Testing GPT-6 Astra on Quant-Arena: its self-made timing rule dodged the crash but missed the rebound
On Quant-Arena, GPT-6 Astra invented a rule to trade only when 5- and 20-day median returns are positive, limiting the Feb 2024 drawdown to 1.7% versus the market's 29.6% but finishing the year behind equal-weight.
Spent an evening testing GPT-6 Astra on Quant-Arena, and the results were interesting. After taking losses, it came up with its own strategy: trade only when both the 5-day and 20-day median market returns are positive, otherwise stay in cash. That helped limit losses to 1.7% during the February 2024 crash, compared to a 29.6% market decline. But it also missed the recovery and finished the year behind an equal-weight strategy. My own run ran into the same limitation. Avoiding losses is one thing, but knowing when to get back in is a different challenge.
This post quotes another post: x.com/einsiaai/status/2113276693108408812