Change Point Detection and Evolution Learning for Dynamic Networks

Sparse inverse covariance matrix estimation is a powerful tool for uncovering dependency structures in high dimensional data. Its non-zero elements represent an undirected graphical model, enabling downstream graph analysis. This paper presents a unified framework leveraging parameter estimates and graph connectivity to detect change points and measure network similarity. Our approach supports consistent qualitative and quantitative analyses using the same sparse estimate. We demonstrate its effectiveness through simulations and apply it to a practical portfolio selection strategy, validated by extensive backtesting.