The short answer
On August 10, the memory sector visibly stratified. The two names that reported earnings after the close on August 5 — SanDisk (+2.31%) and Western Digital (+1.61%) — are up, while the two outside that window — Micron (-0.81%) and Seagate (-0.38%) — are not participating (Micron does not report until September 23; Seagate already reported on July 28). Semiconductors as a group are lower at the same moment (SMH -0.78%, SOXX -0.83%).
That is the exact mirror image of last week. Last week the hardest-hit names were SanDisk and Western Digital, and the one that held up was Micron.
Before going further, accept one constraint: none of today's moves is individually large enough to require an explanation. Measured against each stock's own realized volatility over the past 20 sessions, the largest move in the group is 0.61 of one daily standard deviation. The information here is not in the magnitude. It is in the fact that the sign pattern reversed.
All US figures below are live intraday readings as of August 10, 2026, 10:39 AM ET, roughly 69 minutes into the session. The US market is still trading. These are not closing prices.
What the tape shows
| Ticker | Open gap | Now | vs. open | Day range |
|---|---|---|---|---|
| SanDisk (SNDK) | -0.59% | +2.31% | +2.93% | 6.99% |
| Western Digital (WDC) | +0.07% | +1.61% | +1.54% | 6.52% |
| Micron (MU) | -1.58% | -0.81% | +0.79% | 4.62% |
| Seagate (STX) | -0.95% | -0.38% | +0.58% | 5.03% |
| SMH (semis ETF) | +0.53% | -0.78% | -1.30% | 1.48% |
| SOXX (semis ETF) | +0.55% | -0.83% | -1.37% | 1.80% |
| NVIDIA (NVDA) | -0.23% | -1.64% | -1.41% | 1.80% |
| AMD | -1.23% | -0.98% | +0.25% | 2.62% |
| TSMC (TSM) | +0.13% | +0.71% | +0.59% | 2.01% |
| QQQ (Nasdaq 100) | -0.05% | +0.06% | +0.11% | 0.60% |
| SPY (S&P 500) | -0.07% | +0.12% | +0.19% | 0.28% |
How big are these moves, really? Start with a volatility baseline
Taking a single-day move straight to a news search is the most common error in this sector. The correct first question is whether the move falls inside the stock's own normal range.
HV20 below is self-computed — the annualized 20-day standard deviation of log returns, with the window ending at the 2026-08-07 close, deliberately excluding today's unfinished bar.
| Ticker | HV20 (ann.) | 1 daily sigma | Today | In sigmas | Day range in sigmas |
|---|---|---|---|---|---|
| SanDisk (SNDK) | 159% | 10.04% | +2.31% | 0.23σ | 0.70σ |
| Micron (MU) | 110% | 6.95% | -0.81% | 0.12σ | 0.66σ |
| Western Digital (WDC) | 110% | 6.90% | +1.61% | 0.23σ | 0.94σ |
| Seagate (STX) | 91% | 5.76% | -0.38% | 0.07σ | 0.87σ |
| SMH (semis ETF) | 51% | 3.23% | -0.78% | 0.24σ | 0.46σ |
| NVIDIA (NVDA) | 42% | 2.68% | -1.64% | 0.61σ | 0.67σ |
| SPY (S&P 500) | 14% | 0.90% | +0.12% | 0.13σ | 0.31σ |
The result is unambiguous: the largest reading in the table is 0.61σ, and not one name has travelled a full daily standard deviation. SanDisk's +2.31% is 0.23σ. Micron's -0.81% is 0.12σ. Even the full high-to-low range for today sits inside one sigma for every name in the table.
For that reason this article assigns no mechanism, catalyst, or news story to any individual single-day move today. A reading that falls inside the volatility envelope is a normal observation that is already priced.
