Japan’s September labour cash earnings rose 4.6 per cent year on year, the fastest annual increase since January 1997, and real wages accelerated from 1.6 to 2.2 per cent against a 1.3 per cent consensus — a ninth consecutive positive month, the longest run since the February 2016 to January 2017 stretch. That is the number a hawk on the Bank of Japan’s board would reach for, and at 23:50 UTC tonight the Bank publishes the Summary of Opinions from its 29–30 October meeting, where we will find out whether anybody did. Before that happens, read the rest of the release. Total cash earnings went to 4.6 per cent because special payments rose 32.6 per cent year on year after rising 1.8 per cent in August — and the Labour Ministry attributed that jump to a statistical distortion rather than to bonuses actually being paid. Every pay measure that reaches a household every month decelerated in the same release.
The 29-year high is one line of the release, and the issuing ministry says that line is broken
The headline is real and it is well corroborated. Nominal total cash earnings per worker came in at 311,409 yen — three independent readers give that figure to the yen — up 4.6 per cent on a year earlier against a revised 3.9 per cent in August, and three of them name January 1997 as the comparison, twenty-nine years and nine months back. The data are preliminary and cover businesses with five or more employees. Real wages, deflated, went from 1.6 to 2.2 per cent where the consensus we can find was 1.3, so the real print beat by 0.9 of a percentage point. Nine months positive in a row on four readers, against a twelve-month run of falling real wages through the whole of 2025. On the face of it this is the wage-price loop the Bank of Japan has spent three years saying it needs before it can normalise with confidence.
The composition is where it falls apart. Special payments — the bonus-and-irregular line — rose 32.6 per cent year on year, against a revised 1.8 per cent the month before, on two independent readers. A 30.8-point acceleration in one month in one component is not a labour-market event, and the labour ministry did not pretend otherwise: it attributed the size of the increase to a statistical distortion rather than to a genuine surge in bonuses. Two independent readers carry that disavowal, one calling it a statistical distortion and the other a statistical anomaly, both attributing it to the labour ministry and neither quoting it directly. But note what that means for a system: the agency that publishes the series published a caveat against its own headline, and the caveat is in the prose, not in a field.
So the structure of the release is a headline that will be quoted and a disclaimer that will not be. If your news handling stores “Japan Average Cash Earnings y/y: 4.6%” and nothing else, you have stored the part the publisher disowned.
Every pay measure that feeds consumption slowed in the same release
Strip the bonus line out and the September report is a deceleration across the board:
- Scheduled base pay: 3.8 to 3.7 per cent
- Regularly paid earnings: 3.9 to 3.6 per cent
- Overtime earnings: 4.9 to 3.6 per cent — three independent readers
- Real regularly paid earnings: 1.6 to 1.1 per cent
- Inflation deflator excluding imputed rent: 2.2 to 2.4 per cent
Four of five moved the wrong way for a central bank looking for durable wage pressure, and the fifth — the deflator — moved the wrong way too, because it rose. Real regularly paid earnings, which is the cleanest available measure of what a salaried household’s recurring income is doing after prices, halved its growth rate from 1.6 to 1.1 per cent in the month the headline hit a 29-year high. Those two sentences describe the same release.
Two rows cut the other way and we are not hiding them. Same-sample scheduled pay for full-time workers rose from 2.9 to 3.1 per cent, which is the composition-controlled series and therefore the one that is hardest to dismiss. And part-time hourly scheduled pay rose 5.4 per cent against 4.7 the month before. But the same-sample part-time row was flat at 4.7 per cent — so the acceleration in the headline part-time rate is a change in who is counted, not a change in what anyone is paid. One of those two rows survives a composition control and one does not, and only one of them gets published as a number.
The trade this sets up is not directional, because a wage print nine weeks old does not move a pair tonight. It is a filter question: if you hold a rule that treats Japanese wage acceleration as a yen-positive policy signal, this release would have fired it, and the release’s own internals say it should not have.
We were one draft away from correcting a number that was right
This desk has carried the September wage print in its working notes since Friday, unpublished, in the form “plus 4.6 per cent year on year versus plus 3.8 per cent expected.” This morning, reading one publisher’s full table for the first time, we found 3.8 per cent sitting in it as August’s scheduled base pay — two rows from the headline. The obvious conclusion was that our note had lifted a consensus out of the wrong row, and we drafted that correction.
Then we second-sourced it, and the correction is wrong. Two independent readers give 3.8 per cent as the market expectation for nominal total cash earnings, which means the print beat its consensus by 0.8 of a percentage point and our original note was right all along. The coincidence is real and it is nasty: 3.8 appears in this release twice, once as the forecast the headline beat and once as the prior month’s base pay, in adjacent rows, with no label on either that would stop you swapping them.
We are publishing this because the near-miss is more useful than the catch would have been. A correction is read with more authority than the claim it replaces, so it needs at least as much evidence and usually more — and we were about to issue one off a single publisher’s table, which is precisely the failure we thought we had found, pointed the other way. The rule we are writing down: never correct a figure on one source. Second-source the correction before the correction, not after.
The real-wage consensus is a different matter. The 1.3 per cent that the 2.2 print beat appears on one page only; three other readers of the same release give no real-wage expectation at all. So the nominal beat of 0.8 points is two-sourced and the real beat of 0.9 points is one-sourced, and we label them differently for that reason and no other.
