Noise-Filtered Influence Networks For Portfolio Management

This article presents a practical, model-free framework for monitoring changing connectedness in financial markets. Starting from correlation snapshots, the framework constructs directional and economically interpretable influence signals for portfolio-management purposes. A resampling procedure retains only those relationships that are sufficiently stable to support portfolio action. The method focuses on filtered directional relationships rather than unstable pairwise correlations and can therefore serve as an early-warning tool for stress scenarios. Significant negative influence signals are interpreted as signs that the transmission of systematic factors is beginning to break down. Applications to portfolio selection and risk management are discussed. In particular, the article illustrates a transparent dynamic risk-management strategy in which markets associated with significant negative influence signals are reduced or hedged in the subsequent period, with the aim of lowering drawdowns and realized volatility when the dependence structure fragments.