Market Insights

A session-by-session breakdown of foreign, institutional and retail net flows in Samsung Electronics (005930) and SK Hynix (000660), based on KRX investor-type data through the 2026-08-05 settled close. Key finding: the claim 'foreigners are fleeing Korean memory stocks' and its equally popular opposite, 'foreigners are accumulating the dip,' rest on the same dataset and differ only in the start date. Shift the window by one session and Samsung's foreign net flips from +7,611,626 shares to −747,385 shares. The cause: 07-31 alone accounts for 110% of the four-session net (206% for Hynix). The accurate answer is a third one — foreigners traded with the price direction (same-sign in 7 of 8 sessions), while domestic institutions were the persistent seller (−11.04M shares over three sessions, 14.8x the foreign sell) and retail investors absorbed the supply. All figures pass a zero-sum check across investor classes.

Over the three sessions from the 2026-07-31 close, companies doing the same thing — making memory chips — diverged by 23.36 percentage points: SanDisk +14.79%, Micron +7.59%, Western Digital +0.12%, Seagate −0.37%, SK Hynix −8.21%, Samsung Electronics −8.57%. This piece separates two effects. First, 8.57pp is pure time-zone artifact: Seoul's 8/04 close prints at 01:30 ET while the U.S. readings are 10:30 ET, with an entire U.S. session in between; net of the clock, the true spread is 14.79pp. Second, that remaining 14.79pp is structural — the same ordering repeated in two opposite directions (the 8/03 industry-wide selloff and the 8/04 industry-wide rebound) and lines up near-monotonically with HBM exposure, meaning the market is repricing HBM content per system rather than memory demand. Also covered: how Seoul's V-shaped session with ~8pp of range closing flat falsified our own morning call within the day; options positioning running exactly opposite to the price ranking (strongest name SanDisk buying protection at P/C 1.74 while laggard Micron sits at P/C 0.294); and why the semiconductor excess return is sector rotation rather than new liquidity. All baseline closes independently re-verified via yfinance K-lines and pykrx, with full replication steps and methodological limits.

On 2026-08-03 Korea's two memory majors dropped hard: SK Hynix closed at KRW 1,567,000 (-8.79%) and Samsung Electronics at KRW 239,500 (-8.76%), both underperforming the KOSPI (-5.35%). Foreign net flow as a share of each stock's traded volume flipped from +21.3%/+14.4% on 07/31 to -20.4%/-14.2%, which reads like a textbook foreign reversal. This piece runs three fully recomputable corrections. First, the apparent symmetry is an artifact of collapsing volume: Hynix turnover fell from 10.35m to 5.43m shares (-47.5%) and Samsung from 57.88m to 27.39m (-52.7%), so in share terms foreigners sold back only 50.2% and 46.6%, leaving both names net-bought across the two sessions (Hynix approx +1.10m shares, Samsung +4.463m). Second, the real liquidation came from domestic institutions: on Samsung they sold back 139.3% of Friday's buying, a two-day net of -1.421m shares, pointing the opposite way from foreigners. Third, foreigners sold the semiconductor leg, not Korea, buying Hyundai Motor (+11.0%), Kia (+11.6%) and Samsung SDI (+5.9%) the same session. A two-step percentage-versus-shares checklist and full method limits are included.

On 2026-07-31 three Korean memory-complex names closed near the +30% daily limit: SK Hynix at KRW 1,718,000 (+29.95%), Hanmi Semiconductor at KRW 214,000 (+27.98%), and Samsung Electronics at KRW 262,500 (+26.81%). But non-memory Korean large caps rose a median +9.53% the same session. This piece runs two fully recomputable corrections. First, strip market beta: roughly one third of the headline (31.8%-35.5%) was broad-market, leaving a true memory-specific excess of +16.84% multiplicative, not +30%. Second, fix the calendar: an 'extreme U.S.-Korea divergence of -11.52pp at the 99th percentile' was produced by subtracting Korea's 7/31 close from the U.S. 7/30 close, one trading day apart. Realigned to a two-day window, the spread collapses to 4.67pp and the dislocation vanishes. Includes a reusable two-step checklist and full disclosure of the method's limits.

