At This Pace, Korea’s Two Biggest Buybacks Finish Six Weeks Early — Week of September 4, 2026

Foreigners sold $1.72bn of Korean equities in five sessions, and 69% of that net came out of exactly two stocks. Those two stocks fell 0.58% and 0.36%. The 941 other KOSPI companies fell 2.23%. The index number that gets reported — down 1.50% — describes neither group. What it does describe is a market where the largest buyer is once again not an investor at all, and where two Korean-language filings now let you calculate, to the session, when that buyer stops.

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 2,550–2,551 KOSPI and KOSDAQ common stocks. Week of 31 August – 4 September 2026, five trading sessions. Previous issue: Issue #7, week of 28 August. The full series lives at the Foreign Flow Watch hub.

📉 Both boards fell, and the KOSDAQ fell twice as hard

The path matters more than the endpoint again, and in a different way than last week.

🧮 The two biggest stocks were 16.9% of the decline — the reverse of last week

Aggregate market value of the 943 KOSPI common stocks present in our reference set on both dates fell from $4,128.1bn to $4,070.9bn — a loss of $57.3bn, or 1.39%, against the index’s 1.50%. The two figures agreeing to within a tenth of a point is our check that the reference set tracks the index.

The pair lost $9.7bn between them against an index that lost $57.3bn — 16.9% of the decline, on 48.9% of the market value. Issue #7 measured the same thing seven days earlier and got 196%; Issue #4 got 140% in early August. Three measurements, three different answers, and the useful conclusion is not any one of them. It is that the KOSPI’s weekly print carries almost no fixed information about what happened to Korean companies, because roughly half of it is two memory stocks whose relationship to the other half changes week to week. We set out why that concentration exists in Korea’s chip supercycle.

The KOSDAQ, on the same matched-set basis, fell from $346.2bn to $336.6bn — down 2.77%, close to its index’s 2.97%. That makes it the second week in three that Korea’s small-cap board has underperformed the large-cap one; Issue #6 recorded the widest split of the series at 6.32 points, in the same direction.

🌊 Who sold? Everyone with a name

All three of Korea’s named investor categories — foreigners, institutions and retail — were net sellers on the KOSPI this week. The unnamed residual took the entire other side. That is not the usual shape of a down week, in which at least one domestic group absorbs.

Retail was the largest single seller in the country over the five sessions, at −$1.80bn. Almost all of it landed in one session: Korean individuals sold roughly ₩4.05tn of stock on Friday 4 September, the day both boards rallied. Foreign money, by contrast, was a small net buyer of the KOSDAQ (+$30.1M) — on a board that fell 2.97%. The KOSDAQ seller was domestic institutions at −$512.3M. If you were told only that “foreigners bought the KOSDAQ”, you would have guessed the wrong direction for the index. We wrote up why that inference fails so often in how foreign and institutional flows move Korean stocks.

And 69% of the foreign net selling was two names: Samsung Electronics at −$381.5M and SK hynix at −$809.7M, $1,191.3M of a −$1,719.1M national total. The other 2,549 stocks in our set accounted for the remaining $527.8M between them.

🏦 Is the fourth buyer still buying?

Yes — and at a pace so stable it is measurable to within a few percent per session. The residual bought 3,204,278 SK hynix shares and 9,920,487 Samsung shares over the five sessions, $5.71bn between them. Foreign investors sold the same two names to the tune of −$1.19bn across the same window. This is now the tenth consecutive session of residual buying in each name.

SK hynix’s daily band this week was 632,027–649,919 shares, a 2.8% spread. Last week it was 635,526–650,428. Samsung’s was 1,966,601–2,018,968 against 1,840,988–2,026,114 a week earlier. Neither programme wavered. Both remain far inside their filed daily ceilings — SK hynix averaged 26.6% of its 2,407,000-share cap and Samsung 27.0% of its 7,321,653-share cap. We have no explanation for those two numbers being nearly identical, the caps derive from different branches of the same formula, and we are not claiming it means anything.

Both filed acquisition windows are still open, and we re-checked the disclosure record rather than assuming: SK hynix’s window runs 20 August – 19 November 2026 and Samsung’s 24 August – 21 November 2026, and neither company has filed an amendment, a suspension or a termination between 19 August and 5 September 2026.

⏳ When do the two buybacks run out?

At the pace observed so far, SK hynix’s programme is about 37 sessions long and Samsung’s about 30 — against filed windows of roughly 63 sessions each. Both would finish in October rather than mid-November, some six weeks early.