What actually changed is the grouping
The signal is in the ordering, not the size. Last week versus today (last week measured close-to-close, August 3 → August 7, on a self-aligned window):
| Ticker | Last week (Aug 3→7) | Today, intraday | Rank shift | Earnings window |
|---|---|---|---|---|
| Western Digital (WDC) | -20.29% | +1.61% | ↑ worst to best | Aug 5, after close |
| SanDisk (SNDK) | -0.22% | +2.31% | ↑ worst to best | Aug 5, after close |
| Seagate (STX) | -5.07% | -0.38% | — | Jul 28 |
| Micron (MU) | +6.63% | -0.81% | ↓ best to worst | none (last Jun 24, next Sep 23) |
| SOXX (semis ETF) | +7.60% | -0.83% | ↓ best to worst | — |
The dividing line is not DRAM versus NAND. It is whether a stock had an event window.
- SanDisk and Western Digital both reported after the US close on August 5, and both beat: SanDisk EPS 39.25 versus 34.52 expected (+13.72%), Western Digital 3.56 versus 3.30 expected (+7.87%). Both beat, and both were sold the following session on August 6.
- Micron's last report was June 24 and its next is September 23. It had no event during the entire drawdown. The simplest explanation for Micron "holding up" last week is that it had nothing to trade.
That contrast yields a testable inference: if last week's memory selloff had been about deteriorating DRAM/NAND fundamentals, today's bounce should have included Micron. It did not. The bounce is occurring precisely and exclusively in the two names that an event had already forced out of position. That is the signature of a crowded-position unwind completing, not a re-rating of industry fundamentals.
The sample must be labeled: two names against two, over a single session. The falsification condition is item ③ at the end.
Dispersion is still intact: single stocks are violent, the index is calm
The market-structure question worth asking is not "will it fall" but "is one cluster falling, or is everything falling." Those two states have different causes and different endings.
- Broad de-risking looks like: the index falls alongside, and correlation between single stocks rises. Everything moves together because sellers are compressing gross exposure rather than expressing a view.
- A momentum unwind looks like: single-stock volatility is high while index implied correlation is low. A specific crowded position is being liquidated and the index absorbs it.
Today is the second, with unusually high contrast. The four memory names show intraday ranges of 4.62% to 6.99%, while over the same window SMH ranged 1.48% and SPY 0.28%. The annualized realized-volatility gap is wider still: SanDisk 159%, Micron 110%, Western Digital 110%, against SMH 51% and SPY 14%. At the index level right now, SPY is +0.12% and QQQ is +0.06%.
On the August 10 Korea close, a self-computed implied correlation for the semiconductor basket reads ρ = 0.354, in the 9.2nd percentile of its own series. The trailing 20-day mean of 0.384 is below the prior 20-day mean of 0.467 — still dispersing.
One counter-signal belongs alongside it. The S&P-referenced three-month implied correlation index (COR3M) has turned up on a 20-day mean basis, from 9.06 to 9.25. July averaged 8.72 and August is running 9.59, against a trough of 7.19 on July 10. The magnitude is very small — roughly the 1.3rd percentile over 20 years — so this is a warning, not a conclusion. It is worth stating because a turn upward in correlation is the first signal that a single-cluster unwind is migrating into broad de-risking, and it has just turned.
The most notable reading: stocks are violent, options are priced for calm
Three mutually independent measures point at the same thing. These are not one number converted three ways.
| Measure | Reading | As-of |
|---|---|---|
| Realized vol, HV20 annualized (self-computed) | SanDisk 159% / Western Digital 110% / Micron 110% | window ends 2026-08-07 |
| Realized vol, 252-day percentile | SK Hynix 98.8 / Samsung 98.4 / SanDisk 97.2 / Micron 87.3 | Aug 10 |
| Implied vol, IV Rank | MUU 0.0 / TSM 1.3 / AVGO 7.8 / SNDK 12.6 / NVDA 15.0 / MU 19.6 | Aug 8 |
| Variance risk premium (VRP) | Micron −19.26% / Western Digital −14.91% / MUU −57.03% / SNXX −90.60% | Aug 8 |
The implication: these stocks are genuinely moving at triple-digit annualized volatility while the options market is charging for calm. A single-digit IV Rank means implied volatility is sitting at the very bottom of its one-year range. A deeply negative VRP means implied is below realized, so whoever is buying volatility here is getting it cheap.
