The KOSPI ended the week up 0.63%. The KOSDAQ fell 3.87%. Same country, same four sessions, six percentage points apart — and foreigners sold both.
This is Foreign Flow Watch, our weekly read of where global capital actually moved in Korean equities. Every figure below is computed from settled exchange data for all 2,554 listed common stocks, converted at each session’s own close. Week of 18–21 August 2026 — four sessions, not five: Monday 17 August was a substitute holiday for Liberation Day.
📉 The index split
| Index | 18 Aug | 21 Aug | Change |
|---|---|---|---|
| KOSPI | 6,869.83 | 6,912.95 | +0.63% |
| KOSDAQ | 834.20 | 801.94 | −3.87% |
Three investor groups, four sessions, one direction each:
| Group | Net (KRW) | Net (USD) |
|---|---|---|
| Foreigners | −₩2,033.2bn | −$1.46bn |
| Institutions | −₩2,798.9bn | −$2.01bn |
| Retail | +₩2,553.1bn | +$1.84bn |
Converted at each session’s own USD/KRW close. Week-end reference rate ₩1,390.79/$ on 21 Aug, from FinanceDataReader. We state the method because a single week-end rate would misprice four different days.
Domestic retail absorbed ₩2.55tn of what foreigners and institutions sold. That is the whole reason the KOSPI closed green on a week of ₩4.8tn in combined foreign and institutional selling.
💰 The split ran through one conglomerate
Foreigners were net sellers of Korea. They were not net sellers of Samsung Electronics.
| Stock | Foreign net | Price, 18→21 Aug |
|---|---|---|
| Samsung Electronics (005930) | +₩706.5bn | +4.84% |
| Samsung Electronics preferred (005935) | −₩366.9bn | +9.93% |
| Samsung Electro-Mechanics (009150) | −₩900.5bn | −8.61% |
| Samsung Life (032830) | −₩93.7bn | +5.80% |
Samsung Electro-Mechanics was the week’s single largest foreign sell across the entire market — larger than the Samsung Electronics buy. The component maker fell 8.61% while Samsung Electronics rose 4.84%.
The obvious reading is that one is a supplier to the other, so we checked the size of that link. Samsung Electro-Mechanics’ 2025 annual report names Samsung Electronics and its subsidiaries as the only customer above 10% of revenue, at ₩3,093.8bn — about 27% of the total. That is the largest single relationship and it is still roughly a quarter of the business. It is not enough to make the parent’s share price a proxy for the supplier’s, and this week the two moved 13 percentage points apart.
Two of those four rows move against the flow, and we are printing them rather than dropping them. Foreigners sold ₩366.9bn of the Samsung Electronics preferred line and it rose 9.93% — the best performer in the table they were selling. They sold ₩93.7bn of Samsung Life and it rose 5.80%. If foreign flow set prices, neither of those happens.
🧊 SK hynix: sold hard, up anyway
Foreigners sold ₩525.5bn of SK hynix over the four sessions. It closed the week up 4.09%, from ₩1,662,000 to ₩1,730,000.
Last week these two names — Samsung Electronics and SK hynix — took 73% of everything foreigners put into the KOSPI. This week the pair nets to +₩181.0bn, because the ₩706.5bn into Samsung was nearly cancelled by the ₩525.5bn out of hynix. The concentration did not just shrink. It split in two.
🔻 Where the KOSDAQ went
Foreign selling was not evenly distributed across the two boards:
- KOSPI: −₩1,393.9bn foreign net
- KOSDAQ: −₩315.6bn foreign net
In won terms the KOSPI took four times the selling. In index terms the KOSDAQ fell six percentage points further. A smaller board with thinner books moves more on less money — which is the mechanical part of the answer, and it is most of the answer.
The largest KOSDAQ foreign sells were mid-caps, not a single name: Eo Technics (−₩35.4bn), Silicon2 (−₩30.7bn), Robotis (−₩30.4bn), ST Pharm (−₩25.5bn), Seojin System (−₩22.6bn). No single stock explains a 3.87% index move.
🧾 Scorecard: how Issue #5 held up
Issue #5 published four conditions that would prove it wrong. Here is what happened.
| Condition from Issue #5 | Outcome |
|---|---|
| “If Samsung Electronics and SK hynix give back this week’s gain…” | Held. Samsung +4.84%, hynix +4.09%. Neither gave it back. |
| “If the $3.32bn concentration falls sharply while the index keeps rising…” | Triggered. The pair fell from $3.32bn to +₩181.0bn net while the KOSPI rose 0.63%. We over-weighted the two names. |
| “If the KOSDAQ turns down while foreigners keep selling it…” | Triggered. KOSDAQ −3.87%, foreign net −₩315.6bn. Our “held its gain” reading was one week early, exactly as flagged. |
| “The residual remains our weakest number.” | Not tested this week — we did not publish a residual. |
Two of three testable conditions went against us. That is a worse week than Issue #5 had grading Issue #4, and we print it with the same prominence we would print a hit. A scorecard that only appears when we were right is not a scorecard.
🧩 Prove this issue wrong
- If Samsung Electro-Mechanics recovers next week on continued foreign selling, then the −8.61% was not the flow and we have mistaken coincidence for mechanism.
- If the KOSDAQ rebounds while foreigners keep selling it, the “thin book” explanation is doing less work than we claim and something domestic is setting that index.
- If retail net buying reverses and the KOSPI still holds, then “retail absorbed it” was a description of an accounting identity, not of support.
- Our weakest number this week is the preferred line (005935). It is absent from our reference universe, so its name is inferred from the ticker convention rather than read from a listing record. The flow and price figures are exchange data and stand; the label does not carry the same weight.
🎯 Why it matters
A single KOSPI print would have told you Korea was flat-to-up this week. Underneath it, foreigners and institutions pulled ₩4.8tn out, domestic retail put ₩2.55tn in, the small-cap board fell 3.87%, and the largest single foreign sale in the market was a Samsung component maker whose customer rose at the same time.
The headline was not wrong this week. It was just very nearly empty.
Every figure here is reproducible from our open dataset — 2,554 KOSPI and KOSDAQ common stocks, updated every trading day, CC BY 4.0, DOI 10.5281/zenodo.21833374. This issue makes no causal claim about flows; where flow and price disagree, we print the disagreement.