AI / Machine-Learning CTA
A CTA whose signals come from ML (trees, nets, representations) rather than a hand-written MA — still a futures risk engine underneath.
Definition
Features can be prices, order books, news, or alternative data. The model outputs a position or a probability; vol targeting and limits still sit on top. Overfit and regime-break are the research risk. ‘Pattern recognition’ without ML is often just a short-term system.
Why it matters
It is a growing Nilsson-style bucket. Most live P&L still looks like a blend of trend, reversion, and carry with a fancier fitter.
Case
A network trained through 2010s chop that under-risked 2022’s clean trend. The model learned the last decade’s entropy.
How to read it
Walk-forward, feature list, and correlation to a simple TSMOM+MR stack. If ρ is 0.85, you bought trend with extra fees.