On January 4, 2025, a Saturday, I sat down and did the work. I wrote a twelve-page research note on Micron Technology: DRAM and NAND, the memory cycle, high-bandwidth memory for AI data centers, the only US-based memory manufacturer, the CHIPS Act money, the $100 billion fab in upstate New York. The stock had just been punished for soft guidance — down from $114 to $87 in three weeks — and the note read, in the polite language of research, like a coiled spring. I penciled out the total addressable market for data-center memory in the most bullish AI scenario. And I wrote down, in so many words, the real question: what could cause Micron’s stock price to go up 200% in 2025?
The next day I got as far as pricing out the options. Strikes, expiries, what a real position would cost. It was all sitting there on the screen.
And that was that. I closed the laptop. I never bought the stock, and I never bought an option. The note sat in a folder like a time capsule buried at the exact bottom tick.
PART I — THE NOTEWhat I actually knew, and when
I want to be precise about what was in that note, because hindsight inflates everything. The thesis wasn’t secret. It was three public facts arranged in the right order. One: memory is a brutal cycle, and Micron had just reported the upswing — fiscal 2024 revenue up 61%, first-quarter 2025 revenue a record $8.7 billion, data center over half of revenue for the first time in the company’s history. Two: the AI buildout runs on high-bandwidth memory, HBM supply was effectively sold out, and Micron was one of exactly three companies on earth that could make it — and the only American one, with CHIPS Act billions behind it. Three: the stock had just been marked down 24% in three weeks because of soft consumer demand — PCs and phones — the part of the business the thesis didn’t even need.
The street agreed, tepidly. After the December earnings drop, Bernstein cut its target from $140 to $120. Stifel went from $135 to $130, Cantor from $150 to $130, TD Cowen from $135 to $125. The most bullish target I could find anywhere was $250. The consensus was “strong buy, eventually.” Nobody — not the bulls, not me, not the AI I was interrogating on a Saturday night — wrote down anything like what happened: $87.33 on the day of my note to a $1,213.56 close seventeen months later — $1,255 at the intraday high. The memory supercycle didn’t beat the most optimistic estimate on the street; it quintupled it.
A fair reader should pause here, and so should I: a thesis that outruns the wildest bull on Wall Street by 5× is not evidence I was smart. It’s evidence I bought a lottery ticket’s worth of exposure to a regime change — and then didn’t buy the ticket. What this article can honestly test is not whether the thesis was right. The tape settled that. It’s a colder question: given that the thesis was right, would I have gotten paid? Because the last time I had a correct thesis — autumn 2021, Affirm and Ford, the subject of the first article on this site — the thesis was right and I still lost $89,000. The bottleneck was never the idea. It was the trading.
PART II — THE MACHINESix traders, one brain
Here’s what the first article left me with: a behavioral fingerprint, measured from 378 real round trips. The winning patterns had a shape — Monday-midday entries, near-the-money strikes around 0.45 delta, positions held across weekends for days, sold into strength around 1.5–2× cost. The losing patterns had a shape too — Tuesday-morning chases of Monday’s move, strikes one rung too far out with three days to live, averaging down into holes, riding carcasses to expiry. Same account, same thumb on the buy button.
So I turned each pattern into a machine. Every strategy below starts with $20,000 — what I started 2021 with — trades MU options only, sizes each position at the same fraction of equity I actually used in 2021 (median: 10% of the account per ticket), buys at the ask, sells at the bid, and pays my old broker’s exact 2021 commissions: $9.99 plus $1.25 a contract, each way. The quotes are the real NBBO, pulled hourly for 6,000+ MU contracts across every expiry from January 2025 to July 2026. No strategy sees the future; every rule is mechanical and was written before the backtest ran — though all of them, of course, were written by someone who has seen this movie’s poster. More on that sin in the caveats.
