Nvidia beat again, and everyone is hunting the next Asian chip trade. Last week we spent three days testing 20 Taiwanese stocks for our product. Verdict: there is nothing to pick — and the reason is the interesting part.
On Wednesday night Nvidia delivered another blowout quarter. By Thursday morning the headlines pointed one way: Asian chip stocks, lifted. And everyone asks the same question — what do I buy next? The natural candidate is Taiwan. It is where Nvidia’s chips are actually made.
We went to Taiwan with exactly that thought. Twenty stocks, two years of prices, 113 combinations — graded with the same exam that our Korean coverage passed. (The data arrives with its own charm: dates come in the Republic of China calendar, year 115. We adjusted.)
Here is the result up front: one pass out of 113. But the reason for the failure is not “no signal.” It is that the signal is too strong. What does that mean?
Your instinct is right, by the way. When US chips move, Taiwan really does follow — the strongest link we have ever measured. The problem is what comes after.
Every fruit, it turned out, came from the same tree.
One tree
The Korean market is a fruit stall. There are apples (chips), pears (autos), tangerines (banks). So “the apples ripen first on US chip news” is information worth paying for — and in Korea it is measurably true: the chip sector catches the US signal harder than the market does. That gap is what we sell.
Taiwan’s stall turned out to be different. Every fruit came from the same tree. One company is a third of the index, and the rest are its branches — designing its chips, packaging them, assembling them into servers. When it rains in America, no single fruit gets wet. The whole tree does.
In numbers: the US chip index explains Taiwan’s whole market at 0.293. The best-explained single stock manages 0.244. Not one stock beats its own market. The question “which one do I pick” does not exist in Taiwan. Selection carries no information.
We wondered whether smaller names might differ, so we pulled five small-caps from Taiwan’s junior exchange and re-ran everything. Worse — half the link, and unstable year to year: eleven of twelve pairs failed the out-of-sample test.
| Same window, same method | Sector / best stock | Its own market | Verdict |
|---|---|---|---|
| Korea (chips) | 0.200 | 0.161 | beats the market ✓ |
| Taiwan (best of 20) | 0.244 | 0.293 | loses to the market ✗ |
Half past three in Seoul
Then what about the other trade? Taipei closes an hour before Seoul. Could Taiwan’s afternoon tell you about Korea’s tomorrow?
Our ledger’s answer: Seoul already knows. Measure Taiwan’s close against Korea’s close on the same day and the link is real — 0.36. Let one night pass and it collapses to 0.01. Whatever Taipei learned by 2:30, Seoul has priced by its own 3:30 bell. A one-hour head start is not a product. It is a rounding error.
The two drawers we haven’t opened
To be precise about what died here: Taiwan’s tape. Two drawers remain closed. One is the customs ledger — the method that earned Japan its place in our product. The other is TSMC’s monthly revenue disclosure, the one fundamental signal between quarters. We started collecting it last week, dated, not backfilled. In a few months there will be a sample worth testing.
The news says buy Taiwan. Our ledger says: in Taiwan, picking is the losing move — the market is the stock.
In Korea, the sector beats the market. In Taiwan, there is no market to beat.
How this verdict dies
Our two-year window was one long AI upcycle. If a downturn arrives and Taiwanese stocks start falling differently — fruits ripening apart — we re-run everything. And if TSMC’s monthly revenue, once the sample exists, turns out to move Korean or Japanese names, half of this verdict retires that day. Either way it goes in the ledger, dated.
Method: next-trading-day returns, market controls, walk-forward split, three-part pass criterion (fit, out-of-sample, beats-the-market). All 113 results, the reproduction script and file hashes are archived; the full table is available on request. Windows, FX and a proxy for Japan are known limits and listed with the results.
Data: TWSE and TPEx open government data (政府資料開放授權條款), US prices from our daily collection, Korean prices from KRX official. Findings may be cited with attribution to Asia Pulse (asia-pulse.com).