Network granger causality with inherent grouping structure

摘要: The problem of estimating high-dimensional network models arises naturally in the analysis many biological and socio-economic systems. In this work, we aim to learn a structure from temporal panel data, employing framework Granger causal under assumptions sparsity its edges inherent grouping among nodes. To that end, introduce group lasso regression regularization framework, also examine thresholded variant address issue misspecification. Further, norm consistency variable selection estimates are established, latter novel concept direction consistency. performance proposed methodology is assessed through an extensive set simulation studies comparisons with existing techniques. study illustrated on two motivating examples coming functional genomics financial econometrics.