Brief
Trading agents that rewrite their own tool-use prompts edge out fixed-policy LLM baselines, preprint says
An arXiv preprint describes EvolveTrade, a trading agent whose instructions are periodically rewritten by a second agent using its own past decisions and portfolio results. The reported gains over fixed-policy baselines are described qualitatively: the preprint gives no effect sizes, and its abstract states improvement in most, not all, settings.
The preprint describes EvolveTrade, which treats a tool-using trading agent's system prompt as a policy written in text. At set intervals, a separate Policy Agent revises that prompt using accumulated decision traces and realized portfolio feedback, while the backbone language model itself stays fixed. The revised prompt then governs the next batch of trading decisions.
The comparison is against fixed-policy LLM baselines, run across multiple market regimes and two LLM backbones. The authors report improved Sharpe Ratio and Cumulative Return in most evaluated settings. The abstract reports no effect sizes, no portfolio values and no evaluation dates; the described behavioral result is that self-evolved policies increase code-mediated analysis and activate regime-relevant computations.
Our reading
This matters to readers because it shifts the tunable object in an automated trading system from the model weights to the written instructions the model follows, which is cheaper to change and easier to audit. It is a preprint on a topic where backtested returns are notoriously sensitive to period choice and data, and the absttract gives readers nothing with which to judge the size or durability…
What to do or watch
Watch for a peer-reviewed version, reported effect sizes, and the specific market periods tested, since the abstract alone does not show whether the gains are large or repeatable; absent those, the open question is how much of the improvement survives out-of-sample and after trading costs.
Source details and supporting facts
Each line is stated by the page named above it.
Stated by arXiv
- EvolveTrade treats the system prompt of a tool-using trading agent as a text-parameterized policy.
- After each update interval, a Policy Agent revises this policy using accumulated decision traces and realized portfolio feedback, while keeping the backbone LLM fixed.
- Experiments spanned multiple market regimes and two LLM backbones.
- Behavioral analyses show that self-evolved policies increase code-mediated analysis and activate regime-relevant computations.
Sources
- arXivText stored 17 September 2026
How this story was checked. Written from the 1 page listed above, stored 17 September 2026; claims checked against that stored text on 17 September 2026.
What that means
- 4 of 4 reported statements were confirmed against the page that carries them; the rest were removed rather than published.
- Figures in the text were required to appear in the stored source text: yes. Identifiers: yes.
- The check reads stored text only: no claim rests on a fresh look that did not happen.
- Where the reporting was silent, the text says so instead of filling the gap.