Scuttlebutt

Scuttlebutt

Ex-AHL fund manager on AI: "the fourth thing it suggested would ... destroy the code"

13 lessons: why he has refit his own system about twice in 12 years, why a diversified CTA beats one matched to the portfolio, and his add-only backtest

Scuttlebutt's avatar
Scuttlebutt
Nov 22, 2026
∙ Paid

Top Traders Unplugged · Nov 21, 2026 · 1h 14m · 61m saved

With Rob Carver, Independent Systematic Trader, author and former Fund Manager at AHL

Ex-AHL fund manager on AI: "the fourth thing it suggested would ... destroy the code"

Rob Carver, an independent systematic trader, author and former fund manager at AHL, told Niels Kaastrup-Larsen on Top Traders Unplugged that he is up about 13.6% this year and at a new high-water mark, with a book that is only about 60% trend following. Most of the episode was about AI. New papers have LLM agents designing strategies and running portfolios. Rob's objection is that the same model that proposes a strategy should not be the one that tells its user the backtest works.

"Taking their trading strategy idea and then saying to the LLM model, does this trading strategy work?... I call it marking your own homework. Is this thing working? Yes, it's brilliant. How do you know that? I've written this back testing code that tests it well... Can I really trust that?" — Rob Carver

Rob has recalibrated his own trading system only about twice in 12 years, a debate he traces to his 2001 internship at AHL. He is now writing a book built around a stricter way to backtest.

The full episode is covered here so you can skip it. 74 minutes of audio, 13 minutes of reading.

Here are the 13 lessons that matter.


This Scuttlebutt Summary is available in full to premium subscribers. If you're not one already, please consider subscribing to keep reading.


Key Takeaways

  • Letting an LLM grade the backtest of its own strategy idea is "marking your own homework"

    • Commercial models have already seen the data up to their cutoff, so a true out-of-sample test is close to impossible

  • Rob has recalibrated his trading system only about twice in 12 years, and says constant refitting suits only high-frequency strategies

  • AI cannot produce an original idea, but it can prompt one, the way pulling random books off a library shelf does

  • A diversified trend-following fund beats one matched to the investor's own portfolio

    • Equity-only trend following on top of an equity-heavy portfolio mostly adds more equity beta

  • An AI code review found real bugs in Rob's system and proposed one change that would have broken it

  • Rob's fix is an "add-only" backtest: an idea can be down-weighted but never removed once it is in

1. Tesla's Late Robotaxi

Rob opened with a sarcastic riff on Tesla, which has finally delivered its long-promised robotaxi, 8 years late.

Rob Carver mocked the Cybercab's launch as years late and full of caveats

"The brilliant news is this week we finally have the robotaxis. They're not called that anymore, they're called Cybercabs... And for those who don't know me, everything I've just said was highly sarcastic." — Rob Carver

The car has 2 seats, runs only inside a geofenced area, can be disabled if a subscription lapses and has no steering wheel. The US National Highway Traffic Safety Administration is also looking into how it was certified. Niels added that California has banned Waymo's driverless cars from highways after several incidents.

2. Rob's Book at a New High

User's avatar

Continue reading this post for free, courtesy of Scuttlebutt.

Or purchase a paid subscription.
© 2026 The Golden Age LLC · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for writing, podcasting, and media founders.