The Short Answer
On 2026-08-24, Korea's two memory giants fell on the same session but 2.55x apart: Samsung Electronics −8.70%, SK Hynix −3.41%. The reflexive explanation is that foreign investors were fleeing.
KRX investor-type data for that settled session does not support it. Foreign net selling as a share of same-day volume was nearly identical — Samsung 21.21%, SK Hynix 22.87%. If anything it was marginally heavier on Hynix (0.93x). The same foreign cohort treated the two names almost identically, which cannot account for a 5.29 percentage-point gap in performance.
The variable that actually diverges is domestic institutions: 19.32% of Samsung's volume versus 4.34% of Hynix's — a 4.45x difference.
Data as-of: Korean equities are 2026-08-24 close, settled and final. US figures are intraday, taken 10:34 ET — roughly 64 minutes into the session — and are NOT settled. They can and will change by the close. Read every US number below as provisional.
Was It Foreign Selling?
| Metric (2026-08-24 close, settled) | Samsung 005930 | SK Hynix 000660 | Ratio |
|---|---|---|---|
| Close | ₩257,000 | ₩1,671,000 | — |
| Daily change | −8.70% | −3.41% | 2.55x |
| Volume (shares) | 32,451,480 | 3,993,766 | — |
| Foreign net sell (shares) | −6,882,379 | −913,265 | — |
| Foreign net sell ÷ volume | 21.21% | 22.87% | 0.93x |
| Domestic institution net sell (shares) | −6,269,240 | −173,328 | — |
| Institution net sell ÷ volume | 19.32% | 4.34% | 4.45x |
Three extractable facts:
- Foreign selling intensity was effectively the same across both names, and if anything tilted toward Hynix. "Foreigners fled" cannot explain Samsung's extra 5.29pp of decline.
- Domestic institutional intensity differs by 4.45x — the only variable that moves in the same direction and the same order of magnitude as the performance gap.
- This is a decomposition within a single investor-type table, so the categories are internally consistent. Before attributing the selloff to foreigners, look at the institutional row.
Why Samsung Specifically?
The background sits in two DART filings from three days earlier. The two companies were buying their own stock under fundamentally different legal classifications:
- SK Hynix, 08-19, 주식소각결정 (share cancellation decision), rcp 20260819800340: ₩40.00 trillion, 24,070,000 shares, permanently destroyed after acquisition. The filing states outright that shares outstanding decrease while paid-in capital does not.
- Samsung Electronics, 08-21, 자기주식취득결정 (treasury share acquisition decision), rcp 20260821000616: ₩15.00 trillion, 53,285,968 shares. Item 5, 취득목적 (purpose of acquisition), reads 임직원 주식보상 — employee and executive stock compensation. The acquisition window runs 2026-08-24 to 11-21, and the repurchased shares are ultimately distributed to employees, returning to the float.
In the prior session (08-21) the market bought both as the same event (Samsung +3.87%, Hynix +2.31%). On 08-24, only Samsung gave the entire move back and then some.
⚠️ This causal link is an inference, not an established fact. Within the DART window (08-17 to 08-24) Samsung filed nothing new beyond those two 08-21 documents, and I could not locate a specific same-day trigger filing for 08-24. Korean disclosures are typically published after the close, so it may simply not be visible yet. Treat this as a falsifiable hypothesis, not a conclusion.
An Under-Computed Mechanism: Hynix's Own Decline Made Its Cancellation Larger
Item 11 of the Hynix cancellation filing states explicitly that the 24,070,000 share figure was derived by dividing the fixed ₩40.00 trillion amount by the closing price on the day before the board resolution (2026-08-18: ₩1,662,000). The actual number of shares cancelled will vary with the price during the acquisition window, and the final count will be published via an amended disclosure.
The won amount is fixed; the share count is not. A lower price means more shares retired.
| Pricing basis | Price | Shares retirable | % of 730,492,365 outstanding |
|---|---|---|---|
| 08-18 close (filing basis) | ₩1,662,000 | 24,070,000 | 3.295% |
| 08-21 close | ₩1,730,000 | 23,123,896 | 3.166% |
| 08-24 close | ₩1,671,000 | 23,940,359 | 3.277% |
Hynix's own 3.41% decline on 08-24 pushed the effective cancellation ratio from 3.166% back up to 3.277% (+0.111pp). The acquisition and amendment window runs through 2026-11-19.
Key point: any coverage citing "SK Hynix will retire 3.295% of shares" as a fixed fact is quoting a price snapshot from 2026-08-18, not a final number.
