发表机构
Scalar Field Research（标量场研究）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本研究提出代理型ETF概念，即由大语言模型驱动的自主交易代理执行投资流程的基金，并定义其基础设施六层架构，预测到2030年资产管理规模可达数千亿至数万亿美元。
AI 中文摘要
公共市场中正汇聚三大结构性趋势：交易所交易基金（ETF）已成为主导性的基金载体，主动管理型ETF是其中增长最快的细分领域，而算法系统如今已中介了绝大多数交易量。我们认为，这三者的交汇定义了一个新兴资产类别——代理型ETF（Agentic ETF），即一种投资流程被委托给自主的、由大型语言模型驱动的交易代理的基金，这些代理能够对数据进行推理并在无人工介入的情况下执行交易。我们定义了代理交易，并将其与基于规则的算法交易和机器人顾问区分开来，将代理交易基础设施分解为六个层级，并识别出每个层级中的现有参与者和结构性缺口。随后，我们展示了https://this.url作为集成式代理交易栈的参考实现——涵盖统一的多场所执行注册表、具有持续券商对账的隔离策略代理、第一方市场数据以及计量计算——并将其原语映射到栈的每个层级。一项透明且基于明确假设的规模测算表明，到2030年，代理型ETF的资产管理规模有望达到数千亿至数万亿美元的低位区间。
英文摘要
Three structural trends are converging in public markets: the exchange-traded fund (ETF) has become the dominant fund wrapper, actively managed ETFs are the fastest-growing segment within it, and algorithmic systems now mediate the overwhelming majority of trading volume. We argue that their intersection defines a nascent asset class -- the \emph{Agentic ETF}, a fund whose investment process is delegated to autonomous, large-language-model-driven trading agents that reason over data and execute without a human in the loop. We define agentic trading and distinguish it from rule-based algorithmic trading and robo-advice, decompose the agentic-trading infrastructure into six layers, and identify the incumbent players and structural gaps at each layer. We then present ScalarField.io as a reference implementation of an integrated agentic-trading stack -- spanning a unified multi-venue execution registry, isolated strategy agents with continuous broker reconciliation, first-party market data, and metered compute -- and map its primitives to each layer of the stack. A transparent, assumption-stated sizing exercise suggests the Agentic ETF could plausibly reach hundreds of billions to low trillions of dollars in assets under management by 2030.
