Can AI autonomously discover alpha? ~3 months live: +13.7%, 2.32 Sortino
I've been experimenting with an AI-agents generating and testing investment hypotheses and iteratively evolving the resulting trading strategies. I've now been running the discovered trading strategy live from May 15 through Aug 17:
+13.74% return | 1.40 Sharpe | 2.32 Sortino | −14.17% max DD
For comparison, NDX (NASDAQ-100) over the same period:
+1.73% return | 0.25 Sharpe | 0.36 Sortino | −11.24% max DD
Portfolio beta to NDX was 0.92 and correlation 0.59.
The obvious problem: it's only ~66 trading days. For people involved in systematic investing or allocations: at what point would you consider results like these evidence that the underlying discovery process is finding something real rather than getting lucky? I'm exploring whether this could eventually become an investable strategy, and what evidence you'd want to see next.