Agentic Quantitative Trading: A Survey of Workflows, Systems, and Evaluation

Battle Count: 83

This is a directly relevant survey paper that comprehensively covers agentic quantitative trading across the entire trading pipeline. It provides a taxonomy of workflows, systems, and evaluation methods specifically designed for quantitative trading. The paper identifies key gaps (signal discovery dominance, aggregation reliance, evaluation limitations) and future directions for the field. It reviews 20+ agentic trading systems and 15+ benchmarks, making it a foundational reference for researchers and practitioners in agentic quantitative trading.

As a survey paper, it does not implement a specific system. However, the systems it reviews range from single-agent sequential pipelines to complex multi-agent orchestration frameworks with MCP/A2A protocols, typed JSON contracts, deterministic hard gates, layered memory systems, and evolutionary strategy archives. The complexity of implementing complete agentic trading systems (covering all five pipeline stages) is very high, requiring integration of LLM reasoning, tool use, portfolio optimization, execution algorithms, and risk management.

As a survey paper, reproducibility depends on the underlying systems reviewed. The paper provides detailed taxonomies and comparisons (Tables 1 and 2) that facilitate understanding of the landscape. However, many referenced systems lack public code. The paper identifies reproducibility challenges including information leakage, temporal contamination, and lack of standardized evaluation protocols across the field.