The right unit here is won, not shares, and that is a detail the filings themselves insist on. SK hynix registered ₩40,004,340,000,000 and states that the 24,070,000-share figure was derived from the 18 August close of ₩1,662,000 and “may change with future share-price movement”. Samsung registered ₩14,999,999,992,000 and derives 53,285,968 shares from the 21 August close of ₩281,500, with the same caveat. The cash is the commitment; the share count is an estimate. An English summary that quotes only the share count is quoting the estimate.

Note the gap between the two Samsung rows. By won the programme is 33.9% done; by share count it is 37.0%. The difference is price: Samsung’s stock has fallen from ₩281,500 on the decision date to ₩255,500, so ₩15tn now buys about 58.7 million shares rather than the 53.3 million printed in the filing. A falling share price makes an amount-denominated buyback larger in share terms, not smaller. For SK hynix the same effect is negligible because its price barely moved.

Why the split-date matters for a shareholder: SK hynix’s stated purpose is acquisition for cancellation, and it filed a separate cancellation decision the same day, so those shares are permanently retired. Samsung’s stated purpose is employee share compensation — performance incentives and special performance bonuses — so the share count does not fall; the stock moves from public float to staff. Both are real cash out the door. Only one is a return to shareholders in the sense the Value-Up programme is trying to encourage. We made the same distinction about Celltrion, which cancelled 4% of itself and still ended the year with more shares outstanding.

🧪 A third name switched on — and it fires one of our own tests

Issue #7 published this falsifier: “If the residual appears in a third large-cap name with no corresponding DART filing, then the pattern is a data artefact of our flow source rather than a description of corporate behaviour.” This week it fired.

LG Electronics (066570) went from essentially no residual activity to +386,992 shares — ₩79.3bn, or $58.5M — in five sessions, with no open buyback filing. Over the four weeks of 3–28 August the residual netted just 12,180 shares in the name, worth ₩2.3bn. The switch-on is real, and it is the third-largest residual buy in the market this week.

LG Electronics has no treasury-share acquisition decision in force. It had a ₩100bn treasury-share trust contract with NH Investment & Securities running 2 February – 30 September 2026, and it terminated that contract early on 29 July 2026, ending with 588,589 common shares held (rcpNo 20260729000310, DART). So the literal condition of our test is met.

But the conclusion we attached to it does not follow, and that is our error rather than the data’s. 386,992 shares is 0.238% of LG Electronics’ 162,884,638 shares outstanding. Under Korea’s 5% rule, an existing large holder is required to file an amendment only when its stake moves by one percentage point or more — so a corporate affiliate could buy this much and owe no disclosure at all. The absence of a filing is exactly what you would expect from a legitimate, small, sub-threshold corporate purchase. We wrote a falsifier whose trigger was “no filing” when what it needed to test was “no filing where a filing would have been required.” We restate it at the bottom of this issue.

⚠️ The residual is not always the buyer

Strip out Samsung and SK hynix and the residual was a net seller of $1,068.2M across the rest of the market this week. Nearly all of it landed in a single session.

We do not know what these were, and we are not going to guess. What we did check: the listed share count of all three was unchanged across 1–4 September, so none of this is a cancellation showing up as a trade, and none of the three had filed anything on DART by 5 September that would explain a single-session corporate disposal of that size. We are flagging it because it is the honest counterweight to the section above — the same residual category that looks like a metronome in two stocks looks like a block trade in three others, and readers should hold our identification loosely because of it.

📊 The export-champion scoreboard: 5 sessions against 20

We print both windows every week because one window invents narratives. Where the two columns disagree in sign, that is a reversal, not accumulation.

Three names disagree in sign across the two windows, and in each case the five-session read is the misleading one. LIG Nex1 and HD Hyundai Heavy look sold this week and were bought heavily over the month — this is profit-taking inside an accumulation, not an exit. LG Energy Solution is the reverse: bought this week, sold over the month. And Samsung Electronics is the subtlest case, because both columns are negative: −$381.5M over five sessions against −$323.7M over twenty means foreigners were net buyers of about $58M across the earlier fifteen sessions. All of the month’s Samsung selling, and then some, happened in the last five days.

⚖️ Where flow and price disagreed

We print these every week, because a table of “foreigners bought it and it went up” selected after the fact is not evidence of anything.

