Foreign Flow Watch: Foreign Selling Fell 90%, the KOSPI Fell Anyway — Week of October 19, 2026

Foreign investors net sold $0.80bn of Korean stocks over the five sessions of 19–23 October 2026. Over 12–16 October they had net sold $8.08bn. The selling fell 90% in a week, and the index fell anyway: KOSPI −2.08%, KOSDAQ −5.42%. The seller that dominated our last four issues almost left the tape. What replaced it was not a buyer. Retail had net bought $3.14bn over 12–16 October. Over 19–23 October it net sold $0.54bn. And the two buyback programmes that were the floor under this market for two months are now over — one of them, we can now show from the filing, ended twelve days before we said it was still running.

This is the fifteenth issue of this series. Every figure below comes from exchange-published daily net quantities for 2,539 KOSPI and KOSDAQ listings, valued at settled closes, with one FX rate across the window, ₩1,323.66/$1 (23 October 2026, FinanceDataReader). The universe excludes ETFs and preferred lines. The window is 19–23 October 2026, five settled sessions, beginning at the session immediately after our previous issue ended. Earlier issues are on the Foreign Flow Watch hub.

What “this week” means — and why there are two comparison windows

Five sessions, 19 to 23 October, with no gap from the last issue. We compare them against the five sessions before (12–16 October) and against the twenty sessions ending 23 October (22 September–23 October). Where the two comparisons disagree in sign, we say so.

A single comparison window invents stories. In July this series called a five-session move “rotation” that a twenty-session view showed to be a bounce inside a month of selling, and we have carried both windows since. The comparison window figures from our last issue are restated here at this issue’s FX rate, so they differ in the second decimal from what that issue printed (its rate was ₩1,340.95). The won strengthened 1.3% across the two weeks.

Foreign selling fell 90%. The index fell anyway.

Foreign net selling over 19–23 October ran at −$0.16bn a session, a tenth of the pace of the week before. Retail, a net buyer in both of the two previous windows, net sold $0.11bn a session over 19–23 October. The KOSPI lost 2.08% and the KOSDAQ 5.42%.

All columns: exchange-published daily net quantities valued at that session’s settled close, summed, at ₩1,323.66/$1. The five-session windows cover 2,539 and 2,541 listings; the twenty-session window covers 2,544, because a listing only enters a window if it has both a flow row and a settled price on every session in it. Per-session figures are the window total divided by its session count.

Two signs flip between the five-session and twenty-session views, and both matter. Retail is a net seller over 19–23 October and a net buyer over the 20 sessions to 23 October. That is a reversal, not a trend. Over the 20 sessions to 23 October retail still net bought $5.58bn. The week’s −$0.54bn is a fraction of that. Foreign investors net bought $0.60bn of Samsung Electronics over 19–23 October. Over the 20 sessions to 23 October they net sold $4.01bn of it. Also a reversal, and one week of it. The foreign total does not flip: net selling in both windows, at a tenth of the pace.

Read the table as an accounting identity, not a causal story. The four categories sum to zero by construction. What the identity does say is where the money that stopped leaving went: foreign selling per session fell 90%, and the categories on the other side shrank with it — institutional buying per session fell 89%, retail swung from buyer to seller, the corporate line fell 59%. Less was sold, and less was bought. And the index still fell. We are not claiming flows explain that; we are pointing out that the week with the heaviest foreign selling of the autumn (12–16 October, KOSPI −14.46%) and the week with the lightest (19–23 October, −2.08%) were both down weeks.

Who sold, session by session

Retail sold on the two big up days and bought on the two big down days; Friday, +0.30%, was the one exception. Foreign flow netted to almost nothing until Friday 23 October, when they sold $0.77bn.

Session values at that day’s settled close and the single window rate; KOSPI is the change on the prior settled close (16 October 5,668.05; 23 October 5,550.30).

Three things in this table are worth more than the weekly totals.

First, retail sold into strength and bought weakness on four of the five sessions. Its two net-selling sessions were the two sessions the index rose most (19 October +1.76%, 21 October +3.29%); its two largest net-buying sessions were the two sessions it fell more than 3.5%. Friday, +0.30% with retail +$0.07bn, fits neither pattern. Institutions were broadly the mirror image (their 21 October line was flat). Whatever else “retail turned seller” means, it was not a capitulation — capitulation sells into falls.

