per cent, utilities −10.5, transport and communication −9.1 from plus 0.2, culture and recreation −8.7, healthcare −7.6, food −2.5. Education falling 23.5 per cent is not a rainy-day effect. We do not know what it is, and we are not going to invent it.
Tonight’s print is the one with a consensus, and it has a hole in it
US October CPI lands at 13:30 UTC. The consensus is 0.3 per cent month on month headline and 0.2 per cent core, now on five independent readers after a Tokyo house confirmed both overnight, against September prints of 0.6 and 0.3. The release will not carry a year-on-year rate for the headline index, because October 2025 was never collected during last year’s lapse in appropriations — the Bureau states this on its own page and the Tokyo house states it too.
The consensus dispute is in the decimals and it is ours to name. Six named-house forecasts run 0.20 to 0.30 on the headline, mean 0.272, and one bank identifies the survey consensus as 0.29. Five of the six round to 0.3. So the 0.3 that four calendars store is a rounding of something closer to 0.29, and a 0.26 print is three basis points from the real consensus while reading as a four-basis-point miss against the stored one. The spread between houses is ten basis points. The rounding is wider than the disagreement it conceals, which means a filter computing actual minus consensus on this release is measuring its vendor’s display precision at least as much as the data.
We also owe a partial retraction. At this slot yesterday we criticised a calendar for carrying an uncaveated 3.9 per cent year-on-year consensus for a month with no year-on-year field. One bank names a reproducible method for a synthetic annual rate — average the September and November 2025 index levels — which yields 3.9 per cent headline and 2.7 per cent core. The 3.9 is not invented. It is a particular synthetic computed by a stated method, and the calendar’s defect was the missing label, not the number. We were right that you cannot compute a surprise against it and wrong to imply it came from nowhere.
What this does not tell you
The absence of the currency from this document is an absence from an edited, length-capped summary, and that is a weaker claim than it sounds. The issuer’s own footnote says each member summarised their own remarks within a set character limit and the Governor, as chair, edited them by item. So we can say the yen does not appear in the published Summary of Opinions. We cannot say no member mentioned it in the room. The minutes would settle that, and the October minutes publish on Wednesday 23 December — five days after the 17–18 December meeting has already decided. Any rule that waits for the minutes is structurally one meeting late by design.
The six-versus-two-versus-three split of the policy section is our classification, not the document’s. One Tokyo publisher reads the same fifteen bullets as six pro-hike and four cautious where we read six, two explicit holds and three cautions; the pro-hike count matches and the cautious count does not, which is what you should expect when two readers classify unlabelled text. Treat the six as firm and the rest as a reading.
We have no December hike probability from a named feed. One Tokyo house gives the low 80s on one desk and roughly 80 per cent on another, forty-six minutes apart, neither naming a feed; no other page we opened gives one at all. That is the tenth consecutive confirmation of a standing caution and we would rather print the absence than launder one house’s two estimates into a market probability. We also carry, unpublished and unverified for a fourth asking, a single-sourced Michigan November preliminary sentiment reading of 51.0 from 48.0 with one-year inflation expectations at 4.3 per cent against 4.6 expected — one reader, no second, and we are printing it here so that it is either corroborated or dropped rather than carried a fifth time. Finally, the neutral-rate language is qualitative: one opinion puts neutral considerably above 1 per cent, another says the rate is already inside the estimated neutral range, and the document gives no number for that range.
