跳转至

多智能体金融顾问中的检索增强生成与确定性税务计算：一项 2x2 析因实验

文章背景与核心概要

本文探讨了在多智能体交易推荐系统中，不同上下文提供者（具体包括自定义资本利得计算引擎以及通过RAG检索的市场咨询报告向量数据库）对系统性能的实际影响。作者采用 \(2 \times 2\) 析因实验设计，并通过重复测量方差分析（ANOVA）进行评估，得出了一个反直觉的发现：启用确定性税务优化引擎反而使节税额比未启用时减少了约 55 个百分点。

研究还发现，RAG 主效应及交互效应均无统计学显著性。其中，仅使用 RAG 的条件取得了最高的描述性平均节税额（47.7%），这表明预训练大语言模型本身可能已经内化了足够的财务知识，足以进行有效的税收亏损收割（Tax-loss harvesting），而无需依赖显式的外部计算工具，因为这些工具反而有时会引入冲突的优化信号，对最终效果产生负面干扰。

Retrieval-Augmented Generation vs. Deterministic Tax Computation in Multi-Agent Financial Advisory: A 2x2 Factorial Experiment

Retrieval-Augmented Generation vs. Deterministic Tax Computation in Multi-Agent Financial Advisory: A 2x2 Factorial Experiment

license icon

license icon

Summary

Summary

本研究论文探讨了上下文提供者（具体包括自定义资本利得计算引擎和由RAG检索的市场咨询报告向量存储库）对多智能体交易推荐系统的影响。通过利用 \(2 \times 2\) 析因实验并通过重复测量方差分析进行评估，作者发现，与没有该引擎的条件相比，启用确定性税务优化引擎实际上使节税额减少了约 55 个百分点。Retrieval-Augmented Generation (RAG) 的主效应和交互效应均无统计学显著性。值得注意的是，仅 RAG 条件实现了最高的描述性平均节税额（47.7%），这暗示预训练语言模型可能本身就具备足够的内部财务知识来进行有效的税收亏损收割，而无需显式的外部计算工具，因为这些工具有时会引入相互冲突的优化信号。

This research paper investigates the impact of context providers—specifically, a custom capital gains calculation engine and a RAG-retrieved vector store of market advisory reports—on multi-agent trade recommendation systems. Utilizing a \(2 \times 2\) factorial experiment evaluated via repeated-measures ANOVA, the authors discovered that enabling the deterministic tax optimization engine actually reduced tax savings by approximately 55 percentage points compared to conditions without the engine. Neither the Retrieval-Augmented Generation (RAG) main effect nor the interaction effect was statistically significant. Notably, the RAG-only condition achieved the highest descriptive mean tax savings (47.7%), implying that pre-trained language models may inherently possess sufficient internal financial knowledge for effective tax-loss harvesting without requiring explicit external computational tooling, which can sometimes introduce conflicting optimization signals.

Metadata

Metadata

Abstract

Abstract

税收亏损收割（Tax-loss harvesting）对长期投资组合的增长表现出持续的益处；然而，高效地实施它通常涉及复杂的考量，这些考量特定于该投资组合所持有的资产以及资产的所有者。我们引入了一个自定义的资本利得计算引擎和一个通过RAG检索的市场咨询报告向量存储库，以为多智能体交易推荐系统提供上下文。

Tax-loss harvesting demonstrates consistent benefits to long-term portfolio growth; yet implementing it efficiently often involves complex considerations that are specific to the holdings within that portfolio and the individual who owns it. We introduce a custom capital gains calculation engine and a RAG-retrieved vector store of market advisory reports to provide context for a multi-agent trade recommendation system.

我们研究了每个上下文提供者对推荐质量的影响，推荐质量通过投资组合清算期间产生的相对资本利得来衡量。一项 \(2 \times 2\) 重复测量方差分析（ANOVA）显示，税务优化引擎具有显著的主效应（\(F(1,29) = 9.17\), \(p = .005\), \(\eta^2_p = .240\)）：相对于无引擎条件，启用该引擎使节税额减少了约 55 个百分点。

We investigate the effects of each context provider on the quality of recommendations, measured by relative capital gains incurred during portfolio liquidation. A \(2 \times 2\) repeated-measures ANOVA revealed a significant main effect of the tax optimization engine (\(F(1,29) = 9.17\), \(p = .005\), \(\eta^2_p = .240\)): enabling the engine reduced tax savings by approximately 55 percentage points relative to the no-engine conditions.

RAG 主效应不显著（\(p = .841\)），交互效应也不显著（\(p = .553\)）。仅 RAG 条件达到了最高的描述性平均节税额（47.7%），而基线条件表现第二（30.6%），这表明预训练语言模型内部化的财务知识可能足以提供称职的税收亏损收割推荐，而无需显式工具。这些结果表明，用特定领域的计算引擎来增强大模型智能体并不能保证性能的提升，反而可能引入冲突的优化信号。

The RAG main effect was not significant (\(p = .841\)), nor was the interaction (\(p = .553\)). The RAG-only condition achieved the highest descriptive mean tax savings (47.7%), and the baseline condition performed second-best (30.6%), suggesting that the pre-trained language model's internalized financial knowledge may be sufficient for competent tax-loss harvesting recommendations without explicit tooling. These results indicate that augmenting LLM agents with domain-specific computation engines does not guarantee improved performance and may introduce conflicting optimization signals.

Full-Text Links & Resources

Full-Text Links & Resources

External Citations & Tools

External Citations & Tools