Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Market-Dependent Communication in Multi-Agent Alpha Generation

Jerick Shi, Burton Hollifield

TL;DR

Problem: evaluating whether and how analysts should communicate in multi-agent LLM trading systems. Approach: 5 homogeneous agents generate WorldQuant-style alphas under five organizational structures across 3 market universes, in 450 experiments over 21 months, with returns-based capital reallocation and standard metrics such as total returns, Sharpe ratio, and pairwise allocation similarity. Key findings: communication improves performance in a market-dependent way—competitive dialogue excels in volatile tech stocks and collaborative dialogue excels in stable general stocks—while finance stocks resist benefits; all configurations converge to similar final strategy correlations, indicating transparency does not erode diversity and that behavioral mechanisms drive performance differences. Implications: optimal communication design should align with market regime, and evaluating conversation content beyond superficial quality metrics is crucial for robustness and the practical effectiveness of multi-agent alpha generation.

Abstract

Multi-strategy hedge funds face a fundamental organizational choice: should analysts generating trading strategies communicate, and if so, how? We investigate this using 5-agent LLM-based trading systems across 450 experiments spanning 21 months, comparing five organizational structures from isolated baseline to collaborative and competitive conversation. We show that communication improves performance, but optimal communication design depends on market characteristics. Competitive conversation excels in volatile technology stocks, while collaborative conversation dominates stable general stocks. Finance stocks resist all communication interventions. Surprisingly, all structures, including isolated agents, converge to similar strategy alignments, challenging assumptions that transparency causes harmful diversity loss. Performance differences stem from behavioral mechanisms: competitive agents focus on stock-level allocation while collaborative agents develop technical frameworks. Conversation quality scores show zero correlation with returns. These findings demonstrate that optimal communication design must match market volatility characteristics, and sophisticated discussions don't guarantee better performance.

Market-Dependent Communication in Multi-Agent Alpha Generation

TL;DR

Problem: evaluating whether and how analysts should communicate in multi-agent LLM trading systems. Approach: 5 homogeneous agents generate WorldQuant-style alphas under five organizational structures across 3 market universes, in 450 experiments over 21 months, with returns-based capital reallocation and standard metrics such as total returns, Sharpe ratio, and pairwise allocation similarity. Key findings: communication improves performance in a market-dependent way—competitive dialogue excels in volatile tech stocks and collaborative dialogue excels in stable general stocks—while finance stocks resist benefits; all configurations converge to similar final strategy correlations, indicating transparency does not erode diversity and that behavioral mechanisms drive performance differences. Implications: optimal communication design should align with market regime, and evaluating conversation content beyond superficial quality metrics is crucial for robustness and the practical effectiveness of multi-agent alpha generation.

Abstract

Multi-strategy hedge funds face a fundamental organizational choice: should analysts generating trading strategies communicate, and if so, how? We investigate this using 5-agent LLM-based trading systems across 450 experiments spanning 21 months, comparing five organizational structures from isolated baseline to collaborative and competitive conversation. We show that communication improves performance, but optimal communication design depends on market characteristics. Competitive conversation excels in volatile technology stocks, while collaborative conversation dominates stable general stocks. Finance stocks resist all communication interventions. Surprisingly, all structures, including isolated agents, converge to similar strategy alignments, challenging assumptions that transparency causes harmful diversity loss. Performance differences stem from behavioral mechanisms: competitive agents focus on stock-level allocation while collaborative agents develop technical frameworks. Conversation quality scores show zero correlation with returns. These findings demonstrate that optimal communication design must match market volatility characteristics, and sophisticated discussions don't guarantee better performance.

Paper Structure

This paper contains 32 sections, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Results
  5. Communication Effectiveness Depends on Market Characteristics
  6. All strategies converge equally, but performance still varies
  7. Conclusion
  8. Appendix: Extended Experimental Design
  9. Statistical Design and Computational Infrastructure
  10. Market Data and Universe Construction
  11. Alpha Expression Function Library
  12. Appendix: Prompt Templates
  13. Context Template
  14. Alpha Generation Prompts
  15. Conversation Prompts
  16. ...and 17 more sections

Figures (3)

  • Figure 1: Overview of multi-agent trading framework.
  • Figure 2: Communication effectiveness varies by market characteristics. Error bars show 95% confidence intervals across 30 iterations.
  • Figure 3: Mean pairwise strategy correlations from Month 1 to Month 21. All configurations converge to similar final correlations regardless of information sharing, including isolated baseline agents. Error bars show 95% confidence intervals across 30 iterations.