Multi-Agent Attention for Adaptive Portfolio Allocation
Summary
The article presents MASAAT, a portfolio optimization framework that combines multiple agents, directional price filters, and attention mechanisms. Agents use filters with different thresholds to extract trends at different scales. Cross-sectional attention analyzes relationships among assets, while temporal attention analyzes relationships among observation times; a fusion stage combines those views to propose allocations. Agent proposals are integrated into an ensemble portfolio, and a policy is updated using rewards and replayed experience.
The article also begins an MQL5 implementation, describing parallel trend detection and an attention module. It explains the framework’s intended advantage: diverse agent perspectives may reduce the effect of a single agent’s mistaken view and help account for changing correlations. However, the implementation and evaluation are incomplete in this installment. The article does not provide results on historical data, so its discussion supports understanding of the design rather than conclusions about trading performance or risk reduction.
Key ideas
- Agents use directional filters with different thresholds to represent price movements at multiple scales.
- Cross-sectional attention models relationships among assets, while temporal attention models relationships among time points.
- A fusion mechanism combines both attention views to inform portfolio weights.
- The framework aggregates proposals from multiple agents and updates its policy using rewards and replayed experience.
- This installment starts an implementation but does not report performance tests on historical data.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.