Asia Pulse
Research/Ledger

Your Japan trade is pricing the wrong country

July 28, 2026
NEW YORK · THE TAPE GREATER CHINA · THE INVOICES TOKYO 56–66% OF CHIP EXPORTS · US SHARE ~2%
Key numbers
0.55how hard Tokyo's next session follows the US close — 10 years, 2,621 sessions, walk-forward validated
56–66%Greater China's share of Japan's chip exports, every year since 2021 — never below half
~2%the US share of those exports. Falling.

Psychology trades New York. Revenue ships to Taipei.

No developed market is hotter than Tokyo right now. The Nikkei is up roughly 45% in a year and sits near all-time highs. This week decides whether the rally deserves it: Advantest reports Tuesday, Tokyo Electron Wednesday, and Kioxia — up some 800% this year — follows next week. Western money has poured in to buy “Japan, the AI supply chain,” and the fiscal backdrop has poured fuel on top: defense spending at a twelfth consecutive record, the Bank of Japan at its highest policy rate in three decades, the 30-year JGB above 4%, and Japanese institutions selling US Treasuries by the tens of billions per quarter as money comes home. Japan is, for the first time in a generation, a market the world has to read every morning.

So ask the trader who bought this rally one question: do you know where your Japanese chip company's revenue comes from?

Most will say America. AI is an American story; Nvidia sets the weather and Tokyo follows the shadow. And if you only watch the tape, that answer looks right. We measured ten years of it — 2,621 session pairs — and the pull is real: when the US market moves, the Nikkei follows the next session with beta 0.55 and a 67% directional hit rate. Out of sample the link got stronger. It is one of the cleaner cross-market transmissions we have measured — tighter than our US-to-Korea semiconductor channel.

So far, the consensus survives contact with the data.

Then we opened the customs ledger.

What the ledger says

Japan publishes its trade statistics down to nine-digit commodity codes, monthly, by destination — public-domain government data, the same class of source as the Korean customs ledger we already run. We pulled six full years.

Japan's semiconductors do not ship to America.

0% 50% 100% 20212022 20232024 20252026* Greater China United States
Share of Japan HS-8542 exports by destination, customs-confirmed. Greater China = Taiwan + China + Hong Kong. *2026 covers January–May. Schematic; exact per-country series withheld.
YearGreater China shareUS share
202162%3.3%
202259%3.1%
202356%3.1%
202460%2.8%
202563%2.1%
2026 (Jan–May)66%~2%

Taiwan alone takes a quarter to nearly a third of Japan's chip exports in any given year. Add China and Hong Kong, and Greater China absorbs between 56% and 66% — through a chip glut, an AI boom, and an export-control regime. The US share has never reached four percent in this window, and it has been drifting down — even as the Nikkei's sensitivity to New York has not budged.

Hold those two facts next to each other. They describe two different Japans.

Two channels, one mispricing

The US close is Tokyo's psychology channel. It sets the next morning's tape with mechanical reliability. Ten years of sessions say so.

Greater China is Tokyo's revenue channel. It is where the invoices go. Six years of customs filings say so.

NEW YORK TAIPEI · SHANGHAI TOKYO psychology · the tape · β 0.55, next session revenue · the invoices · 56–66% of chip exports to the US · ~2%
Two channels, one market. The tape prices the dashed line; the invoices travel the solid one.

These channels can disagree — and the days they disagree are the days the tape carries information. When New York sells off but Taiwanese orders are firm, Tokyo's morning gap down is psychology mispricing revenue. When New York rallies while mainland demand rolls over, the gap up is the same error in reverse. You can only see the divergence if you are watching both channels. Most Japan exposure watches one.

This week is a live experiment. The earnings calls at Advantest and Tokyo Electron will be scored, in most coverage, against an American question — how big is AI capex?

The ledger says the sharper question is the one almost nobody will ask: what are Taipei and Shanghai ordering?

The contrast with Korea

Korea does not work this way, and the difference is worth a moment. In Korea, US transmission is sector-specific: Micron moves SK Hynix (R² 0.165) while healthcare barely responds (R² 0.040). In Japan, at the index level, we found no semiconductor-specific channel beyond the market-wide one — the US close moves the whole Nikkei, not chip names in particular. Same American shock, one strait apart, two different propagation structures.

(We tried to establish a Japan sector-specific channel and could not, with licensable data. That is a measurement boundary, not a proof of absence — we would rather say so plainly than dress it up.)


The transmission coefficient is derived from licensed research data; we publish the coefficient, never the underlying index values. The trade shares come from public-domain customs statistics. Nine-digit commodity detail, monthly series, and per-country breakdowns beyond the table above are deliberately not in this post.

Data: Trade Statistics of Japan (Ministry of Finance / e-Stat), Government of Japan Standard Terms of Use. Findings may be cited with attribution to Asia Pulse (asia-pulse.com).

Discover more
Research / Methodology
Six hypotheses we rejected in Korean biotech
The transmission coefficient works in semiconductors. It does not work here — and the reason turned out to be the mechanism itself.