What Were US Memory Names Doing? (Intraday, Not Settled)
Figures below were pulled live at 10:34 ET, roughly 64 minutes into the session — these are not closing prices:
| Tier | Ticker | Intraday change |
|---|---|---|
| Mega-cap platform | GOOGL | +1.09% |
| MSFT | +0.65% | |
| Broad market | SPY | −0.27% |
| QQQ | −1.11% | |
| Semis (ex-storage) | AVGO | −1.48% |
| NVDA | −2.09% | |
| TSM | −2.56% | |
| SMH (sector ETF) | −2.80% | |
| AMD | −3.16% | |
| Storage media | WDC (HDD) | −6.73% |
| MU (DRAM) | −6.73% | |
| STX (HDD) | −6.94% | |
| SNDK (NAND) | −9.21% |
Two extractable facts:
1. This is still concentrated, not broad. At the same timestamp GOOGL and MSFT were positive and SPY was down only 0.27%. Cross-sectional dispersion from GOOGL to SNDK reached 10.30 percentage points. That is not the shape of broad de-risking — the risk is contained within the storage bucket.
2. But the market is not differentiating by technology. DRAM (MU), NAND (SNDK), and HDD (WDC, STX) are three product lines with entirely different supply-demand structures, yet all four names sat in a narrow −6.73% to −9.21% band, less than 2.5pp apart. If this were a NAND oversupply story, HDD names should not move in lockstep. If it were a DRAM cycle story, SNDK should not be leading the decline. The pricing treats "storage media" as a single undifferentiated basket.
The contrast is the most interesting cross-market observation of the day: Korea separated the two names sharply (2.55x), while US trading at this timestamp barely separated them at all.
Are These Declines Statistically Unusual?
Normalizing each name by its own realized volatility (HV20 taken from the 2026-08-21 settled options snapshot; daily vol = annualized ÷ √252):
| Ticker | Intraday move | HV20 daily-equivalent | In sigma |
|---|---|---|---|
| MU | −6.73% | 6.12% | 1.10σ |
| SNDK | −9.21% | 9.34% | 0.99σ |
| SMH | −2.80% | 2.84% | 0.99σ |
Key point: measured against each name's own trailing 20-day realized volatility, all three readings sit near 1σ. The absolute numbers look dramatic, but statistically this is an ordinary day for these securities. What is genuinely abnormal is not the magnitude of the moves — it is the cross-sectional concentration.
⚠️ Caveat: HV20 is a 08-21 snapshot while today's move is intraday and unsettled. If the decline widens into the close, the sigma values rise accordingly. Separately, both MU and SNDK sat at the 0.0 percentile of implied volatility in the 08-21 session (self-collected sample, n=104 trading days from 2026-03-20 — not 252), meaning almost no one had paid for protection against a move of this size beforehand.
Verifiable Forward Checkpoints
| Date | Event | Primary source |
|---|---|---|
| 2026-08-25 Korea close | Whether Samsung foreign net selling converges, and whether domestic institutions still account for ~20% of volume | KRX investor-type data |
| 2026-09-18 | SK Hynix re-disclosure deadline on its "undetermined" answer to the Japan fab inquiry | DART rcp 20260821800524 |
| 2026-11-19 | End of Hynix cancellation acquisition window / amendment disclosure | DART rcp 20260819800340 |
| 2026-11-21 | End of Samsung treasury acquisition window | DART rcp 20260821000616 |
Falsification condition: if on 2026-08-25 Hynix's decline catches up to Samsung's and foreign net selling widens beyond 3 million shares, the conclusion here — domestic-institution-driven and concentrated in a single company — is void, and the move should instead be read as a sector-wide repricing.
Method and Sources
- Korean prices, volume, investor-type net flows: KRX same-day investor-type trading data, 2026-08-24 close, settled.
- Korean disclosures: primary DART filing text; rcp numbers given inline and independently verifiable at dart.fss.or.kr. Samsung's 취득목적 = 임직원 주식보상 and Hynix's Item 11 amount-to-share conversion note were both read from the filing body, not from secondary reporting.
- US intraday quotes: Tiger market data API, pulled live 2026-08-24 10:34 ET, not settled values.
- Implied and realized volatility: self-collected options snapshots; latest settled session 2026-08-21. IV percentile sample n=104 (from 2026-03-20), not 252 trading days.
- For the full breakdown of the two buybacks' legal classifications, see our earlier piece: "SK Hynix's KRW 40T vs Samsung's KRW 15T: Why Only One of These Korean Memory Buybacks Actually Retires Shares."
This is public-layer market structure analysis. It contains no position, sizing, or trade execution information, and is not investment advice.