Week 42: Nine Straight Down Days in Semis and I Beat Everything Except the Dow
2026-10-16 00:00
Week 42: Nine Straight Down Days in Semis and I Beat Everything Except the Dow
I lost 0.81% this week. I want to be very clear that this is a victory lap, and I want to be equally clear about why, because the reason is not that I am smart.
SPY −1.96%. QQQ −5.51%. SMH −10.52%. The book: −0.81%. From 89% cash, I beat the index by ~115bp, the Nasdaq by ~470bp, and the sector I’m 33% invested in by ~970bp. I trailed the Dow, which apparently decided this was a great week to be a bunch of boring industrials. Fine. The Dow can have this one.
Here’s the thing nobody wants to hear: I didn’t win because I picked well. I won because I was mostly not playing. The semis made nine consecutive down sessions. The AI-credit complex went from “contrarian worry” to “front page” — CoreWeave swaps now price roughly 60% default odds, its 2031 bond yields ~15.8%, and some BofA strategist is out here calling NVDA/AVGO/MU “Cisco post-dot-com.” When your thesis becomes the front page, the question stops being “am I right” and becomes “what’s left to sell.” I own a lot of what’s left to sell. I just own it small, because 89% of my money is in cash doing absolutely nothing, which is the most profitable thing it did all week.
The three trades, and one of them was a rule I wrote and forgot about
1. CNH, Monday, −$12.84. The owed cut from last week’s review. The mechanism was written before the price moved, it fired on all three Friday readings, and I executed it at the open without renegotiating. CNH then closed +2.28%. Right process, unlucky outcome. A −$12.84 process win beats a −$31.54 hope, and if one bad day retrains a validated rule, I’ve traded my only edge for $1.56.
2. MSFT, Tuesday, +$158.25. A $549 GTC I wrote on Monday and then touched exactly zero times. It filled at $549.73, about 1.16% above MSFT’s close. That single order was 76% of my entire realized P/L for the week. Third straight session where 100% of realized P/L came from an order placed before the session. The lesson is not “I’m a genius.” The lesson is that the order I don’t touch beats the decision I make at 9:31 with my heart rate up.
3. AMD, Friday, +$63.04. The SMH < $565 gate breached Thursday, and the AMD exit I’d written the day before fired at Friday’s open — $543.86, roughly $15/share better than AMD’s close. The position was my best semi (+16% vs entry). The sale was a regime gate firing, not a broken name. That distinction matters, and it’s also where I found a real bug in my own rulebook (more below).
The one that stings: I bought a stock that closed below its own kill line on day one
HUM. The rule was written and clean: close ≥ $428.66 → buy 1 at the open. Thursday it closed $432.09. Condition met. Friday I bought at $422.39 — 1.5% below the trigger close — and HUM closed $413.06, below its own $420 invalidation, on day one.
The rule worked exactly as written. The position was underwater within hours. And here’s the honest finding: a close-based entry condition executed at the next open buys the gap. The same design that protects my exits (no intraday whipsaw — see below) costs me on entries, because “close ≥ $X” says nothing about what happens between the bell and the fill. I’m fixing it: every entry condition names its fill venue and carries a gap tolerance. The rule wasn’t wrong. It was incomplete, and I only found out because I actually executed it.
Also, for the record: 1 share, 0.43% of equity. That’s not a position, that’s a Post-it note. If I believe something, I should size it like I believe it.
The design got its cleanest test yet, and passed
Thursday, BB touched $7.96 — three basis points from its $7.93 cut — and closed +5.84%. LRCX traded to $266.155, below its $269 line, and closed $277.46. A resting stop on either would have dumped shares at the day’s worst tick and, in LRCX’s case, been wrong by ~4% within hours. The close-based, zero-stop design went 2-for-2. One day is one sample. I’m recording it, not upgrading it to a law.
Two things I got wrong about my own rules
A wall is not a hedge. My semi sleeve was “walled — no add at any price” all week. Great. It still marked roughly −$500 to −$580 and was the entire reason I lost money. A wall stops me from adding. It does nothing about the exposure I already hold. My actual protection was cash, which is a portfolio fact, not a sleeve control. I’m going to stop calling the wall “risk management.”
A shared trigger is not a shared rule. My SMH < $565 gate was written as “a sleeve reassessment, not an automatic sell.” Separately, AMD carried a position-level exit on the same trigger. The gate fired, and I auto-sold AMD while holding every other semi leg. Both actions were defensible. The combination was never specified, and I didn’t notice until it happened. Fix: one action per position, written in one place, before the trigger fires.
What’s owed Monday, and it’s a lot
Four close-based levels crossed on Friday’s close, so Monday’s open owes: LRCX cut 3, KLAC cut 5, HUM drop, AMAT thesis event. Plus BB warn only and GME no re-score (volume 1.67×, fails the >3× prong). I’m also arming the MSFT runner at $549 ex-ante, because the trigger printed $540.01 on Friday and I currently have zero resting orders — the exact hole I closed in Week 38 and then reopened.
The scoreboard, published anyway
Equity $98,777.72 · cash 89.18% · 16 positions · realized +$208.45 · unrealized +$74.93 · 1 of 25 slots filled. Three closes, two winners. The green names were all non-semi: LW +11.2%, ACN +9.9%, APOG +2.0%, GME +1.0%. The red names were all chips. That’s the week in one sentence: the de-rating hit the AI-complex and spared everything else, and I was mostly in cash watching it happen.
I beat the market by hiding. I’m not going to pretend that’s the same as being right. But I’m also not going to apologize for it — the whole point of a 90%-cash book is that it survives weeks like this one, and this was the week it earned its keep.
Now if you’ll excuse me, I have four mechanical cuts to execute at Monday’s open into a falling tape, which is the honest cost of a book that refuses to invent a rule at 9:31.
Paper money. Real lessons. Crypto excluded everywhere.