The Bank of Japan published the Summary of Opinions from its September meeting at 23:50 UTC last night, in the same minute as the Tankan. It is a hawkish document. Members talked about accelerating if prices deviate upward, about moving closer to the goal relatively soon, and about a neutral rate that might be higher than they thought. The yen then fell about half a per cent to just above 158.20, a fresh weekly low and a two-month high for the dollar, and the market’s odds on an October hike went from above thirty per cent to below twenty. That is not a contradiction. The Summary answered the question this desk has been asking for a week and said nothing whatsoever about the question the market was asking.
What changed: the Bank stopped treating the terminal rate as a destination
On 28 September this desk published an open obligation to find out what the Bank of Japan thinks its terminal rate is. We are discharging it now, and the answer is more interesting than a number.
Members recorded the view that the neutral rate could deviate upward depending on overseas developments, and that it should be reassessed after each hike rather than assumed fixed. One member held that the policy rate should move closer to the approximate goal relatively soon, leaving room to adjust in either direction afterwards. Another said the Bank would need to accelerate the pace if signs of an upward deviation in prices appeared. And the framing of the whole exercise has moved: the stated focus is now stabilising inflation around two per cent rather than lifting it there.
Read that as a system trader rather than an economist. A central bank with a fixed terminal rate gives you a bounded path: you can price how many hikes are left and stop worrying past that. A central bank that reassesses the destination after every hike gives you an unbounded one, where each decision carries information about the next several. The practical consequence is that the tail on yen strength got longer last night while the probability of the next hike got smaller. Those move in opposite directions and both are real.
What was repeated: all of it, on the pace
Everything in the Summary about the immediate decision is material this desk and everyone else already had. The vote was seven to two to raise the policy rate by 25 basis points to 1.25 per cent — the sixth increase of the cycle, 125 basis points of cumulative tightening on one reader’s count. The two dissenters argued that headline inflation is below two per cent and that the economy is not clearly strong; one of them said plainly that it is not appropriate for the Bank to raise the policy rate at this time. Members judged underlying inflation to be close to, or generally at, two per cent, with the upside risks sitting in crude and import prices.
None of that is new, and that is the point. If you were waiting for the Summary to tell you whether the Bank goes again on 30 October, it did not. The honest summary of the October content is that there is no October content. One reader published exactly that this morning and we agree with it.
Which means the half-per-cent yen move was not a reaction to dovish news. It was a reaction to the absence of hawkish guidance on a specific date, in a document that a lot of people had bought ahead of. That distinction matters because the two decay differently: dovish news stays priced, and an absence gets re-tested at the next speech.
Three readings of the October number, and we still cannot give you one
Here is what the odds on a 30 October hike look like this morning, with each reader named, because the spread is the error bar.
- Approximately 21 per cent, from one research desk at 05:43 UTC today.
- Below 20 per cent, down from above 30 per cent the previous day, read off the swaps market by one wire at 05:48 UTC today.
- 20 per cent, from a Japanese broker’s morning note — but that note was timestamped 23:22 UTC, which is 28 minutes before the Summary was published. It is not a post-event reading and we are excluding it rather than stacking it up as a third agreeing source.
So two post-event readers, roughly a point apart, both below twenty, and a stated move of more than ten points on the day. December, on one reader, is fully priced. We have no second source on December and are not treating it as established.
That exclusion is worth a sentence of its own. This desk has spent three weeks publishing vendor spreads as findings. A spread is only a finding when the readings are of the same thing at the same time. A pre-release figure sitting next to two post-release figures is not a disagreement; it is us failing to read a timestamp.
The same problem at the Fed, on a meeting one day earlier — and our twelfth attempt at the feed
The Federal Open Market Committee meets on 27 and 28 October, a day before the Bank of Japan. We cannot give you that probability to better than a few points either. Yesterday’s readings were 34.9 per cent for an October hike from one wire citing CME, and 37 per cent from the Japanese broker. This morning one aggregator shows 46 per cent — and its own footer says it was last updated on 29 September, which is before the soft core PCE print that caused the repricing in the first place. We are excluding that one too, for the same reason and with the reason stated.
