Thursday’s Tokyo session was reported as an AI selloff. It was not one. The Nikkei 225 closed at 61,077.90, down 1,110.73 points or 1.79 per cent — three independent readers agree on all three figures — and on the same session the Topix rose 1.16 per cent to 3,940.39, with 1,127 Prime Market issues advancing against 408 declining. One reader puts a single stock, Advantest, at roughly 1,097 points of that 1,110.73. The index everybody quotes said risk-off; the market underneath it said one name. Then New York did the real version: the Nasdaq Composite fell 5.52 per cent, the Russell 2000 fell 4.45 per cent, and dollar–yen traded 158.50 down to 157.54. The distinction is not pedantry. Thursday was a concentration event, which mean-reverts when the name stops falling. Tonight is a breadth event, which does not.
Thursday in Tokyo was one stock wearing an index’s clothes
The close is settled. Three publishers — a German wire carrying Jiji, a Japan-focused English outlet, and a US aggregator — all print 61,077.90 and −1.79 per cent, and the first two both print −1,110.73 points against a 62,188.63 Wednesday close. That is as clean a three-way confirmation as this desk gets.
What sits inside it is the interesting part. The Topix closed up 45.01 points at 3,940.39, a gain of 1.16 per cent. So the two headline Tokyo indices moved 2.95 percentage points apart in opposite directions on one session. Breadth on the Prime Market was 1,127 advancers, 408 decliners, 11 unchanged — nearly three up for every one down. Toyota rose 16.68 per cent and Nintendo 6.31 per cent on the day the Nikkei “sold off”.
The two contributor arithmetics we can read do not reconcile with each other, and both point the same way. One reader attributes about 1,097 index points of decline to Advantest alone — 98.8 per cent of the entire fall, from one stock. Another says Advantest, Tokyo Electron and SoftBank Group together weighed about 1,800 points, which is 689 points more than the index actually lost, meaning everything else added back roughly that much. Those two statements cannot both be precise, but they agree on the structure: a handful of chip names fell hard and the rest of Japan went up. A third reader gives the percentages — Advantest about −15, Tokyo Electron about −9.5, SoftBank Group about −6.9 — and no reader gives all of it, so take the concentration as established and the exact decomposition as not.
If your Thursday Japan handling triggered on a Nikkei percentage, it fired on one company’s earnings reaction. A Topix-keyed or breadth-keyed rule would have correctly done nothing.
Tonight is the broad one, and the gap into the open is 6.1 per cent on a single source
New York was not concentrated. The Nasdaq Composite closed at 24,315.00, down 1,421.51 or 1,421.52 points depending on the reader, −5.52 per cent — three-way agreement on the level and the percentage, and the point change differs by one hundredth. The Nasdaq 100 fell 6.36 per cent to 26,629.81. The Dow fell 3.44 per cent. The S&P 500 fell 4.35 or 4.36 per cent. And the Russell 2000 fell 4.45 per cent, which is the figure that settles the breadth question: small caps do not fall 4.45 per cent because three semiconductor names are being de-rated.
Against the record closing high of 27,599.89 set on 6 October — two Reuters-wire readers agree on the level and the date — the 24,315.00 close is 11.90 per cent lower. That is past the correction threshold with room to spare, and nobody published it, because every correction story we could open was filed before the close. One put the index at 24,796.06 and called the correction “on pace to confirm”; one put it at 24,712.60 and said “more than 10 per cent below”; a third described it as already pushed into correction territory from an intraday level near 25,500. Three readers, three modal verbs, three levels spanning nearly 800 points, and the close was below all of them. One of those pages also fails its own arithmetic: 24,796.06 against a 25,736.51 prior close is −3.65 per cent, not the −3.2 per cent it prints.
The number that actually decides your Tokyo open is single-sourced and we are labelling it as such. One Tokyo publisher reports CME Nikkei futures trading down to 57,035 and settling at 57,350, quoted in yen. None of the three pages that gave us the cash close carries a futures level at all; the one page whose headline suggested it might would not open. Taking that 57,350 at face value against the 61,077.90 cash close gives a gap of 3,727.90 points, or 6.10 per cent — and 6.62 per cent to the 57,035 low. It would put the open roughly 3,436 points below Thursday’s intraday low of 60,786.19.
A 6 per cent indicated gap is large enough that we would normally refuse to publish it off one reader. We are publishing it with the warning rather than cutting it, because the asymmetry runs the wrong way: if the figure is right and you sized as though it were a normal gap, the mistake is expensive, and if it is wrong you lose nothing by having been ready. Treat it as one publisher’s number and not as a level.
The yen round-tripped, and the line we wrote about on Wednesday was never the structure
The pair did the whole circuit. Through Thursday’s Tokyo hours it fell from about 158.43 to roughly 157.82 and steadied near 157.90. New York took it back up to 158.50, then sold it to 157.54 on one Tokyo reading and 157.53 on another — a one-pip disagreement between two pages from the same publisher, which is the most honest kind. One wire has it closing at 157.64, down 0.49 per cent; a Tokyo publisher independently puts the close about 0.5 per cent lower. By 07:02 Japan time the pair was around 157.70.
Now mark that against what this desk published twenty-four hours ago. We gave Wednesday’s close as 158.46 and the 200-day moving average as three different numbers — 158.60, 158.63 and 158.65 across three Tokyo publishers. Tonight’s 157.64 is 82 pips below our published close and 96 to 101 pips below the lowest and highest of those 200-day readings. The call was right and it has now been right twice running.