Second, the foreign line netted to almost nothing over four sessions and then sold $0.77bn on Friday 23 October. It was not idle in between: foreign investors net sold $0.44bn on 20 October. They net bought $0.40bn on 21 October. One session is not a return of the seller, and we will not call it one. It is, however, the whole week’s foreign total in a single day.

Third, the corporate residual fell from $0.48bn to $0.04bn between Tuesday and Thursday and never recovered. That is the SK hynix buyback, and it is the subject of the next section.

Did the buybacks end? Samsung’s did — on 6 October, and we missed the filing

Samsung Electronics completed its $11.33bn programme (₩15.0tn, the filing’s figure) on 6 October 2026 and filed the execution report on 12 October. Our last issue, published on 18 October, said the programme was 99.1% spent, estimated it would finish on 19 October, and said no execution report had been filed. The first figure was within 1% of the truth. The second and third were wrong.

Source: Samsung Electronics, 자기주식취득결과보고서, DART rcpNo 20261012000421, filed 12 October 2026. Acquisition period 24 August–6 October 2026, on a trade-date basis. The filing states the shortfall against budget was less than one share’s price. Treasury holding after the programme: 136,727,099 common shares, 2.34% of common stock, as of 12 October.

The filing answers a question our last issue left open. We wrote that the constraint on Samsung’s programme “was not the price — it was how much the companies chose to spend per day,” and that a falling share price should have pulled the finish date forward but did not. Here is the mechanism: Samsung bought a fixed number of shares each session, not a fixed amount of money. Two million shares a day, every day, from 25 August to 2 October (1.8 million on the first day), until the remaining 1,056,305 on 6 October closed the budget to within a share’s price. A fixed-share order spends less won when the price falls, so the budget lasts longer. That is why the finish date moved out as the stock fell. On the 27 full two-million-share days the company’s spending ranged from ₩499bn to ₩567bn a session, purely because the price moved; the share count did not.

Our residual proxy for Samsung read the ending correctly and we misread the proxy. In this window the corporate line in Samsung ran between −₩69bn and +₩65bn a day and netted to $0.00bn across the five sessions. We saw that and called it “slowing to 6% of average,” then projected the last 0.9% of budget onto a 19 October finish. The 0.9% was never going to arrive, because it was measurement error, not unspent money: across 29 sessions the exchange’s corporate category under-counted a ₩15.0tn programme by ₩140bn, or 0.93%. We now know the size of the proxy’s error floor for a large programme, and it is about 1%.

SK hynix: 99.2% spent, and the bid went to zero on 22 October

The exchange-measured spend on SK hynix’s $30.2bn (₩40.0tn) cancellation programme stood at ₩39.68tn, 99.2%, at the 23 October close. Its residual ran at $0.43–0.47bn a session through Wednesday 21 October and then stopped: $0.02bn on Thursday, −$0.01bn on Friday.

Per-session residual and foreign net buying for SK hynix (000660), valued at that day’s settled close; prices in dollars at the window rate. For SK hynix, foreign investors net sold $0.69bn over 19–23 October. Its residual over the same five sessions summed to +$1.37bn.

The budget arithmetic closes. At the 16 October close the programme was 94.7% spent with ₩2.12tn left. Three sessions at ₩565–626bn bought ₩1.80tn of it. The exchange’s cumulative figure stopped at ₩39.68tn, 0.8% short of the ₩40.0tn budget — the same order of gap as Samsung’s 0.93% under-count, and we read it the same way: the programme is spent, and the shortfall is the proxy’s error floor, not unspent money. Our last issue estimated exhaustion on 20 October. The last session of buying was 21 October. One session early, in the same direction as every SK hynix estimate this series has made.

There is one other explanation for a stop on 22 October, and we owe it to readers to print it. SK hynix reports third-quarter results on 28 October, and Korean issuers commonly pause buybacks in the days before results. A pause and a completion look identical in the flow data. What separates them is the execution report: if the programme is complete, a filing like Samsung’s should arrive within a few trading days — Samsung’s came three trading days after its last purchase (6 October to 12 October, with 9 October a holiday). The same lag from 21 October is 26 October. As of 24 October SK hynix has not filed one. That filing is this issue’s first falsifier.

What else lives in the corporate line: a $1.05bn block that showed up as $0.39bn

With the SK hynix programme over, the corporate residual for the rest of the market was negative this week: −$0.29bn once SK hynix is removed. The largest single item was a −$0.39bn print in Hanwha Ocean on 19 October, and the filings show what it was — and how much of it the flow data did not see.