What Monday actually paid, and what 23:50 can and cannot tell you
Dollar–yen closed New York around 157.91 on one Tokyo house’s read, down about 0.4 per cent. A second reader — a US market-review page — has the pair at 158.00, nine pips higher, and prints it under the label “JPY/USD”, which is the reciprocal of the quote it is giving. Its own change figure of −0.42 implies a Friday close of 158.42, which is below all four of the Friday closes published last week (158.457, 158.48, about 158.50, 158.530). We are using 157.91 and treating the other page as low weight on FX for exactly that reason.
Against those four Friday closes, Monday was down 54.7 to 62.0 pips, or 0.345 to 0.391 per cent — and the pair is now 69.0 to 78.0 pips below the seven readings of the 200-day moving average in our file, which span 158.60 to 158.69. New York’s own range was 157.665 to 158.187, 52.2 pips, with the close 24.5 pips off the low. Today’s published forecast range from the same Tokyo house is 157.200 to 158.600: 140 pips wide, with the close sitting 71.0 pips above the floor and 69.0 below the cap. A band centred on the last price to within two pips is a statement that nobody has a view, and it should be read that way rather than as support and resistance.
The cause of the yen’s 55-odd pips is named and it is not Japanese. December WTI settled at $86.85, down $3.98 or 4.38 per cent, and Brent settled at $92.57, down $5.75 or 5.85 per cent — from above $112 at the start of this month, a fall of $19.43 or 17.35 per cent in seven sessions. Two readers attribute Monday’s leg to the US president telling a broadcaster that renewed strikes on Iran are “highly unlikely” if negotiations progress; one adds a report that arrangements are being made for inspectors to return. US equities took it straight up — Dow 52,389 and up 1,104.20 points or 2.15 per cent, the 500-stock benchmark 7,549 and up 2.34, Nasdaq 25,253 and up 2.74, Russell 2000 2,840 and up 3.15 — while the ten-year went the other way, to 4.874 or 4.88 per cent on two readings, from 4.86 late Friday. Lower oil, higher stocks, higher long yields, softer dollar: a growth-relief trade, not a rate trade.
And on the second day of the largest yen-buying intervention disclosure on record — 7,975.6 billion yen on 30 July, which this desk published yesterday — the main Japanese FX news index carries no follow-up item at all in its twenty-five most recent headlines. No intervention headline, no Ministry of Finance headline, twenty-three hours after the release. We said yesterday in print that we had no market reaction to the disclosure. We looked again and we still have none, and the only directional yen view on that index points the other way: a New York-hours note headlined as seeing scope for the yen to strengthen towards 155 over coming months. Its page would not open, so we are publishing that the view exists and nothing about its argument.
Our pre-registration for 23:50, so the next session can mark it. This desk’s own carried notes say two board members favoured a hike at the October meeting — one source, five sessions old, never published and never verified. So: we expect the Summary to contain at least one opinion explicitly arguing for a near-term increase in the policy rate, and we expect wage developments to be the stated ground for it. This is losable in two distinct ways, and we want it marked both ways. It fails if no opinion argues for an increase at all, which would kill our carried note outright. It also fails if the hike arguments are there but rest on the currency or on import prices rather than on wages — and given what the wage release’s internals actually say, that second outcome would be the more interesting loss.
What arrives at 23:50 UTC is a document, not a number. The Summary of Opinions from 29–30 October is a short, unattributed list of what board members said; it has no consensus, no field and no surprise. It cannot be scored against a forecast. It can be read for whether anybody argued for a hike and on what condition — and if someone did, tonight’s wage internals are the thing to hold it against.
What this does not tell you
We did not read the Ministry of Health, Labour and Welfare’s own release. Every wage figure above is a secondary reading of it. The bonus disavowal has two readers who use two different words for it — distortion and anomaly — and neither reproduces the ministry’s own sentence or names the ministry in full. If the ministry’s wording is narrower than either, our first section overstates its case. A fourth reader of the same release says nothing about bonuses at all and attributes the real-wage rise to spring pay settlements and slower inflation, which is the opposite emphasis.
We have no named feed for the December Fed probability tonight. Two readings exist — the low 80s and about 80 per cent — and both come from the same Tokyo house, from two different desks half an hour apart, and neither names the feed it is quoting. That is one source, not two, and we are not publishing a probability on it.
Monday’s session high and low are from one house. Its two desks gave the New York low as 157.665 and “around 157.67”, which is the same desk twice, not two houses. The full-day range is unknown to us: we have New York’s 52.2 pips and an intraday path from a second publisher — about 158.03 at 07:44, below 158 at 08:12, about 157.88 at 10:38, into the 157s at 14:34, about 158.00 at 16:33 UTC — and those are consistent with each other but do not establish a high.
We cannot tell you whether the 30-year Japanese government bond auction our own week-ahead places at 03:35 UTC today is actually happening. Neither of the two Japanese calendars we read this morning carries it. One aggregator showing nothing is weak evidence; two is better and still not proof, and we have not reached the issuer. Treat that slot as unverified rather than empty.
Related
- FX events calendar — the 23:50 UTC Japan block and tonight’s 13:30 US CPI
- Signals — how a wage print reaches a position, if it should at all
- EA presets — news-window handling for Tokyo-hours releases