On 2026-07-30 Samsung Electronics' board did two things on the same day: it revised Q2 2026 results upward to KRW 171.50tn revenue (+130.00% YoY) and KRW 89.49tn operating profit (+1,813.83% YoY), and it declared only a routine quarterly dividend of KRW 374 per share with zero buyback. This piece re-verifies every figure against first-party DART filings from Korea's Financial Supervisory Service and introduces a reusable analytical method — the 'clean sample test'. SK Hynix's refusal to detail a buyback has been widely excused as an ADR-related regulatory gag; Samsung faces no such constraint, upgraded its earnings that very day, and still returned essentially nothing. That makes the regulatory explanation non-load-bearing. It also unpacks a commonly miscalculated trap: Kioxia's KRW 60tn is not KRW 60tn of buyback ammunition.

On 2026-07-29 SK Hynix reported Q2 2026 results: operating profit KRW 60.54tn, +557% YoY, a 76% operating margin — all records. The stock closed down 9.61% the same day, after reversing from +4.45% to -9.61% intraday (24.1% range). This piece explains the paradox with four reproducible calculations: the miss was entirely in revenue not margin; net profit hides a KRW 62.17tn one-off gain (P/E moves from 3.7x to 6.6x); the buyback precondition can now be back-solved from the cash statement; and why a 2x leveraged ETF still cannot break even even after the underlying fully recovers. Includes full definitions, workings, and falsification conditions.

On 2026-07-28 global memory-chip stocks crashed together, with KOSPI down 10.84% in a single session, its 5th worst in 30 years. Yet the Dow closed higher, VIX did not spike, and the Korean won was flat. This article lays out a five-check framework for separating a genuine systemic crisis from sector-level deleveraging, with the full loss-gradient dataset and the explicit conditions that would invalidate the conclusion.

The last week of July 2026 brings a triple catalyst convergence: Goldman raises FOMC rate hike probability to 35%, SK Hynix Q2 earnings arrive July 29, and Big Tech reports en masse. Options markets are flashing extreme signals — LRCX IV Rank 99.5, AMAT 87.2, AMD 80.6. We analyze the fund flow divergence (TSM +$66.8M vs MU -$152M), institutional de-leveraging in CCASS, and three FOMC scenarios for the memory sector.

Intel Q2 2026 revenue hit $16.1B (+25% YoY), with Data Center/AI at $6.3B beating consensus by 14% and Foundry's 18A entering risk production. We analyze three signals: semiconductor recovery broadening from memory to logic, Intel Foundry's catch-up trajectory, and the hidden margin threat from oil prices breaking $100.

A use-case-driven roundup of quant tools for individual investors, organized into 4 categories: broker terminals for A-share live trading, backtesting platforms, overseas-market tools, and open-source frameworks. With pros, cons, target users, and the key compliance line: only official broker terminals are legal for automated A-share trading.

Monte Carlo simulation reveals: when underlying annualized volatility exceeds 70%, a 2x leveraged ETF loses money even if the stock rises 40% per year. We quantify the exact vol threshold where leverage turns from amplifier to destroyer.

SK海力士7月22日发布2026年Q2财报。尽管DRAM现货价格连涨8周，存储股却从6月高点回落30%。我们拆解三大关键信号：HBM利润率走势、回购计划扩展、Q3指引——叠加韩国杠杆ETF资产从24万亿韩元缩水至17万亿的结构性去杠杆背景。

SOXX两日跌超5%，台积电/ASML财报兑现抛压叠加7月OpEx。但期权市场罕见逆向信号密集爆发——AMD PCR达4.0σ极端值、KLAC PCR触5.25历史100%胜率。解读聪明钱与恐慌的博弈。

用29只AI产业链标的的期权Put Skew数据，拆解存储芯片暴跌背后的真实恐慌分布。数据显示设备/代工恐慌反降，存储被单独惩罚。

台积电将于7月16日公布2026年Q2财报。存储股从6月高点回调30%，SK海力士ADR刚登陆纳斯达克之际，我们解读5个趋同信号，判断存储半导体周期是否正在形成拐点。