This matters most for leveraged products. Path decay in a 2x instrument scales with the square of realized volatility, and realized volatility is currently at its one-year high — a cost the option pricing has not marked. The reason leveraged memory instruments are fragile right now is not "sentiment is too optimistic." It is that decay is colliding with peak realized volatility.
Asia: rotation, with the buyer base downgrading
The following are final closing prices for August 10 in Korea, which closes before the US session opens.
| Ticker | Close | Change | Foreign net, % of volume | Institutions | Retail (derived) |
|---|---|---|---|---|---|
| Samsung Electronics | ₩230,000 | −0.43% | −26.9% | +3.8% | +23.1% |
| SK Hynix | ₩1,420,000 | −0.14% | −8.7% | +2.1% | +6.6% |
| Hanmi Semiconductor | ₩205,500 | +6.86% | +16.9% | +12.2% | −29.1% |
Two readings deserve isolating:
- Foreign net selling in Samsung, as a share of volume, has risen for three straight sessions: 12.0% → 8.5% → 26.9%, a new high on the third (same measure throughout, so the sequence is comparable). Yet the stock fell only 0.43% — retail absorbed it.
- Foreign flow in Hanmi Semiconductor flipped positive for the first time, from −18.6% in the prior session to +16.9%, and the stock rose 6.86%.
Non-memory Korean names rose broadly the same day: Samsung SDI +4.90%, Hyundai Motor +3.16%, LG Electronics +2.70%, LG Energy Solution +2.08%.
Read together: money is changing position within and across sectors, not leaving Korea. But the composition of the buyer base is deteriorating — in Samsung's case, foreigners sold and retail absorbed.
⚠️ Methodology note: the four-way zero-sum check on KOSPI market-level investor-type aggregates did not pass today. This article therefore draws no market-level conclusion about whether foreign capital is exiting Korea, and reports only exchange-disclosed single-stock data. The "retail" column above is derived by subtraction, not directly disclosed.
What would invalidate this article
A judgment that cannot be falsified is worthless. The five conditions below are fixed in advance. Any one of them triggering retires the corresponding conclusion.
| # | Trigger | What it invalidates |
|---|---|---|
| ① | Memory HV20 percentile falls from ~98 to below 80 and holds for three sessions | "Dispersion is intact" — the shakeout is over |
| ② | Any memory name's IV Rank jumps from single digits back above 40 | The implied-versus-realized divergence — pricing is catching up |
| ③ | Micron weakens for a second and third consecutive session while SanDisk/Western Digital keep strengthening | "The dividing line is the event window" — upgrade to genuine DRAM-versus-NAND fundamental divergence |
| ④ | S&P-referenced COR3M 20-day mean rises above 10 | "One cluster is moving" — it has become broad de-risking and diversification is failing |
| ⑤ | Foreign net selling in Samsung stays ≥20% for a fourth session and price breaks down | "Rotation" — local absorption has failed; re-model as contagion |
Sources and methodology
- US equities: Tiger live quote interface, as of August 10, 2026, 10:39 AM ET, roughly 69 minutes into the session. The market is still open; these are not closing prices and will change before the close.
- HV20: self-computed, annualized 20-day standard deviation of log returns, window ending at the 2026-08-07 close, excluding the unfinished August 10 bar.
- Last week's returns: August 3 → August 7, close-to-close, on a self-aligned window. Pre-packaged weekly fields were not used, since they can blend intraday prices into a weekly comparison.
- Earnings EPS: primary earnings-date fields, not media summaries.
- IV Rank / VRP: internal options signal library, as of August 8.
- Korean equities: August 10 closing values. Foreign and institutional figures are exchange-disclosed; retail is derived.
This is public-layer market-structure analysis. It contains views, data, and facts only, and no position or trade recommendations.