研究者は自己進化する LLM 取引エージェントのために EvolveTrade を導入

一次情報源文書記録されたソース 2026年9月17日 6:56 UTC

出版社: arxiv.org
ソースリンク: arxiv.orghttps://arxiv.org/abs/2609.17632
ソースの種類: 一次文書 — 私たちが直接読む公式発表、論文、提出書類、またはファーストパーティのページ。

コンテキスト60秒で理解できる

ここから始めましょう

重要な用語

大規模言語モデル (LLM): テキストを生成および分析するために大規模なテキストコーパスでトレーニングされた言語モデル。
システムプロンプト: モデルの動作、ポリシー、および応答スタイルを設定する優先度の高い命令。
堅牢性: ノイズ、シフト、または敵対的な入力の下でパフォーマンスを維持するモデルの機能。

何が起こったのか

Researchers introduced EvolveTrade, a self-evolving framework for large language model LLM trading agents. The framework treats the of a tool-using trading agent as a text-parameterized policy, which is revised after each update interval using accumulated decision traces and realized portfolio feedback.

The framework treats the of a tool-using trading agent as a text-parameterized policy, which is revised after each update interval using accumulated decision traces and realized portfolio feedback.

This approach enables LLM trading agents to adapt their behavior under changing market regimes, improving their performance and .

The development of EvolveTrade has significant implications for the field of LLM trading agents, highlighting the importance of adapting the reusable procedure governing tool use to build more robust agents.

ソースの詳細: arxiv.org ↗

なぜそれが重要なのか

EvolveTrade enables LLM trading agents to adapt their behavior under changing market regimes improving their performance and . This is a key direction for building more robust LLM trading agents.

EvolveTrade improves the performance of LLM trading agents by enabling them to adapt their behavior under changing market regimes.

The framework achieves improved Sharpe Ratio and Cumulative Return over fixed-policy LLM baselines in most evaluated settings.

Behavioral analyses show that self-evolved policies increase code-mediated analysis and activate regime-relevant computations.

Case-level policy-to-return attributions trace how policy-induced allocation changes contribute to realized return differences.

Interactive Mechanism

インタラクティブなメカニズム: 実際にどのように機能するか

この開発の背後にある基盤となるテクノロジーをインタラクティブに探索します。

Simulate Tool Error (API 404 / Network drop)Require Human Approval Guardrail

Agent Lifecycle Stage:

User Intent & Planning: "Audit customer refund request #4092 and settle payment."

Tool Calling: Emits structured JSON call crm_get_transaction(id='4092').

Guardrail & Verification:🛡️ Paused: High-value action requires human operator sign-off.

Final Settlement: Refund recorded, email receipt dispatched, and audit log stored.

Core takeaway: An AI agent is not just a language model—it is a closed loop of planning, tool invocation, and environment feedback. Production systems require self-healing retries and strict human approval guardrails.

インタラクティブコンセプトチェック+10 Points

What is AI? Quiz

A route planner searches possible journeys using explicit rules. What does this illustrate about AI?

次に見るべきもの

The development of EvolveTrade has significant implications for the field of LLM trading agents. It highlights the importance of adapting the reusable procedure governing tool use to build more robust agents.

The impact of EvolveTrade on the performance of LLM trading agents will be closely watched.

The framework's ability to adapt to changing market regimes will be a key area of focus.

The potential applications of EvolveTrade in other areas such as finance and economics will be explored.