Federated Causal-Driven Autonomous Investment Advisory Framework with Privacy-Preserving Multi Agent Intelligence Analysis

Abstract

Financial systems are requiring an increasing amount of collaboration across entities; however, institutional, regulatory, and technical restrictions on data sharing and ownership limit the ability to create a true autonomous financial advisor. Centralized investment advisors are limited by their inability to scale with growing amounts of financial data as well as they expose this sensitive information to potential risks such as cybercrime. This paper provides a Federated Learning-Based Autonomous Investment Advisor that integrates five new and distinct modules: Differentially Private Federated Market Representation Learning (DP-FMRL), which enables safe and efficient feature extraction from multiple datasets; the Causal-Adaptive Federated Risk Propagation Network (CA-FRPN), which models interdependencies within the system; the Multi-Agent Reinforcement Federated Strategy Synthesizer (MARFSS); which synthesizes strategic policies in real-time based upon input received from all agents; the Privacy-Preserving Explainable Decision Tensorization Engine (PP-EDTE); which generates decisions in a manner that is understandable to users; and finally, the Secure Hierarchical Federated Execution and Validation Optimizer (SH-FEVO), which validates the execution of the overall federated strategy in a reliable and auditable fashion for real time scenarios. The proposed solution achieves end-to-end privacy protection as it enhances causal reasoning, flexibility and transparency during the investment process.