A0464
Title: High-dimensional-responses-assisted heterogeneous nodal influence analysis
Authors: Dongxue Zhang - Southwestern University of Finance and Economics (China) [presenting]
Abstract: An $m\times n$ matrix network data is considered with m network nodes and n-dimensional responses for each node, where both $m$ and $n$ can diverge to infinity. The heterogeneity of network nodal influence is addressed by different influence parameters of each node, which are expressed through high-dimensional responses using a specific link function. By allowing heterogeneous error variances, a response-assisted network influence model is proposed to integrate information on the matrix response variable and network structures across both $m$ network nodes and $n$ dimensions of responses. Since the traditional maximum likelihood estimation method is invalid in this case, an optimal generalized method is built on the moment's estimation method to avoid estimating unknown error variances by restricting the diagonal of the weighting matrix in quadratic moments. The consistency and asymptotic normality of the estimator are established. In addition, a homogeneity test has also been developed to examine the influence of heterogeneity. Extensive simulation studies and an empirical study of fund and stock matrix network data are presented to demonstrate the usefulness of the proposed model.