The most instructive disagreement is not in the table at all. SK hynix was the largest foreign net sell in Korea this week at −$809.7M and the stock fell 0.36%. Doosan Enerbility is the mirror image: it was the KOSPI’s largest market-value gainer in Issue #7 on foreign net selling of $4.6M, and this week it fell 10.20% on foreign net selling of $113.3M. Same direction of flow, opposite outcomes, one week apart. Anyone building a rule from a single week of Korean flow data is building it on this.

🧾 Scorecard: how Issue #7 held up

Issue #7 published four testable conditions and one self-flagged weak number. One condition fired against us, one held, two remain open, and the weak number got worse.

One thing Issue #7 got right that deserves recording. It flagged that the residual was running at “roughly twice an even pace, which would exhaust both well before their November end dates,” and offered two explanations — front-loading, or contamination. With ten to twelve sessions instead of five, the pace has not slowed at all, which favours the first. Our running record grading ourselves: Issue #4 scored Issue #3 at two failures of three; Issue #5 scored Issue #4 at three holds and one open; Issue #6 scored Issue #5 at two failures of three; Issue #7 scored Issue #6 at three tests that could not run plus one correction against itself. We print the bad weeks at the same size as the good ones.

🧩 Prove this issue wrong

• If either programme is still buying in mid-November at anything like the observed rate, our run-rate projection over-counted the residual and part of it was never the issuer. Testable weekly for ten more weeks, and this replaces Issue #7’s weaker “stops or halves” test.
• Restated third-name test: if the residual switches on in a name at a size that would require a disclosure — above 1% of shares outstanding for an existing large holder, or a new 5% crossing — and no such filing appears within five business days, then our residual is measuring something other than corporate share dealing. LG Electronics at 0.238% does not meet that bar; it would need to be roughly four times larger.
• If SK hynix’s eventual execution report shows materially less than ₩12.86tn bought through 4 September, our attribution is an upper bound we leaned on. The same test applies to Samsung and ₩5.08tn. This is the authoritative check and it is the one we cannot run yet.
• If the 3 September corporate selling in POSCO International, Woori Financial and POSCO DX turns out to be a data artefact rather than real trades, then the same feed produced the buying we identified, and the whole residual method needs re-validating rather than caveating.
• Our weakest number this week is the USD/KRW mapping, as set out in the scorecard: up to roughly 0.8% on any dollar figure. Won figures are exact.

🗂️ One more line from the filing record

Not a flow item, but it is in the same week’s Korean disclosure record and has no English equivalent. On 4 September 2026 SK hynix filed a rumour-and-report clarification responding to a Korea Economic Daily story that its US subsidiary Solidigm is pursuing a ₩5tn pre-IPO. The company’s answer: Solidigm is “reviewing various measures to strengthen competitiveness, but nothing has been decided,” with a commitment to re-disclose by 3 December 2026 (rcpNo 20260904800642, DART). That is not a denial and it is not a confirmation. It is a dated obligation to say more, which is a thing Korea’s disclosure regime produces and most English coverage never surfaces. If you hold SK hynix, 3 December is now on your calendar.

📚 Lingo Check

🎯 Why it matters for K-Export Stars

A single KOSPI print told you Korea fell 1.50% this week. Underneath it, the two largest companies in the country barely moved, everything else fell more than twice as hard, retail sold $1.80bn, foreigners sold $1.72bn, institutions sold $1.13bn — and the buyer of last resort was two corporate treasuries executing programmes they had disclosed, in Korean, in August.

None of that requires a view on memory pricing. It requires reading two filings and doing division. The filings state the won committed, the start date, the end date and the purpose; the settled flow data states what got bought each session; the ratio of the two gives you a completion date that no wire has printed. That date is the kind of thing that changes how you size a position, and it exists only because Korea makes issuers say in advance what most markets only reveal afterwards. For how these flows are constructed and why they mislead as often as they inform, start with how foreign and institutional flows move Korean stocks; for the two stocks at the centre of it in dollar terms, see Samsung and SK hynix stock prices in USD.

Disclosure of interest. The author holds positions in Samsung Electronics (005930) and SK hynix (000660), the two stocks this issue is largely about, and in other Korean equities not discussed here. No position was opened or closed in any stock named above in connection with this article.

Disclaimer: This article is for informational and educational purposes only and is not investment advice, a recommendation, or an offer to buy or sell any security. Figures are as of 4 September 2026 and go stale quickly; prices, flows and programme-completion estimates in particular. Do your own research and consult a licensed adviser before investing.

Next issue: Foreign Flow Watch, week of September 11, 2026 — the KOSPI rose 3.33% and foreigners sold the two stocks that produced four-fifths of the gain.