On 19 October Hanwha Systems, an affiliate of Hanwha Ocean’s largest shareholder Hanwha Aerospace, sold its entire 22,169,660-share stake in Hanwha Ocean — 7.24% of the company — in an off-hours block at $47.14 (₩62,400) a share, $1.05bn in total, against a price-return swap signed on 16 October with Kiwoom Securities and six other counterparties. The Hanwha Aerospace group’s holding fell from 60.42% to 54.34% on a shares-and-equivalents basis. Sources: Hanwha Aerospace large-holding report, DART rcpNo 20261019000344; Hanwha Ocean largest-shareholder change notice, DART rcpNo 20261021800343.

Hanwha Ocean’s volume that session was 23.1 million shares against roughly one million on an ordinary day, and the stock fell 14.92% across the week, from $49.11 (₩65,000) at the 16 October settled close to $41.78 (₩55,300) on the 23rd.

Here is the part that matters for this series. A corporate seller of 22.17 million shares should appear in the corporate category as −22.17 million. It appeared as −8.12 million, with institutions +8.09 million on the other side. About 14 million shares — nearly two-thirds of the block — netted out inside a single investor category and are invisible in the four-way split. Which category absorbed them we cannot tell from the exchange data; the swap counterparties are securities firms, and a securities firm’s own account is classified as an institution, so the shares that did show up moving to institutions are consistent with the filing. The rest is not visible to us.

Two lessons for how we read the residual from here. First, the corporate line is a buyback proxy only while a buyback is the only large thing in it; with both programmes finished, that condition no longer holds for this market, and we will not describe the market-wide residual as “the buyback” again. Second, the four-category split can understate a large block by more than half, so when the residual moves in a single name on a single day, the filing is the measurement and the flow figure is the clue.

Was this a semiconductor exit?

SK hynix alone was 85.6% of the week’s foreign net selling: foreign investors net sold $0.69bn of SK hynix over 19–23 October. Samsung Electronics was the week’s largest foreign purchase: net bought $0.60bn over 19–23 October. Measured from the 16 October settled close, SK hynix ended the week up 0.39% at $964.7 (₩1,277,000) and Samsung was unchanged at $162.8 (₩215,500).

Behind SK hynix: Samsung Electro-Mechanics −$0.20bn (the stock fell 5.56%), Daeduck Electronics −$0.10bn (−5.69%), Samsung SDI −$0.06bn (−7.45%), Woori Financial −$0.05bn (−3.99%). On the buying side after Samsung: Samsung E&A +$0.23bn, a stock that rose 37.16% in the week; S-Oil +$0.05bn, which fell 13.86%.

So “foreigners sold Korean chips” is true of exactly one chip stock this week and false of the other. Over the 20 sessions to 23 October it is true of both. Foreign investors net sold $11.89bn of SK hynix over the 20 sessions to 23 October. They net sold $4.01bn of Samsung Electronics over the 20 sessions to 23 October. Together that is 79.6% of the market-wide foreign net selling of −$19.98bn over the 20 sessions to 23 October. The Samsung reversal is one week old. We note it; we do not build on it.

Scorecard

Our last issue listed five things that would show it was wrong. Here is where each stands, and the ones that went against us are in the same table as the ones that did not.

We said no report had been filed. One had been, six days earlier.

Our last issue’s Scorecard said, of the 2% test: “Neither company has filed an execution-status report.” Samsung’s 자기주식취득결과보고서 was filed on 12 October. We published on 18 October. The filing was public for six days before we wrote that sentence, in the one place — the issuer’s own DART page — that this series exists to read.

How it happened is simple and not flattering. Our buyback tracker reads the exchange’s investor-category data and the original acquisition decision; it does not poll DART for the result report. When the tracker said 99.1%, we reasoned that a report follows completion and completion had not happened, so there was no report to look for. The reasoning was circular: we used our estimate to decide whether to check the one document that would have corrected the estimate. From this issue the DART filing list for both issuers is checked before the Scorecard is written, and the check is printed whether or not a filing is found.

Two things the error did not change. The 99.1% was within 1% of the truth, so no flow figure in that issue was wrong. And the headline call this series made on 26 September — that both programmes would be spent in October, not in the November their disclosures allowed for — is now confirmed by filing for Samsung (6 October) and by exchange data for SK hynix (21 October, pending its filing). We were right about the month and wrong about a document, and we would rather say both than only the first.

The retail bid, three issues

Earlier rows are as printed in those issues, at their own window rates (₩1,340.02 and ₩1,340.95); this row is at ₩1,323.66. The differences are in the second decimal and do not change any sign.

