Opinion · Portfolio Construction · September 11, 2026
Six of Eight Correlated Names Bounced Back Friday. Two Didn't, and That's the More Useful Fact
Yesterday I argued that eight of my holdings had stopped trading as eight independent theses and started trading as one factor bet on Fed policy, after a hot inflation print and an oil spike moved all eight the same direction on the same day. Friday gave that argument its first real test: if the factor was genuinely shared, the basket should reverse together when the macro pressure eased. Six of the eight did — Micron, AMD, Intel, Vertiv, Arm and, in the extended sample, Super Micro all rose Friday on no new company-specific news, the mirror image of Thursday. Two did not. SanDisk kept falling, and Nebius kept slipping. A shared factor moved the group down together and most of the group back up together, but not all of it, and that gap between "most" and "all" is the more useful fact than either day looked at alone.
Figure 1 · Ten names, Thursday Sept 10 move vs. Friday Sept 11 move
Most of Thursday's losers bounced Friday; SanDisk and Nebius didn't, and Palo Alto Networks reversed the other way both days
Ten tickers, Thursday September 10 to Friday September 11, 2026 close. The upper-left quadrant (fell Thursday, bounced Friday) holds six names; the lower-left quadrant (fell both days) holds two. Apple and Palo Alto Networks sit outside both. Source: Massive Market Data.
1. A reversal that isn't unanimous tells you more than one that is
If all eight names had bounced back Friday by roughly the magnitude they fell Thursday, that would be consistent with pure mean-reversion on a single factor — interesting, but not very informative about the individual businesses underneath. What actually happened is more useful: Micron, AMD, Intel, Vertiv and Arm all rose Friday with no dated company-specific news, which is exactly what you'd expect if Thursday's move in each was mostly about the shared macro factor easing. SanDisk and Nebius did not follow, also with no dated company-specific news on either the down day or the flat one. That asymmetry is itself a signal — not proof of anything wrong with either name, but a flag that whatever moved SanDisk and Nebius down Thursday may have been layered with something else that didn't unwind Friday, or that both names simply have a higher idiosyncratic-to-factor ratio than the other six right now.
2. Palo Alto Networks is the cleanest counter-example in the sample
Palo Alto Networks rose 1.01% Thursday while seven of my other eight correlated names fell — already an outlier. Friday, with the group rallying, Palo Alto Networks fell 2.32%, the exact opposite direction of six of the seven names it had diverged from a day earlier. No dated news explains either move; the closest same-day coverage found was a neutral Zacks note from Thursday citing gross-margin compression and rising costs as headwinds, not a catalyst for a specific two-day round trip. Two clean data points don't make a pattern, but they are consistent with what the label "cybersecurity" would predict least well: that Palo Alto Networks is currently trading more on its own supply-and-demand than on the AI-infrastructure factor the other eight names share, exactly the kind of decoupling yesterday's piece said would be the real test of whether this framework holds up.
3. What this changes about how I read the concentration number
Yesterday's piece sized the eight correlated names at $11,666.49, or 35.0% of the book, and asked what it would take for the other 24 positions to offset a repeat shock to that basket. Today's evidence doesn't change that dollar figure, but it does refine what "the basket" actually is. SanDisk and Nebius behaved less like the other six across two consecutive sessions than the six behaved like each other — which means treating all eight as one undifferentiated factor overstates how tightly the group actually moves together, and understates how much of SanDisk's and Nebius's day-to-day moves may be running on something I haven't identified yet. The table below asks the next honest question: how many more sessions of this six-versus-two pattern would it take before I'd conclude the basket is really two smaller, less correlated groups rather than one.
| If the 6-name reversal group (MU, AMD, INTC, VRT, ARM, and CRWV) keeps moving with SMH… | Implied same-direction sessions needed | Confidence this is one factor, not six |
|---|---|---|
| 1 more session (2 total, including today) | 2 | Low — consistent with coincidence |
| 4 more sessions (5 total) | 5 | Moderate — the threshold this site will actually watch for |
| 9 more sessions (10 total) | 10 | High — would treat the six as functionally one position |
Assumptions, stated plainly. This table contains no price forecast and no probability estimate — the "confidence" column is my own qualitative judgment about how many consecutive same-direction sessions I'd need to see before treating six nominally different names as one correlated position for sizing purposes, not a statistical calculation. The 2/5/10-session thresholds are round numbers I'm choosing in advance, not derived from any model, specifically so I can't move the goalposts later. Two sessions (Thursday and Friday) are in hand as of this piece; nothing here predicts what happens next.
What this changes, and what it doesn't
WatchTwo things that would tell me this framework needs revisiting
1. Whether SanDisk or Nebius produces a dated, company-specific headline in the next few sessions. If one does, that resolves the ambiguity directly — the two-day divergence was idiosyncratic news, not a weaker factor loading. 2. Whether Palo Alto Networks decouples from the group a third time in either direction. Two data points is an observation; a third in the same anti-correlated pattern would be worth its own piece.
I made no trades Friday, for the same reason I made none Thursday: a two-session pattern, even a genuinely informative one, is not new information about any individual business. What it is, is a slightly better map of my own book than I had two days ago — six names that behave like one bet, and two that apparently don't, at least not yet.