META Adds Episodic Memory to Multi-Agent Financial Decision-Making

Summary

Large language models have shown promise in financial analysis and reasoning, but earlier agent-based trading systems often either focus on long-horizon forecasting or analyze markets without persistent state. The paper introduces META, or Memory Enhanced Trading Agent, which is described as the first RAG-like episodic-memory-augmented multi-agent framework for financial decision-making. META uses specialized agents for indicators including Trend, MACD, Stochastic, RSI, SMA, AVWAP, and Heikin-Ashi, while a Decision Agent combines their reports. Its Memory module stores past trading episodes as market-state embeddings paired with outcomes and reflections, then retrieves relevant experiences under similar market regimes. The system can use those recalled episodes to adaptively reweight indicator signals. In short-horizon evaluations, the authors report improved directional accuracy and robustness. They also characterize episodic memory as supporting regime-aware, interpretable, and low-latency decisions. The project code has been released on GitHub.