The nuclear basket, one line

The falsifier we opened on 13 September closed in our last issue as not fired and, by its own three-part construction, unfireable. We are not reopening it. For the record, foreign net flow in the four names this week was close to zero in each — Doosan Enerbility −$3.4M, KEPCO KPS +$0.7M, KEPCO E&C +$9.6M, BHI +$10.3M — and over twenty sessions Doosan Enerbility remains the basket’s one material foreign sale at −$0.38bn.

What we cannot tell you

One FX rate across a five-session window in which the won itself moved 1.3% against the previous window’s rate. Named again rather than quietly carried.

Whether SK hynix’s buying stopped because the budget is spent or because results are due on 28 October. The flow data cannot separate those; only the execution report can, and it has not been filed. Our reading is “spent”, for the budget-arithmetic reason given above, and it is a reading.

Where two-thirds of the Hanwha Ocean block went. The filing says 22.17 million shares changed hands; the four-category split shows 8.1 million moving from the corporate line to institutions. The remainder netted inside one or more categories we cannot see into. Breaking the institutional line into its components — pension funds, asset managers, proprietary desks — is published separately and is not used in this issue.

Who the other corporate sellers were. With the two programmes finished, the market-wide residual excluding SK hynix was −$0.29bn, and apart from Hanwha Ocean it is spread across names in amounts too small to trace to a filing. “Other corporations” is a category, not a company.

Exchange investor-category data is always as of the previous session. Buyback progress figures are as of the 23 October close. The SK hynix end date of 21 October is our inference from the data, not a disclosure.

Prove this issue wrong

One condition each, and each can be judged on the next window — the five sessions of 26–30 October 2026, settled on the 31st.

If SK hynix files its execution-status report and it shows any purchase after 21 October 2026, our residual called the end of the programme on the wrong day. We expect the filing by 26 October, Samsung’s three-trading-day lag applied; if it has not been filed by 30 October we will print that, and the pause-before-results explanation gains ground.

If SK hynix’s residual exceeds $0.10bn in any single session of 26–30 October, the programme we say ended on 21 October did not end then.

If retail’s net buying per session in 26–30 October is above +$0.50bn, this week’s turn to net selling was a one-week artefact.

If foreign net selling per session in 26–30 October exceeds $1.00bn, this week’s 90% drop was a pause in the selling, not a change in it.

If the 19–23 October quantities move when we re-check them on 27 October, every total in this issue shifts. The snapshot we read this time is saved, so we will be able to say which session changed.

Lingo Check

Residual buyer (기타법인) — the fourth investor category in Korean exchange data: corporations trading shares, including an issuer buying its own. Most feeds publish only the other three, so it is recovered by subtracting them from zero. This issue shows it holding a buyback one week and an affiliate’s block sale the next.

Execution-status report (자기주식취득결과보고서) — the filing a Korean company makes after a direct buyback ends, listing every day’s shares and price. Samsung’s, filed 12 October, is the document that corrected this series.

Off-hours block trade (시간외매매 / 시간외대량매매) — a large pre-arranged trade crossed outside the regular session at a negotiated price. It appears in the day’s volume and in the investor-category totals, but only to the extent the buyer and seller sit in different categories.

Price-return swap (PRS, 주가수익스왑) — a contract in which a holder sells shares to a counterparty, usually a securities firm, and keeps the economic exposure to the price. Hanwha Systems’ 22.17 million Hanwha Ocean shares were sold against one. The shares move; the risk, in large part, does not.

Settled close (확정 종가) — the exchange’s confirmed closing price. Our flow table carries a provisional price for the newest session, so every figure here is valued at the settled close from the price table instead.

Disclosure

The author holds shares in Samsung Electronics (005930) and SK hynix (000660). This article reports exchange data and filings about the market as a whole; it is not a recommendation to buy or sell any stock. Where a number is our estimate rather than a disclosed figure, the sentence containing it says so.

Flows: exchange-published daily net quantities by investor category for 2,539 KOSPI and KOSDAQ listings (ETFs and preferred lines excluded), valued at each session’s settled close and summed across 19–23 October 2026, at a single rate of ₩1,323.66/$1 (23 October 2026, FinanceDataReader). Comparison windows (12–16 October, 2,541 listings; 22 September–23 October, 2,544 listings) at the same rate. Buyback progress: exchange investor-category data against disclosed authorisation amounts, as of the 23 October close. Filings: DART, linked inline. Index levels: exchange-published daily closes.