But the more useful correction is to our own framing. One Tokyo publisher points out that dollar–yen has been inside the daily Ichimoku cloud more or less continuously since mid-October, with today’s cloud top at 158.44 and the floor at 157.18. If that is right, the 200-day “break” we have now written about twice was a break of one line sitting inside a band — which is exactly why it kept failing. On that reading the live structure is 157.18 to 158.44, tonight’s close sits 46 pips above the floor, and the thing to watch today is not a moving average at all but whether 157.18 gives way. One publisher’s forecast range for the session is 156.900 to 158.500.
Elsewhere: euro–dollar rose from 1.1203 into the 1.1240s. Euro–yen extended its fall to 176.83. The US 10-year yield traded down to the 4.81 per cent area and closed near 4.84, with the 2-year down 11.45 basis points and the 30-year down 2.19 — a 9.26 basis point bull steepening in the 2s30s, which is what a genuine risk-off bid looks like and not what a sector rotation looks like.
Four readers, four causes, and only one of them is a regime
Everyone agrees on the price. Nobody agrees on why, and the disagreement is not cosmetic — the four available explanations have different half-lives.
The Tokyo version is political and causal: one publisher writes that the opposition Democrats’ sweep of both chambers led to the view that AI regulation will advance and data-centre construction will stall, and that this is what took US equity indices sharply lower. A second Tokyo page frames it as wariness of tighter regulation around AI data centres plus concern about an investment slowdown, and notes that 4 November had rallied hard on the reduction in political uncertainty from the same election result.
The Reuters wire inverts the causal direction. It names the cause as AI spending worries and heightened scrutiny of data-centre expansion, and casts the election as the thing whose gains were erased rather than the reason: Thursday’s slide wiped out the previous session’s advance after Democrats swept both houses, as investors weighed the prospect of greater scrutiny of further AI buildout. One of those wire pages adds that Alphabet and Meta earnings failed to reassure investors about returns on AI spending.
A third reader removes the election entirely: Wednesday’s rotation out of AI and into the broader market turned into a broad selloff on Thursday, hitting memory, storage, chip equipment and data-centre suppliers, with one memory name down about 12 per cent and investors questioning how fast the buildout converts into supplier orders. A fourth attributes it to record leverage and weakening sentiment, citing a newsletter, with AI regulation appearing as context.
Those are not four phrasings of one story. A regulatory regime is persistent, cumulative and yen-positive, and you hold through Tokyo. A leveraged positioning unwind in semiconductors is mean-reverting, and you do not. The tape gives you no way to choose between them, and the honest consequence is that tonight is a session to carry smaller size rather than a directional view — not because the direction is unclear, but because the duration is, and duration is what sizing is for.
One of those Tokyo pages adds two Washington personnel rumours as further dollar-negative drivers. We could not second-source either, so we are not naming them, and no figure here depends on them.
The marks you are reading disagree about gold’s sign
Before you reconcile your overnight book, know what the overnight book is made of. Two wires we read tonight both price spot gold within four cents of each other — $4,088.02 on one, $4,088.06 on the other — and disagree about which way it went. The first says up 0.68 per cent. The second says down 0.64 per cent, and gives the dollar change as −26.34, which is internally consistent. Back out the implied previous closes and they are $4,060.41 and $4,114.39: $53.98 apart. One reader thinks gold rose about $27.61 overnight and the other thinks it fell $26.34, and they land on the same price.
This is not a price disagreement. It is a reference-point disagreement, and it is invisible unless you check, because both pages look authoritative and the level agrees. Anything in your system that stores a daily percentage change rather than two timestamped levels inherits whichever reference its vendor chose.
The same fault runs through the equity marks at smaller amplitude. The Dow close is 51,523.06 on two readers and 51,528.16 on a third — 5.10 points — and the implied previous closes differ by 5.20, so again the move agrees and the levels do not. The S&P 500 is 7,353.73 against 7,353.85. West Texas crude settled at $90.52 on the one page that calls it a settlement, 90.54 on another that does not, and that same first page quotes $91.06 elsewhere in its own body — three numbers, one instrument, one page partly responsible.
And the 10-year carries a units trap worth a filter rule: one reader gives the move as −5.12 basis points, another as −1.17 per cent. Both describe the same roughly six basis point decline from 4.894. A rule that thresholds on “change greater than 5” fires on one feed and sleeps through the other.
What this does not tell you
The futures gap is one publisher’s number and the single most consequential figure here. We could not corroborate it, and a 6.1 per cent indicated move deserves better than one source. If it is wrong, most of the second section above is wrong with it.
We do not know Thursday’s Tokyo cash close to everyone’s satisfaction either: one publisher’s headline carries −2.2 per cent against the −1.79 per cent three readers print, and its page would not open, so we cannot say whether the body agrees with its own headline.
The Advantest decomposition is approximate in its own source and contradicted in magnitude by another. We are confident the Thursday fall was concentrated; we are not confident it was 98.8 per cent one stock.
We have no option expiry board for today and no FX implied volatility reading at all — neither wire we read tonight publishes one. Twelve hours ago this desk published one-week dollar–yen implied volatility at 6.23 per cent, the lowest since 22 October, on the tenor covering today’s US payrolls. Between that sentence and this one the Nasdaq fell 5.52 per cent. Treat the 6.23 as stale and do not size off it.
The Japanese 10-year is “below 3 per cent” on one Tokyo publisher and we have no level. And we have not established whether the correction threshold being crossed changes anyone’s behaviour, or is simply a number that newspapers like.
