美股

Why Did 'Foreign Investors Barely Moved' Mask a 340 Billion Won Rotation in Korean Memory Stocks?

2026-09-01 · 10 min · AlphaGBM
korea-exchangeforeign-investor-flowsSK-hynixsamsung-electronicsmemory-chipssector-rotationnet-vs-gross-flowsKRXbroadcomimplied-volatilitysemiconductor-cycleSeptember-2026

A Near-Zero Net Number Hiding a Rotation 11.7x Its Size

At the September 1, 2026 close on the Korea Exchange, foreign investors' combined net position change across the two memory leaders was just 29.19 billion won (roughly $21.3 million). Read on net alone, foreign investors "barely moved" that day.

Split into its two legs, the conclusion reverses entirely:

Direction Ticker Foreign Net Shares Value at Session Close
Sell leg Samsung Electronics (005930.KS) −707,536 shares ₩184.67B (approx. $0.13B)
Buy leg SK Hynix (000660.KS) +91,835 shares ₩155.48B (approx. $0.11B)
Net ₩29.19B (approx. $21.3M)
Gross two-way turnover ₩340.14B (approx. $0.25B)

Gross turnover was 11.65 times the net figure. The reading "foreign investors did nothing today" concealed an intra-sector reallocation more than eleven times its own size.

Why "Net Inflow/Outflow" Is the Most Misleading Metric in This Regime

Net flow carries information only when capital moves in the same direction. When capital switches between names inside one sector, the two legs cancel and net approaches zero — precisely when repositioning is most intense.

This is not a theoretical concern. On September 1, 2026, reading net alone yields "foreign investors have no view on Korean memory." Reading the legs yields "foreign investors are actively cutting Samsung and adding SK Hynix." The two readings imply opposite position actions.

Decision rule: when net approaches zero, first establish whether both legs are small or both legs are large and offsetting. Only the former means nothing happened.

The Slow Variable Confirms: Foreign Ownership Ratio

Flow data can be distorted by single-session noise, but the foreign ownership ratio is a slow-moving variable whose direction is harder to fake. Both agreed on September 1:

Ticker Foreign Ownership 08-31 Foreign Ownership 09-01 Direction
SK Hynix 50.55% 50.57% Up
Samsung Electronics 46.72% 46.71% Down

Ownership shifts align directionally with same-session net share flows, and the zero-sum check passes — this is not data noise.

Worth isolating: SK Hynix foreign ownership had declined for five consecutive sessions (from 50.75% on 08-25 down to 50.55% on 08-31). September 1 marks the first uptick of this stretch.

Six-Session Net Share Series

Session Hynix Foreign Net Samsung Foreign Net Hynix Foreign Ownership
08-25 −1,234,309 −5,003,317 50.75%
08-26 −227,762 +2,518,326 50.69%
08-27 −159,385 +1,381,786 50.63%
08-28 −190,235 −1,927,155 50.60%
08-31 −211,935 −2,293 50.55%
09-01 +91,835 −707,536 50.57%

Hynix foreign flow turned net buy for the first time after five consecutive selling sessions, and diverged from Samsung for the first time in the stretch. This is the structural inflection point in the series.

Methodology: Why This Table Uses Net Shares, Not "Percent of Volume"

Many Korea flow analyses use "net flow as a percentage of session volume." That ratio cannot be compared across sessions.

The reason: the Korea Exchange retroactively revises volume after the close, while net share counts are not revised. The same session's ratio therefore yields different values computed today versus three days later. Recomputing the 08-27 and 08-28 sessions from source for this article produced 9.49% and 35.26%, against 9.18% and 27.92% computed live on those days — the entire discrepancy comes from a revised denominator.

Conclusion: use net shares for cross-session comparison; use ratios only as single-session cross-sectional readings. As a single-session figure, SK Hynix's combined (foreign + domestic institutional) net selling on September 1 was 2.05% of session volume.

A Hypothesis That Got Falsified

On August 31 data, a natural hypothesis was "Samsung domestic institutions are engaged in sustained one-way distribution." September 1 data falsified it:

One-way distribution did not persist. The August 31 session should be classified as a single-session event, not the start of a trend.

US Equities on 08-31 Provide Independent Evidence in the Same Direction

The same "stop trading the sector as one block" phenomenon appeared independently in the most recent settled US session (2026-08-31):

Ticker Close Change
SanDisk (SNDK) 1,566.70 +5.50%
Micron (MU) 958.73 +2.77%
NVIDIA (NVDA) 220.78 +1.48%
AMD 470.72 +1.10%
Semiconductor ETF (SMH) 556.63 +0.64%
Broadcom (AVGO) 370.34 +0.42%
Nasdaq ETF (QQQ) 716.76 +0.05%
S&P 500 ETF (SPY) 767.05 −0.30%
TSMC (TSM) 415.32 −0.53%

The meaningful structure is not the direction of any single name but this: on the same session the semiconductor ETF closed higher, TSMC closed lower. Foundry and design diverged under one sector label, while broad indices (SPY −0.30%, QQQ +0.05%) gave no direction at all.

In short: US equities showed selection between segments within an industry; Korea showed selection between names within one segment. Same direction.

An Unexplained Anomaly: Broadcom Options

Across the nine-name US cross-section on 08-31, Broadcom's options pricing is the sole outlier:

Ticker IV Percentile ATM IV 20-Day Realized Vol Session Change
AVGO 95.3 79.95 44.04 +0.42%
SNDK 0.9 77.50 97.39 +5.50%
NVDA 1.8 33.64 44.38 +1.48%
MU 0.0 55.01 55.03 +2.77%
SPY 3.6 11.02 10.60 −0.30%

Broadcom's at-the-money implied volatility sits 35.91 points above its own 20-day realized volatility, while the stock moved only +0.42% on the session. The other eight names all sit at IV percentiles of 0.0–3.6.

Price barely moving while options are bid to the top of the sample range typically indicates a name-specific, short-dated event being priced.

Three constraints must be stated alongside this reading, or it will be over-interpreted:

  1. Broadcom's IV percentile is drawn from a self-collected 108-trading-day sample (samples across these nine names range from 82 to 111 days), not a standard 252-day window. "Top of sample" does not mean "highest of the year."
  2. No day-over-day comparison is valid. All nine names rolled contracts that session (days-to-expiry jumped from 3/5 to 6/7/9), and Broadcom's prior session had only 17 contracts of open interest with null implied vol. The claim "Broadcom IV went from nothing to 95.3" does not hold; only the same-day cross-section is usable.
  3. No primary source for a catalyst had been identified as of writing. By discipline, an unlocatable catalyst warrants more caution, not less.

What This Means for Analysis

  1. Sector-level conclusions are losing resolution. Questions framed as "is memory up?" or "is semis strong?" no longer carry enough information to guide positioning in the current structure. Useful judgments must descend to the individual name.
  2. Net-flow metrics require paired leg decomposition. When net approaches zero, establish whether both legs are small or offsetting.
  3. Index level currently gives no direction. The three broad benchmarks closed 08-31 between −0.30% and +0.05% — neither risk-on nor defensive. Against that backdrop, large single-name moves more likely reflect idiosyncratic causes than systematic signal.

Falsifiable Verification Conditions

Data Scope and Timing Disclosure


This article is based on publicly available market data and is for informational purposes only. It does not constitute investment advice. Data sources: Korea Exchange (KRX) investor-type trading data; US equity and options snapshots.

想要更深入的期权数据与实盘分析?了解 AlphaGBM →