Agentic Artificial Intelligence in Finance: A Comprehensive Survey

Battle Count: 50

This is a directly relevant comprehensive survey covering the full spectrum of agentic AI applications in quantitative trading and financial markets. It addresses multi-agent trading systems, portfolio optimization, risk management, market microstructure impacts, liquidity provision, price discovery, arbitrage, and regulatory frameworks. The paper provides detailed technical architectures (MADDPG, FINMEM, FS-ReasoningAgent, ABIDES-MARL), empirical results (Sharpe ratios, drawdowns), and discusses systemic implications for market stability. It is essential reading for anyone working at the intersection of AI and quantitative finance.

The paper discusses highly complex multi-agent architectures requiring sophisticated infrastructure (edge/cloud computing, cognitive architectures, multi-agent system models, LLMs, RL algorithms). Implementation involves multiple specialized agents with distinct roles (fact reasoning, sentiment analysis, trade execution, risk monitoring, reflection), layered memory systems, communication protocols, and coordination mechanisms. The three-layer technology stack (Infrastructure, Intelligence, Decision and Learning) and progressive testing pipelines (simulated lab, testing bed, red-team, partial deployment) indicate substantial engineering complexity. However, as a survey, it does not provide a single implementation blueprint.

As a comprehensive survey paper, it does not present a single reproducible experiment. It synthesizes findings from numerous cited works across Scopus-indexed papers and preprints. Some referenced frameworks (e.g., ABIDES-MARL, FINMEM, LLMFactor) may have associated code, but the survey itself does not provide a unified reproducible pipeline. The evaluation tables (e.g., MADDPG results on 100 Shenzhen Stock Exchange stocks 2018-2021) reference specific experimental setups from cited papers.