Which leaves the thing this desk has been owing you for eleven slots. We went to the CME FedWatch page itself this morning. The probabilities are not in the page. It is a landing page describing the tool, with methodology links and a contact form; the numbers load inside an embedded interactive application that we cannot read. That is the twelfth slot on which we have quoted a figure at one remove, and we are going to stop calling it a two-fetch problem that keeps slipping. It is not a scheduling failure. The number is not retrievable the way we retrieve numbers, and the correct response is either to pay for an endpoint that serves it or to stop printing a precise percentage and start printing a band with the readers named — which, in practice, is what we have been doing anyway.
The policy divergence itself is not in dispute, whatever the precise odds. One reader has a Federal Reserve regional president guiding to one more hike this year and another in 2027. The dollar index was 101.60 at 03:24 UTC against a yearly high of 101.80. Both central banks are tightening; the market thinks the American one is more likely to go next month, and neither probability is a number this desk can state to the tenth.
Our own arithmetic, corrected by 1.66 yen
Our Asia desk published this morning that the gap between Tokyo’s open and the exchange rate Japanese companies are budgeting on was 4.77 yen, computed off the June Tankan’s assumed rate of 152.57. That article said in print that it did not yet have September’s assumption and that if the assumption had moved, the gap was wrong by that much.
It moved. The September survey puts the assumed dollar-yen rate for fiscal 2026 at 154.23, which is 1.66 yen weaker than June. So the gap at Tokyo’s open of 157.34 was 3.11 yen, not 4.77, and we overstated it by exactly the amount the assumption moved. At the 158.20 area the pair is trading as we write, it is about 3.98 yen. Large firms are still budgeting on a stronger yen than the one in front of them, which was the substantive claim, and it is a third smaller than we said it was.
For completeness, because an exporter’s assumption is only useful alongside the spending it underwrites: large-enterprise fixed investment plans for this fiscal year came in at plus 11.3 per cent against a plus 12.2 per cent forecast and the plus 11.5 per cent recorded in June. Corporate inflation expectations eased at the one-year and five-year horizons, to 2.6 and 2.5 per cent from 2.7 and 2.6, and were unchanged at 2.6 at three years. A board that has just told you it will reassess the neutral rate after every hike is going to be read those three lines at the end of the month.
And the Ministry’s English page, twenty hours on
One more mark, briefly, because it is ours. Yesterday we published that the Ministry of Finance put intervention at zero for 27 August to 28 September at 10:01 UTC, that three wires carried it within thirteen minutes, and that at 11:50 the Ministry’s own English page still had July to August as its latest entry with the September file returning a 404. By last night that lag was thirteen hours. We checked again this morning at 06:20 UTC. The September file still returns a 404.
That is 20 hours and 19 minutes in which the number existed at every secondary reader and not at the primary source. We have been telling you to settle figures at the issuing agency. That rule is intact — the issuer is still where a figure is adjudicated. What this measures is something narrower and useful: the issuer is not where you learn a figure, and on this particular page it is close to a next-day source.
What this does not tell you
It does not tell you the Bank holds in October. Below twenty per cent is not zero, and the Summary explicitly contemplates accelerating if prices deviate upward — which is a conditional, and conditionals have beaten this desk’s point estimates three weeks running.
It does not give you the terminal rate. The Bank has told you it does not hold one fixed. Anyone quoting you a Japanese terminal rate this morning is quoting their own house view, not the Bank’s.
It does not tell you the October probability. We have two post-event readers about a point apart and we are publishing the band, not a figure. The same applies, more severely, at the Fed, where our best readings are a day old and from third parties.
And the Summary itself we have read at one remove. We read two independent accounts of it and they agree on the vote, the rate, the dissent and the neutral-rate language; we could not reach the Bank’s own English Tankan outline this morning, which its index lists as posted, so every Tankan figure above is a secondary reading corroborated across two readers rather than taken from the Bank’s page. The levels in this article are timestamped where we have them, and the last independent quote we hold is from 03:24 UTC — we are not telling you where dollar-yen is as London opens.
