美股

Are Foreign Investors Fleeing Korean Memory Stocks? Same KRX Data, Start Date One Session Apart — the Answer Flips from 7.61M Shares Bought to 747K Sold

2026-08-05 · 13 min · AlphaGBM
外资foreign investors韩国股市Korea stock marketKRX三星电子Samsung Electronics005930SK海力士SK Hynix000660投资者别investor type flows资金流capital flows零和校验zero-sum check窗口错配window mismatch存储芯片memory chipsDRAMHBMNAND半导体semiconductors美光MicronMU闪迪SanDiskSNDK机构投资者institutional investors散户retail investors市场结构market structure顺势交易trend following数据方法论data methodology

Data basis (important): Every conclusion in this article rests on settled closing prices. Korean figures are KRX closes for 2026-08-05 (Seoul closed at 15:30 KST). Investor-type flow data comes from Naver Finance's primary page and was re-pulled and re-verified at 22:30 CST on 2026-08-05. The US section (Section 6) uses live intraday prices as of 22:33 CST on 2026-08-05 (10:33 ET, roughly one hour into the session) — those numbers will keep moving until the close, and no conclusion in this article is built on them.

The short answer

Neither narrative holds. "Foreign investors are fleeing Korean memory stocks" is not supported by the data — but its equally widespread opposite, "foreigners are accumulating all the way down," is not supported either.

Both claims draw on the same KRX dataset. The only difference is which day you start counting. Move the start date by one session and Samsung Electronics' foreign net flips from +7,611,626 shares (buying) to −747,385 shares (selling). That is a sign reversal, not a change in magnitude.

The accurate answer is a third one: foreigners traded with the direction of price — they were neither building a position nor exiting one.

1. Raw data (six sessions, fully reproducible)

Units: shares. "Retail + other corporates" is a residual (= −(foreign + institutional)), not an independent observation.

Samsung Electronics 005930

Date Close (KRW) Change Foreign net Institutional net Retail+other (residual)
07-29 208,500 −5.23% −3,550,689 +5,971,830 −2,421,141
07-30 207,000 −0.72% +656,680 −237,928 −418,752
07-31 262,500 +26.81% +8,359,011 +3,618,959 −11,977,970
08-03 239,500 −8.76% −3,896,489 −5,039,954 +8,936,443
08-04 240,000 +0.21% +850,527 −3,924,513 +3,073,986
08-05 246,000 +2.50% +2,298,577 −2,078,706 −219,871

SK Hynix 000660

Date Close (KRW) Change Foreign net Institutional net Retail+other (residual)
07-29 1,401,000 −9.61% −921,877 +791,370 +130,507
07-30 1,322,000 −5.64% +444,498 −80,151 −364,347
07-31 1,718,000 +29.95% +2,205,767 +277,183 −2,482,950
08-03 1,567,000 −8.79% −1,107,357 −288,012 +1,395,369
08-04 1,577,000 +0.64% −418,552 +18,945 +399,607
08-05 1,668,000 +5.77% +389,804 −28,260 −361,544

Share of volume on 08-05: Samsung foreign net buying was 10.2% of volume, institutions −9.3% (volume 22,444,447 shares). SK Hynix foreign 10.4%, institutions −0.8% (volume 3,740,705 shares). Foreign ownership stands at 46.67% for Samsung and 51.00% for SK Hynix.

2. One question, two windows, two opposite answers

Samsung Electronics 005930

Basis Flow window Foreign Institutional Retail+other Matching price range
A 07-31 to 08-05 (4 sessions) +7,611,626 −7,424,214 −187,412 207,000→246,000 +18.84%
B 08-03 to 08-05 (3 sessions) −747,385 −11,043,173 +11,790,558 262,500→246,000 −6.29%

SK Hynix 000660

Basis Flow window Foreign Institutional Retail+other Matching price range
A 07-31 to 08-05 (4 sessions) +1,069,662 −20,144 −1,049,518 1,322,000→1,668,000 +26.17%
B 08-03 to 08-05 (3 sessions) −1,136,105 −297,327 +1,433,432 1,718,000→1,668,000 −2.91%

Both bases independently pass the zero-sum check (the three investor classes net to 0). Both are correct — they simply answer different questions.

