A1423
Title: Global high frequency price forecasting through distributed consensus using local microstructure
Authors: Juan Diego Sanchez-Torres - Western Institute of Technology and Higher Education (Mexico)
Juan Francisco Munoz-Elguezabal - Western Institue of Technology and Higher Education (ITESO) (Mexico) [presenting]
Abstract: The purpose is to introduce a novel distributed convex optimization framework for high-frequency financial forecasting that addresses the fundamental econometric challenge of achieving global consensus across heterogeneous market environments while preserving local adaptation capabilities. The methodology extends classical diffusion strategies through dynamic regularization mechanisms, implementing both combine-then-adapt (CTA) and adapt-then-combine (ATC) protocols within leader-follower network topologies. The framework provides theoretical convergence guarantees under convexity assumptions while maintaining computational tractability for millisecond-level market microstructure analysis. Empirical validation using high-frequency cryptocurrency market data across nine distributed agents processing 77.76 million observations demonstrates superior econometric performance: 21\% accuracy improvement over traditional centralized approaches with sub-100 millisecond processing latency. The distributed architecture exhibits remarkable robustness, showing less than 8\% performance degradation during regional node failures, making it particularly suitable for fragmented cryptocurrency markets where traditional econometric models fail to capture cross-exchange dependencies and regional microstructure heterogeneity effectively.
