Brief
EvolveTrade rewrites a trading agent's system prompt from its own decision traces, leaving the model fixed
A new arXiv preprint describes EvolveTrade, a framework that revises a tool-using trading agent's system prompt after each update interval using accumulated decision traces and realized portfolio feedback. The evidence available to us is the abstract alone, and it reports no code, cost, data-handling or availability details.
What changes here is the policy layer, not the model. EvolveTrade treats a tool-using trading agent's system prompt as a text-parameterized policy. After each update interval, a Policy Agent revises that policy using accumulated decision traces and realized portfolio feedback, while the backbone LLM is held fixed. The revised policy is then used for the next batch of decisions, so the agent's procedure for gathering evidence, invoking tools, verifying signals and managing risk is what adapts over time.
The abstract reports that across multiple market regimes and two LLM backbones, EvolveTrade often improves Sharpe Ratio and Cumulative Return over fixed-policy LLM baselines, achieving the improved SR and CR in most evaluated settings. Behavioral analyses are said to show self-evolved policies increasing code-mediated analysis and activating regime-relevant computations, with case-level attributions tracing how policy-induced allocation changes relate to realized returns.
Our reading
For anyone shipping tool-using agents, the reusable artifact under revision is the prompt that decides how tools are called, checked and risk-managed — a surface most teams currently freeze by hand and then leave alone. If the reported behaviour holds, the interesting work shifts from model selection to maintaining the policy document: versioning it, tracing which revisions drove which outcomes,…
What to do or watch
Ask for the decision traces and the policy-update mechanism before treating the gains as portable: the unresolved question is whether the reported improvement survives an independent run with the same backbones, and whether the same trace-driven policy revision transfers to non-trading tool-use agents.
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 across multiple market regimes and two LLM backbones show that EvolveTrade often improves Sharpe Ratio and Cumulative Return over fixed-policy LLM baselines, achieving the improved SR and CR in most evaluated settings.
- 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.