| Strategy | Enters | Exits |
|---|---|---|
| Best self only the patterns that made money in 2021 | Monday midday, only on a ≥2% dip from the 5-day high; calls at ~0.45 delta, 10–24 days out | Sell into strength at 1.75× cost or once deep in the money (delta ≥ 0.92); 2-session time-stop if never green; out the day before expiry; never averages down |
| As I traded the modal 2021 behavior, frequencies from the ledger | ~3 entries/week, tilted Thu–Fri; 94% calls; median ~3 days to expiry; mostly at-the-money to 12% out | Morning-heavy; quick harvests near 1.5×; quick punts when red; 11% of lots ride to expiry; occasional averaging down. Run 51 times; median shown |
| Worst self only the patterns that lost money in 2021 | Tuesday 10:30 a.m., chasing any +1.5% prior day; calls ~10% out of the money, ≤3 days to expiry | Averages down once when 40% underwater; sells only on capitulation (bid ≤ 30% of cost) or rides to expiry |
| Best in, worst out | Best self’s entries | Worst self’s exits |
| Worst in, best out | Worst self’s entries | Best self’s exits |
| Just buy the stock | 229 shares on day one, $9.99 commission | Never |
| Monthly ATM call the simplest options expression of the thesis | First trading day of each month: one at-the-money call ~30–60 days out, 10% of equity | Sold at month-end; rolled |
PART III — THE RACEEighteen months, six accounts
Same thesis. Same stock. Same starting stake. Same commission schedule. The only variable is behavior.
SIX WAYS TO TRADE A THESIS THAT WENT UP 11×
account equity, daily, log scale · marked to the bid · click legend to toggle · hover to inspect
FIG. 1 — The race.
PART IV — THE LEDGEREvery strategy, by the numbers
FIG. 2 — Final and maximum values are daily closes, marked at the bid. Fees are commissions only, at 2021 TD Direct rates. “As I traded” shows the median of 51 randomized runs.
PART V½ — THE BLOTTEREvery trade, named
Backtests earn trust by being auditable, so here is every position each strategy took — the exact contract, expiry and strike, when it was entered and exited, at what price, and why it closed. Filter by strategy; the strikes vary trade to trade because each rule set picks its own rung on the chain.
TRADE BLOTTER — exact contract, entry, exit
PART V — ENTRY vs EXITWhich half of the behavior mattered
The two hybrid strategies exist to answer the question the 2021 post-mortem kept circling: was the damage done at the entry or at the exit? Crossbreed the machines and the answer falls out.
THE FOUR CROSSES: FINAL VALUE OF $20,000
same tape, entries × exits · log scale
PART VI — THE HONEST PARTWhat this does and doesn’t prove
Every backtest is a flattering mirror, and this one has specific distortions you should know about before letting any number above rearrange your convictions:
The strategies were built in hindsight. The rules are mechanical and were frozen before this backtest ran, but they were distilled from a post-mortem of my own 2021 trades — which means “best self” is, unavoidably, curve-fit to a different market’s lessons. The honest claim is not “this strategy works.” It’s “this is what my documented good habits would have done on this specific tape.”
One stock, one regime, one of the great single-name runs of the decade. MU went up 11× in eighteen months. Nearly any long-calls process looks brilliant on that tape, and every one of them dies on the tape where MU goes to $40 instead. This is a study of behavior spread — the gap between versions of me on identical fuel — not of expected returns. The right benchmark for humility is sitting in the table above: the stock, bought once and never touched.
The fills are hypothetical and polite. I cross the full spread (buy at ask, sell at bid) and pay 2021 commissions, which is honest. But quotes are hourly snapshots — real fills happen between them, better and worse; there is no slippage model beyond the spread; and as the option accounts compound into six and seven figures, they’re quietly assuming size that weekly MU options would not always have absorbed without moving. Past a few hundred contracts, these curves describe physics with the friction turned off.
The randomized trader is a distribution, not a line. “As I traded” makes random choices with the 2021 ledger’s frequencies, so I ran it 51 times and plotted the median run; the full range is in the table. The others are deterministic: same tape in, same trades out, every time.
No taxes, no FX, no margin, no early assignment. All longs, cash-secured, and every position is closed or expires; the 2021 account’s punitive currency-conversion fees are left out of both sides.