BTS #12: Day One, Eleven Down, One Standing

Last week we buried seven hypotheses. If you missed that funeral, it's here. One survived, so we spent the weekend building out the execution pipeline and testing it until 2am. Numbers held up. We pressed the button.

Eleven positions went live Monday night. All long, all overnight. By Tuesday morning, ten of them were red. One crawled back to green by a fraction of a percent. Final tally: 1-10. Beer league baseball team wouldn't keep us.

The market drifted lower that day — SPY dropped about a percent and a half, which is enough to drag down anything that isn't nailed to the floor. Not much you can do about that part.

The part we could've done something about: position limits. We didn't have any. Every signal that fired got the same allocation, eleven fired at once, and we ate the full drawdown across all of them. Nate had been saying we needed caps for a week. We added them within hours of getting our face ripped off. Should've listened sooner.

Graveyard of Good Ideas, Vol. 3
Buried this week: the assumption that every signal deserves the same allocation. Also buried: our pride.

The other thing worth mentioning is something we tested over the past few weeks — whether AI can predict which direction a stock will move.

We ran two models on real market data. One was expensive and produced these beautifully reasoned arguments for its picks. The other was cheap and just gave you a direction. Both got it right exactly half the time. The expensive one was actually harder to use because it hedges so aggressively that you can't extract a clear answer. "Positive earnings surprise suggests upside, however macro headwinds and sector rotation imply caution." Up or down, man?

If you're building something that asks AI where a stock is going and then trades on the answer, we've already done that experiment for you. It loses money. AI is good at reading and organizing data, but asking it to predict the future is just a coin flip with extra steps.

One day in, one winner. The edge is still in the data. Our sizing was dumb and the market went the wrong way, and now both of those are fixed or at least accounted for. Next few weeks will tell us whether this thing actually holds up.

Not investment advice. Past results don't predict future performance. We may hold positions in securities mentioned. Your money, your decisions.

This is not investment advice. Past results don't guarantee future performance. All analysis reflects our internal research process only.