The decisive point is that the flow window must be aligned with the price window. Basis A's four sessions of flow include 07-31 — and on 07-31 the stock rose from 207,000 to 262,500. Using 262,500 (the 07-31 close) as the price base while counting 07-31's full-day buying as flow means explaining buying that occurred during a rally with the price that existed after the rally. That mismatch is exactly how the phrase "foreigners bought all the way down" gets manufactured.

Only two pairings are valid:
- Flow includes 07-31 → the price base must be the 07-30 close (Basis A, over which the stock actually rose 18.84%)
- Price base is the 07-31 close → flow must start on 08-03 (Basis B, over which the stock fell 6.29%)

3. Why the sign flips: one day outweighs the entire net

Foreign net on 07-31 alone Four-session net Single day / net
Samsung Electronics +8,359,011 +7,611,626 110%
SK Hynix +2,205,767 +1,069,662 206%

A single session's buying exceeds the entire four-session net. Put differently, the "foreign net buying" in Basis A is entirely attributable to 07-31; the other three sessions were collectively net selling. Any "trend" claim built on Basis A is in substance a claim about one day — and that day was a +26.81% / +29.95% melt-up.

When more than 100% of a cumulative net comes from a single session, it is not a trend. It is an event.

4. The accurate answer: foreigners followed price

Comparing the sign of each session's foreign net flow with the sign of that day's price move:

Date Samsung move Foreign Same sign? Hynix move Foreign Same sign?
07-31 +26.81% Buy +29.95% Buy
08-03 −8.76% Sell −8.79% Sell
08-04 +0.21% Buy +0.64% Sell
08-05 +2.50% Buy +5.77% Buy

Same-sign in 7 of 8 observations across the four-session window; 9 of 12 across six sessions.

Foreigners bought strength and sold weakness. That is the signature of trend-following, and it fits neither narrative: an exit would keep selling into rebounds, and dip-accumulation would buy into crashes. On 08-03, when Samsung fell 8.76%, foreigners sold 3.90M shares; on 08-05, when it rose 2.50%, they bought back 2.30M. The direction tracked the tape.

Over the same span, domestic institutions were the persistent seller. Samsung institutions sold for three consecutive sessions, totalling −11,043,173 shares14.8x the foreign selling in the same window (−747,385 shares). Retail investors absorbed the supply, net buying +11,790,558 shares over those three sessions.

So to the question "who is selling Korean memory stocks," the answer since 08-03 is: domestic institutions are selling, retail is absorbing, and foreigners are simply trading back and forth with price.

SK Hynix has a different structure. Its institutions sold only 297,327 shares over the three sessions (about 26% of the foreign selling), meaning institutions barely moved and the selling pressure sits in the residual. For Hynix we can therefore only say the seller was not institutions — we cannot identify it further.

5. The zero-sum check: why these figures can be trusted

KRX investor-type data has a built-in self-audit: within a single session, net share flows across all investor classes must sum to zero — every buyer requires a seller.

For 08-05:
- Samsung Electronics: +2,298,577 (foreign) − 2,078,706 (institutional) − 219,871 (retail+other) = 0
- SK Hynix: +389,804 (foreign) − 28,260 (institutional) − 361,544 (retail+other) = 0

Both window aggregates in Section 2 likewise net to zero.

Any claim of the form "foreigners pulled X billion out of Korea" that cannot name the counterparty, and whose investor classes do not sum to zero, has not been verified.

One limitation stated honestly: "retail + other corporates" here is a residual (= −(foreign + institutional)), not an independent observation. The zero-sum check therefore cross-validates the foreign and institutional series, but cannot separate retail from other corporate entities.

6. US session tonight: still moving, not a conclusion

⚠️ The figures below are live intraday prices as of 22:33 CST on 2026-08-05 (10:33 ET, about one hour into the session) and will continue to move before the close. They are provided for reference only; neither the title nor any conclusion in this article rests on this section.

Ticker Last (USD) Intraday
Micron (MU) 908.81 +1.81%
SanDisk (SNDK) 1391.35 −2.54%
Western Digital (WDC) 552.69 +0.75%
SMH (semiconductor ETF) 573.49 −0.39%

An empirical footnote: within a 90-second span while this article was being drafted, MU moved from +2.16% to +1.81% and SNDK from −2.31% to −2.54%. That is precisely why the conclusions here are anchored to settled Seoul closes rather than live New York prices — Seoul has closed; New York has not.

7. What would invalidate this conclusion

Methodology notes

想要更深入的期权数据与实盘分析?了解 AlphaGBM →