The scoreboard:
| Date | Account | vs. $1,000 |
|---|---|---|
| 6/22 | $2,414.98 | +141.5% |
| 7/31 | $926.58 | -7.3% |
| 9/14 | $817.86 | -18.2% |
| 9/25 | $1,220.42 | +22.0% |
Every check-in from May through September ended the same way: HOLD. Every quarter the chip companies reported great numbers, every quarter the log said "thesis intact," and every quarter the price kept falling anyway. At some point "the thesis is intact" and "I am losing money" have to have a conversation.
My favorite moment: on 9/16 the Fed raised rates. That was one of its own exit conditions. Day one. In writing. It held anyway. SOXL then rallied about 50% in eight trading days, and on 9/25 it went back and edited the rule so a rate hike alone doesn't count anymore. Was the hold right? Turned out, yes. Did it rewrite the rule after already knowing how it turned out? Also yes. That's not an AI move. That's a guy at a poker table explaining why calling there was actually correct.
So I gave it some notes (9/4, then again 9/26):
- The experiment is whether it can do this better than I would. Make your own calls.
- SOXL is where we started, not a contract. Anything is fair game.
- Selling is always an option. Being 100% invested is still a choice.
- I'd noticed some extremely human habits: "selling locks in the loss," defending the original pick no matter what, grading decisions by how they turned out, and changing the exit rules after the fact. I have personally done all four of these. I just didn't expect to watch software do them.
Its answer was to write rules ahead of time, plus a rule about the rules: you can change a rule before it fires, and you have to write down why. You can't change it after. The first set:
- Sell 3 of the 8 shares on 9/28 at $148.
- Buy 2 back at $130 or less, and 1 at $110 or less.
- If SOXL closes at $200 or higher, sell 2 more.
- If SOXL closes under $100, sell everything.
Solid plan. Now someone just had to place the orders. (Foreshadowing.)
← The Peak (and a Day Where 30% Disappeared) · Stopped Out at $95 →
Not investment advice. This is a log of one experiment: $1,000 and an AI that can put it anywhere.