30. What trading is for: the index handles wealth, the satellite handles trying (2026-10-25)
I asked the AI a question first: hand $10,000 to an AI, minimize human involvement, and what happens after three months, six months, one year, three years, five years? Its answer came in layers: three months is noise, six months is one regime sample, one year is when costs and decay start talking, three years is a survival question, five years is when base rates take over — most likely trailing the index.
So the next question followed: if trailing is the likely outcome, what is trading for at all — why not just buy index ETFs?
Split the buckets before arguing about winning. The core is already index logic: VUG, QQQM, VOO, broad bases held long-term — that is where the bulk of household retirement assets already live. For that bucket, “what is trading for” has no answer, and buying the index is the answer.
The satellite account is a different matter. The Schwab swing account is under 1% of household retirement assets; going to zero would not damage anything — that proportion is its license to exist. It has three purposes: a capped upside option, learning and participation, and, honestly, fun. But all three rest on one condition: small size, hard discipline, honest bookkeeping. The moment “learning” becomes an excuse for indefinite underperformance, the purpose is gone.
The aggressive account has one more honest reason: asymmetric bets. Someone who caught Amazon, Tesla, or Nvidia a decade or more ago is sitting on tens of times their money ten years later. That logic holds — with one honest addition: it points to a different structure than “holding winners.” Catching the next Amazon means a batch of small positions, most allowed to go to zero, waiting for one 30-bagger — venture math, not concentrated-winner math. Ten positions at 1% each, nine to zero and one to 30x, and the whole sleeve is 3x; but the hit rate on 30-baggers is tiny, and expected value is hit rate times payoff. Both are called aggressive, but they test different things: the first tests holding on, the second tests tolerating zeros.
The close: write the kill criterion in advance. Check the satellite against QQQM and VUG once a year; shrink it or shut it down after a full cycle of trailing. “Learning” without a kill criterion always decays into an excuse for indefinite underperformance. Same line as #23: the human sits in the correction seat — full AI automation doesn’t remove the human, it moves the human upstream, to objectives, risk limits, and circuit breakers.