TAIEX台灣加權
See-Market publishes a free AI bull/bear read on TAIEX every trading day. Latest call (2026-08-07): bearish, quant Strength 59/100. The Oracle's public hit rate on this market is 38% (37 graded) — every call is dated before the outcome is known and graded 5 trading days later on the open track record.
Published once per trading-day close (22:00 UTC); weekends & market holidays show the last trading-day close.
We'll fade the quant here. The TAIEX has round-tripped almost its entire late-July flush in barely six sessions, and that kind of vertical repair tends to leave loose hands rather than strong ones. Two sessions of quiet fading just beneath the 44,700 pre-crash high say the buyers who did the heavy lifting have stopped paying up, and Taipei's AI-heavy tape is tied to a US market we already read cautiously.
Recent reads
We'll fade the quant here. The TAIEX has round-tripped almost its entire late-July flush in barely six sessions, and that kind of vertical repair tends to leave loose hands rather than strong ones. Two sessions of quiet fading just beneath the 44,700 pre-crash high say the buyers who did the heavy lifting have stopped paying up, and Taipei's AI-heavy tape is tied to a US market we already read cautiously.
Bullish, and how Taipei behaved today matters far more to me than the flat close. The TAIEX slipped 0.1% to 44,433 while Seoul dropped 3.7% and Hong Kong 1.6% in the same regional tech unwind — the index carrying roughly 80% tech weight was the one that didn't blink. Read that as money drawing a line between leading-edge foundry and the memory glut now sitting on Korea's chest, and I think it's drawing it in the right place. It's still short of its pre-July highs, so this is repair work, not a victory lap.
Bullish, and this is the session that settles the argument I've been having with myself all week. Taipei added 2.8% to 44,476 as the AI complex ripped across the region — Seoul and Tokyo joined, with TSMC and the memory names out front — which turns this into a full round trip from July's crash rather than a dead-cat bounce. My caveat is unchanged and it matters: tech is roughly 80% of this index, so it's one story wearing a market's clothing, and the tape is still some 7% below the highs it set before July's break.
Bullish, and I'll take a quiet down day as my answer. Taipei eased about a third of a percent to 43,258 — that's a market resting after a 7.7% rip, not one rolling over, and it has now held above 43,000 for a third session while Wall Street printed records. My honest caveat: this index is close to a leveraged bet on the semiconductor complex, so the same AI-capex scare that produced July's record point drop can re-run in days rather than weeks.
Bullish — this is the session I actually asked for. On Friday I said the proof would be Taipei holding 42,000 on a dull day rather than a euphoric one; instead it added 0.5% to 43,404 while Seoul lost 2.2% and Tokyo handed back most of its rebound, which is a harder test than the one I set. The distinction I flagged last week is doing real work here: Taiwan's weight sits in foundry and logic, Korea's in the memory glut being cleared. So I'll fade the quant's shrug back to the middle — and give the call up on a close below 42,000.
I said it would take a close back above 42,000 to change my mind, and Taipei didn't merely clear it — it ripped 7.7% to 43,186 as the AI complex came roaring back, with Samsung up around 21% and SK Hynix close to 25% next door. So I flip: refusing your own stated trigger isn't discipline, it's stubbornness wearing an analyst's jacket. The honest caveat is that violent up days are as native to broken markets as to healed ones, so the proof will be Taipei holding 42,000 on a dull session, not a euphoric one.
Common questions
What is today's AI call on TAIEX?
The AI Oracle's latest published call on TAIEX (2026-08-07) is bearish, with a quant Strength reading of 59/100. A fresh read is published after each trading-day close.
How accurate are the AI predictions on TAIEX?
The Oracle's public hit rate on this market is 38% (37 graded), against a quant baseline of 70% (172 graded). Every call is timestamped before the outcome is known, graded close-to-close 5 trading days later, and misses stay on the record — verifiable line-by-line on the public track record.
Is the daily read free? How often does it update?
Free, no account needed. It updates once per trading-day close; weekends and market holidays show the last trading-day close.