Agentic Causal Macro Intelligence: A Mixture-of-Experts Framework for Regime-Conditional Systematic Trading

This paper presents a production-grade systematic trading architecture that replaces static, correlation-dependent financial modeling with explicit causal transmission paths. Developed for deployment within the ATRX Intelligence Limited infrastructure, our framework implements Pearl’s do-calculus and Directed Acyclic Graphs (DAGs) using the DoWhy library to evaluate counterfactual macroeconomic shocks across a 32-node topological network. To bridge the latency gap between slow-moving macro data and high-frequency execution, the system employs an M1-frequency (minute-level) Hidden Markov Model (HMM) to detect market microstructure regimes in real time. Capital routing and position sizing are dynamically governed by an Agentic Mixture-of-Experts (MoE) neural network conditioned on a continuous Causal Alignment Score. In live production testing, the Causal Macro Transmission Framework achieves an independently verified Annualized Sharpe Ratio of 2.1, a net strategy return of +27.4%, and a maximum drawdown of -8.4% under sub-second execution latencies.